RTX 5090 Laptop GPU (24 GB)

24 GB·NVIDIA·graphics card·221 figures, from 3 exhibits · named in 12 exhibits

This roster is in beta and still being tested: every figure is proved against the page it came from, and the way they are read and shown is still changing.

Language 98
Pictures 16
Power, temperature, and what it costs 19
Memory 3
Everything else this shelf measured for it 85

Narrow these 221 figuresevery modelmodel: FLUX.2 [klein] 4B · 35 of 221 figuresmodel: gemma4:26b · 89 of 221 figuresmodel: HiDream-O1-Image-Dev · 3 of 221 figuresmodel: Kandinsky 5.0 Image Lite · 2 of 221 figuresmodel: Krea 2 Turbo · 2 of 221 figuresmodel: MiniCPM-V 4.5 · 10 of 221 figuresmodel: Mistral Small 3.2 · 32 of 221 figuresmodel: nomic-embed-text · 3 of 221 figuresmodel: Qwen3.8-Flash-Next · 37 of 221 figuresmodel: Z-Image base · 2 of 221 figuresmodel: Z-Image Turbo · 2 of 221 figuresmodel: not stated · 4 of 221 figuresevery power cappower cap: 95 W · 72 of 221 figurespower cap: 150 W · 130 of 221 figurespower cap: stated in words, not as a number · 2 of 221 figurespower cap: not stated · 17 of 221 figuresevery exhibitexhibit: A laptop, asked the desktop's questions · 168 of 221 figuresexhibit: A Rig Your Friend Already Owns · 16 of 221 figuresexhibit: Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench · 37 of 221 figures

Narrow by model any FLUX.2 [klein] 4B 35 gemma4:26b 89 HiDream-O1-Image-Dev 3 Kandinsky 5.0 Image Lite 2 Krea 2 Turbo 2 MiniCPM-V 4.5 10 Mistral Small 3.2 32 nomic-embed-text 3 Qwen3.8-Flash-Next 37 Z-Image base 2 Z-Image Turbo 2 not stated 4

FLUX.2 [klein] 4B, as this shelf’s pages spell itFLUX.2 klein-4B · FLUX.2 klein-4B (our sketch artist)HiDream-O1-Image-Dev, as this shelf’s pages spell ithidream-o1-devKandinsky 5.0 Image Lite, as this shelf’s pages spell itkandinsky5-liteKrea 2 Turbo, as this shelf’s pages spell itkrea2-turboMiniCPM-V 4.5, as this shelf’s pages spell itminicpm-v4.5Mistral Small 3.2, as this shelf’s pages spell itmistral-small3.2:24bZ-Image base, as this shelf’s pages spell itz-image-base (undistilled, 768 only)Z-Image Turbo, as this shelf’s pages spell itz-image-turbo

Grouped by the model tag the model roster knows, where it knows one; the page’s own words for it stay on every row.

Narrow by power cap any 95 W 72 150 W 130 stated in words, not as a number 2 not stated 17

150 W, as this shelf’s pages spell it140 to 150 W · 140 to 150 W (n=19) · 140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them) · 140–150 W (n=374; 150 W in 335) · 150 W in every one of the 24 samples of its own 2 Hz trace · 150 W over their short stages (n=11 and n=5 at five-second intervals) · 150 W, flat · 150 W, flat (n=385) · 150 W, flat — 82 samples at 2 Hz across the three recipes' timed batches · 150 W, flat — every one of 1,140 samples at 2 Hz · 150 W, flat — every one of 1,141 samples at 2 Hz · a limit of 135-150 Wstated in words, not as a number, as this shelf’s pages spell itat every posture

Grouped by the watts each page states — one control per wattage, however its page spelled it — and the page’s own words stay on every row. A string naming two wattages is a pair and not a spelling, so it keeps a control of its own.

Narrow by exhibit any A laptop, asked the desktop's questions 168 A Rig Your Friend Already Owns 16 Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench 37

The fourth facet plan §1 promised and the first edition put only on the shelf: a reader who arrived from an article wants that article’s rows, not all of this card’s. Each count under “where it appears” links the control that narrows to it.

The rail is CSS and nothing else — no script runs on this page. It narrows by model and by power cap independently, and the line above says what each one is set to; the size of the two together is the one number it cannot count. A group or a band with nothing left in it goes with its rows, so what you are left looking at is what matched. A narrowed view is not part of the address, so it cannot be linked — each card’s own page is a URL, and that is the thing to share.

Language — decode, prefill and first token

Grouped by model and by what was measured, and inside a group sorted by power cap ascending — alphabetical and numeric, never by value. The heading of a group carries everything that does not change inside it; each row carries the value and the conditions that do. A condition the page does not state reads not stated, which is a third thing and not an uncapped card. A † means the same reading is printed in a second place on its exhibit — one reading, two printings, not two figures.

value	what moves inside this group
gemma4:26b first token, msmsollama · 0.32.13 · batch or steps not stated
345.52ms	a q8_0 KV cache95 Wnot stated1 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
525.93ms	a q8_0 KV cache95 Wnot stated2 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
702.42ms	a q8_0 KV cache95 Wnot stated4 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
326.27ms	KV cache q4_095 W131,072one parallel slot18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
372.49ms†	a q8_0 KV cache95 W131,072one parallel slot18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
372.49ms†	KV cache q8_0 (the default)95 W131,072one parallel slot18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page labels this row the ladder's own rung, for reference: the comparison table at the top of this same section prints the same reading with the posture's own cap in its cell. Two printings of one measurement are two rows on this roster, and each names the other's section — here its own, since both printings sit inside this page's one cited section from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
435.96ms	KV cache f1695 W131,072one parallel slot18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
744.92ms	a q8_0 KV cache95 Wthe window filled to 131,072one parallel slot18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
336.23ms	KV cache q4_0140 to 150 W131,072one parallel slot21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
372.92ms†	KV cache q8_0 (the default)140 to 150 W131,072one parallel slot21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page labels this row the ladder's own rung, for reference: the comparison table at the top of this same section prints the same reading with the posture's own cap in its cell. Two printings of one measurement are two rows on this roster, and each names the other's section — here its own, since both printings sit inside this page's one cited section. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
721.95ms	a q8_0 KV cache140 to 150 Wthe window filled to 131,072one parallel slot21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
401.70ms	a q8_0 KV cache140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)not stated1 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
528.06ms	a q8_0 KV cache140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)not stated2 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
694.10ms	a q8_0 KV cache140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)not stated4 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
372.92ms†	a q8_0 KV cache140–150 W (n=374; 150 W in 335)131,072one parallel slot21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
439.78ms	KV cache f16150 W in every one of the 24 samples of its own 2 Hz trace131,072one parallel slot21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
371.01ms	a q8_0 KV cache150 W, flat (n=385)131,072one parallel slot00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
gemma4:26b J per 1,000 tokensJ / 1k tokollama · 0.32.13 · batch or steps not stated
857.75J / 1k tok	a q8_0 KV cache95 Wnot stated1 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
504.27J / 1k tok	a q8_0 KV cache95 Wnot stated2 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
396.86J / 1k tok	a q8_0 KV cache95 Wnot stated4 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
1,573.32J / 1k tok	a q8_0 KV cache95 Wthe window filled to 131,072one parallel slot18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
622.11J / 1k tok	KV cache f1695 W131,072one parallel slot18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
672.50J / 1k tok†	a q8_0 KV cache95 W131,072one parallel slot18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. The page prints the three runs' means beside the median so the coarseness is visible: a mixture-model run here is about two seconds long, eight to ten samples at 2 Hz, and the three means spread widely for the same speed — the joule and per-100-W cells inherit it. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
672.50J / 1k tok†	KV cache q8_0 (the default)95 W131,072one parallel slot18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page labels this row the ladder's own rung, for reference: the comparison table at the top of this same section prints the same reading with the posture's own cap in its cell. Two printings of one measurement are two rows on this roster, and each names the other's section — here its own, since both printings sit inside this page's one cited section from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
744.29J / 1k tok	KV cache q4_095 W131,072one parallel slot18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
1,718.01J / 1k tok	a q8_0 KV cache140 to 150 Wthe window filled to 131,072one parallel slot21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
787.76J / 1k tok†	KV cache q8_0 (the default)140 to 150 W131,072one parallel slot21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page labels this row the ladder's own rung, for reference: the comparison table at the top of this same section prints the same reading with the posture's own cap in its cell. Two printings of one measurement are two rows on this roster, and each names the other's section — here its own, since both printings sit inside this page's one cited section. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
955.03J / 1k tok	KV cache q4_0140 to 150 W131,072one parallel slot21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
829.49J / 1k tok	a q8_0 KV cache140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)not stated1 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
626.27J / 1k tok	a q8_0 KV cache140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)not stated2 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
405.97J / 1k tok	a q8_0 KV cache140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)not stated4 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
787.76J / 1k tok†	a q8_0 KV cache140–150 W (n=374; 150 W in 335)131,072one parallel slot21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. The page prints the three runs' means beside the median so the coarseness is visible: a mixture-model run here is about two seconds long, eight to ten samples at 2 Hz, and the three means spread widely for the same speed — the joule and per-100-W cells inherit it. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
700.31J / 1k tok	KV cache f16150 W in every one of the 24 samples of its own 2 Hz trace131,072one parallel slot21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
747.97J / 1k tok	a q8_0 KV cache150 W, flat (n=385)131,072one parallel slot00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. The page prints the three runs' means beside the median so the coarseness is visible: a mixture-model run here is about two seconds long, eight to ten samples at 2 Hz, and the three means spread widely for the same speed — the joule and per-100-W cells inherit it. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
gemma4:26b per stream, tokens/stok/sa q8_0 KV cache · ollama · 0.32.13 · context and batch or steps not stated
118.368tok/s†	95 W1 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
99.950tok/s	95 W2 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
69.255tok/s	95 W4 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
155.231tok/s†	140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)1 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
128.550tok/s	140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)2 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
87.994tok/s	140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)4 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
gemma4:26b summed tokens/stok/sa q8_0 KV cache · ollama · 0.32.13 · context and batch or steps not stated
118.368tok/s†	95 W1 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
199.413tok/s	95 W2 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
275.732tok/s	95 W4 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
155.231tok/s†	140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)1 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
256.955tok/s	140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)2 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
351.997tok/s	140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)4 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
gemma4:26b tokens/stok/sa q8_0 KV cache · ollama · 0.32.13 · 131,072 · one parallel slot · batch or steps not stated
118.424tok/s†	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
155.251tok/s†	140–150 W (n=374; 150 W in 335)21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
155.288tok/s	150 W, flat (n=385)00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
gemma4:26b tokens/s (spread of three)tok/sollama · 0.32.13 · one parallel slot · batch or steps not stated
116.125 (115.846–116.716)	KV cache q4_095 W131,07218:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
118.424 (117.802–119.239)†	KV cache q8_0 (the default)95 W131,07218:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page labels this row the ladder's own rung, for reference: the comparison table at the top of this same section prints the same reading with the posture's own cap in its cell. Two printings of one measurement are two rows on this roster, and each names the other's section — here its own, since both printings sit inside this page's one cited section from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
121.379 (120.822–123.017)	KV cache f1695 W131,07218:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
56.174 (56.039–56.374)	a q8_0 KV cache95 Wthe window filled to 131,07218:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
155.251 (155.128–155.593)†	KV cache q8_0 (the default)140 to 150 W131,07221:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page labels this row the ladder's own rung, for reference: the comparison table at the top of this same section prints the same reading with the posture's own cap in its cell. Two printings of one measurement are two rows on this roster, and each names the other's section — here its own, since both printings sit inside this page's one cited section. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
155.712 (155.641–155.908)	KV cache q4_0140 to 150 W131,07221:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
72.159 (72.024–72.171)	a q8_0 KV cache140 to 150 Wthe window filled to 131,07221:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
165.656 (165.545–166.175)	KV cache f16150 W in every one of the 24 samples of its own 2 Hz trace131,07221:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
gemma4:26b tokens/s per 100 Wtok/s per 100 Wa q8_0 KV cache · ollama · 0.32.13 · 131,072 · one parallel slot · batch or steps not stated
148.7tok/s per 100 W	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. The page prints the three runs' means beside the median so the coarseness is visible: a mixture-model run here is about two seconds long, eight to ten samples at 2 Hz, and the three means spread widely for the same speed — the joule and per-100-W cells inherit it. The exhibit prints this ratio; the roster computes none. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
126.7tok/s per 100 W	140–150 W (n=374; 150 W in 335)21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. The page prints the three runs' means beside the median so the coarseness is visible: a mixture-model run here is about two seconds long, eight to ten samples at 2 Hz, and the three means spread widely for the same speed — the joule and per-100-W cells inherit it. The exhibit prints this ratio; the roster computes none. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
133.7tok/s per 100 W	150 W, flat (n=385)00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. The page prints the three runs' means beside the median so the coarseness is visible: a mixture-model run here is about two seconds long, eight to ten samples at 2 Hz, and the three means spread widely for the same speed — the joule and per-100-W cells inherit it. The exhibit prints this ratio; the roster computes none. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
MiniCPM-V 4.5 the page says: minicpm-v4.5 first token, msmsa q8_0 KV cache · ollama · 0.32.13 · one parallel slot · context not stated
149.86ms	95 Wmeasured here on its text path only18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
132.94ms	140 to 150 Wtext path only21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
MiniCPM-V 4.5 the page says: minicpm-v4.5 J per 1,000 tokensJ / 1k toka q8_0 KV cache · ollama · 0.32.13 · one parallel slot · context not stated
913.16J / 1k tok	95 Wmeasured here on its text path only18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
1,188.72J / 1k tok	140 to 150 Wtext path only21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
MiniCPM-V 4.5 the page says: minicpm-v4.5 tokens/s (spread of three)tok/sa q8_0 KV cache · ollama · 0.32.13 · one parallel slot · context not stated
86.808 (85.795–87.364)	95 Wmeasured here on its text path only18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
121.733 (121.722–121.762)	140 to 150 Wtext path only21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
Mistral Small 3.2 the page says: mistral-small3.2:24b first token, msmsa q8_0 KV cache · ollama · 0.32.13 · one parallel slot
214.37ms	95 W98,304every layer forced onto the card18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
210.94ms	95 W65,536not stated18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
208.97ms	140 to 150 W98,304every layer forced onto the card21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
203.72ms	140–150 W (n=374; 150 W in 335)65,536not stated21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
191.65ms	150 W, flat (n=385)65,536not stated00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
Mistral Small 3.2 the page says: mistral-small3.2:24b J per 1,000 tokensJ / 1k toka q8_0 KV cache · ollama · 0.32.13 · one parallel slot
3,227.99J / 1k tok	95 W98,304every layer forced onto the card18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
3,131.56J / 1k tok	95 W65,536not stated18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. On the dense model the runs are twenty seconds long and the three means agree within a few per cent, which is the page's own reason for printing one mean here and three above. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
3,124.35J / 1k tok	140 to 150 W98,304every layer forced onto the card21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
3,019.40J / 1k tok	140–150 W (n=374; 150 W in 335)65,536not stated21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. On the dense model the runs are twenty seconds long and the three means agree within a few per cent, which is the page's own reason for printing one mean here and three above. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
2,904.35J / 1k tok	150 W, flat (n=385)65,536not stated00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. On the dense model the runs are twenty seconds long and the three means agree within a few per cent, which is the page's own reason for printing one mean here and three above. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
Mistral Small 3.2 the page says: mistral-small3.2:24b tokens/stok/sa q8_0 KV cache · ollama · 0.32.13 · 65,536 · one parallel slot · batch or steps not stated
28.839tok/s	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
44.897tok/s	140–150 W (n=374; 150 W in 335)21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
44.702tok/s	150 W, flat (n=385)00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
Mistral Small 3.2 the page says: mistral-small3.2:24b tokens/s (spread of three)tok/sa q8_0 KV cache · ollama · 0.32.13 · 98,304 · every layer forced onto the card · one parallel slot
28.851 (28.721–28.993)	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
44.946 (44.917–44.963)	140 to 150 W21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
Mistral Small 3.2 the page says: mistral-small3.2:24b tokens/s per 100 Wtok/s per 100 Wa q8_0 KV cache · ollama · 0.32.13 · 65,536 · one parallel slot · batch or steps not stated
31.9tok/s per 100 W	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. On the dense model the runs are twenty seconds long and the three means agree within a few per cent, which is the page's own reason for printing one mean here and three above. The exhibit prints this ratio; the roster computes none. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
33.1tok/s per 100 W	140–150 W (n=374; 150 W in 335)21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. On the dense model the runs are twenty seconds long and the three means agree within a few per cent, which is the page's own reason for printing one mean here and three above. The exhibit prints this ratio; the roster computes none. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
34.4tok/s per 100 W	150 W, flat (n=385)00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. On the dense model the runs are twenty seconds long and the three means agree within a few per cent, which is the page's own reason for printing one mean here and three above. The exhibit prints this ratio; the roster computes none. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
Qwen3.8-Flash-Next writing speed, tokens a secondtok/sa limit of 135-150 W · one request at a time · context not stated
17.5tok/s	Q2_0llama.cpp · 3cf03257128K (128,037 tokens) · 8 threads · 24 of 48 layers of experts in system memory2026-10-03T06:07:35Z → 06:31:09Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe clean re-read, the figure of record; llama.cpp's memory limit reached, as the page's caption says, and the two llama.cpp attempts agree within about 6 % on every figure from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The Q2_0 pair (the laptop) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-q2-0-pair-the-laptop
13.0tok/s	IQ3_Sllama.cpp · 3cf03257128K (128,037 tokens) · a 4,096-token micro-batch · 8 threads · 42 layers in system memory2026-10-05T14:43:47Z → 14:59:47Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe page's control, registered after the first readings: a control, no band; measured while two of our audit jobs ran on the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § Writing the answer (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #writing-the-answer-the-laptop-the-iq3-s-file
14.2tok/s	IQ3_Sllama.cpp · 3cf03257128K (128,037 tokens) · a 512-token micro-batch (its default) · 8 threads · 34 of 48 layers' experts in system memory2026-10-02T23:03:15Z → 23:40:55Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentences from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § Writing the answer (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #writing-the-answer-the-laptop-the-iq3-s-file
102.5tok/s	Q2_0the Strata engine · 0.1.25128K (128,037 tokens) · reading 8,192 tokens at a time2026-10-03T05:09:47Z → 05:13:25Zon
“the draft head on” — § The Q2_0 pair (the laptop) from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The Q2_0 pair (the laptop) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-q2-0-pair-the-laptop
69.8tok/s	IQ3_Sthe Strata engine · 0.1.25128K (128,037 tokens) · reading 8,192 tokens at a time · about a third of the experts on the card2026-10-03T00:46:41Z → 00:50:45Zon
“with the model's draft head” — § Writing the answer (the laptop, the IQ3_S file)measured while another job's dry run shared the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § Writing the answer (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #writing-the-answer-the-laptop-the-iq3-s-file
26.6tok/s	Q2_0llama.cpp · 3cf0325732K (32,038 tokens) · 8 threads · 24 of 48 layers of experts in system memory2026-10-03T06:07:35Z → 06:31:09Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe clean re-read, the figure of record; llama.cpp's memory limit reached, as the page's caption says, and the two llama.cpp attempts agree within about 6 % on every figure from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The Q2_0 pair (the laptop) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-q2-0-pair-the-laptop
19.4tok/s	IQ3_Sllama.cpp · 3cf0325732K (32,038 tokens) · a 4,096-token micro-batch · 8 threads · 42 layers in system memory2026-10-05T14:43:47Z → 14:59:47Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe page's control, registered after the first readings: a control, no band; measured while two of our audit jobs ran on the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § Writing the answer (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #writing-the-answer-the-laptop-the-iq3-s-file
20.7tok/s	IQ3_Sllama.cpp · 3cf0325732K (32,038 tokens) · a 512-token micro-batch (its default) · 8 threads · 34 of 48 layers' experts in system memory2026-10-02T23:03:15Z → 23:40:55Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentences from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § Writing the answer (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #writing-the-answer-the-laptop-the-iq3-s-file
115.2tok/s	Q2_0the Strata engine · 0.1.2532K (32,038 tokens) · reading 8,192 tokens at a time2026-10-03T05:09:47Z → 05:13:25Zon
“the draft head on” — § The Q2_0 pair (the laptop) from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The Q2_0 pair (the laptop) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-q2-0-pair-the-laptop
87.1tok/s	IQ3_Sthe Strata engine · 0.1.2532K (32,038 tokens) · reading 8,192 tokens at a time · about a third of the experts on the card2026-10-03T00:46:41Z → 00:50:45Zon
“with the model's draft head” — § Writing the answer (the laptop, the IQ3_S file)measured while another job's dry run shared the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § Writing the answer (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #writing-the-answer-the-laptop-the-iq3-s-file
32.5tok/s	Q2_0llama.cpp · 3cf032574K (4,038 tokens) · 8 threads · 24 of 48 layers of experts in system memory2026-10-03T06:07:35Z → 06:31:09Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe clean re-read, the figure of record; llama.cpp's memory limit reached, as the page's caption says, and the two llama.cpp attempts agree within about 6 % on every figure from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The Q2_0 pair (the laptop) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-q2-0-pair-the-laptop
22.5tok/s	IQ3_Sllama.cpp · 3cf032574K (4,038 tokens) · a 4,096-token micro-batch · 8 threads · 42 layers in system memory2026-10-05T14:43:47Z → 14:59:47Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe page's control, registered after the first readings: a control, no band; measured while two of our audit jobs ran on the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § Writing the answer (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #writing-the-answer-the-laptop-the-iq3-s-file
24.4tok/s	IQ3_Sllama.cpp · 3cf032574K (4,038 tokens) · a 512-token micro-batch (its default) · 8 threads · 34 of 48 layers' experts in system memory2026-10-02T23:03:15Z → 23:40:55Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentences from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § Writing the answer (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #writing-the-answer-the-laptop-the-iq3-s-file
85.6tok/s	IQ3_Sthe Strata engine · 0.1.254K (4,038 tokens) · reading 8,192 tokens at a time · about a third of the experts on the card2026-10-03T00:46:41Z → 00:50:45Zon
“with the model's draft head” — § Writing the answer (the laptop, the IQ3_S file)measured while another job's dry run shared the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § Writing the answer (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #writing-the-answer-the-laptop-the-iq3-s-file

Where one condition moved, and nothing else

seconds per render, median (min–max) — batch or steps apart

The same flux.2 klein-4b

2.40 s (2.35–2.98) — 1024×1024, steps held at 4, six renders per cell The sketch ladder — the number the game lives on
0.71 s (0.69–1.27) — 512×512, steps held at 4, six renders per cell The sketch ladder — the number the game lives on
1.37 s (1.36–2.02) — 768×768, steps held at 4, six renders per cell The sketch ladder — the number the game lives on

Every row above is the same card and the same FLUX.2 klein-4B. The only thing that moved is the batch or steps. That is why nothing on this page is ranked, and why a condition is part of a row rather than a heading above one.

Pictures — seconds per render

16 rows — same rule as above.

value	what moves inside this group
FLUX.2 [klein] 4B the page says: FLUX.2 klein-4B (our sketch artist) median seconds per render, mixed resolutionssidentical mixed resolution-and-step sets on every rig · quantisation, runtime, power cap, context, concurrency and measurement window not stated
1.39 s
from A Rig Your Friend Already Owns § The table that answers the question read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the table’s own heading · read from a-rig-your-friend-already-owns · section id #the-table-that-answers-the-question
FLUX.2 [klein] 4B the page says: FLUX.2 klein-4B seconds per render, median (min–max)squantisation, runtime, power cap, context, concurrency and measurement window not stated
2.40 s (2.35–2.98)	1024×1024, steps held at 4, six renders per cell
from A Rig Your Friend Already Owns § The sketch ladder — the number the game lives on read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the table’s own heading · read from a-rig-your-friend-already-owns · section id #the-sketch-ladder-the-number-the-game-lives-on
0.71 s (0.69–1.27)	512×512, steps held at 4, six renders per cell
from A Rig Your Friend Already Owns § The sketch ladder — the number the game lives on read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the table’s own heading · read from a-rig-your-friend-already-owns · section id #the-sketch-ladder-the-number-the-game-lives-on
1.37 s (1.36–2.02)	768×768, steps held at 4, six renders per cell
from A Rig Your Friend Already Owns § The sketch ladder — the number the game lives on read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the table’s own heading · read from a-rig-your-friend-already-owns · section id #the-sketch-ladder-the-number-the-game-lives-on
FLUX.2 [klein] 4B the page says: FLUX.2 klein-4B (our sketch artist) the exhibit's own multiplier against the 96 GB cardidentical mixed resolution-and-step sets on every rig · quantisation, runtime, power cap, context, concurrency and measurement window not stated
2.9×
the exhibit prints this ratio; the roster computes none from A Rig Your Friend Already Owns § The table that answers the question read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the same section · read from a-rig-your-friend-already-owns · section id #the-table-that-answers-the-question
HiDream-O1-Image-Dev the page says: hidream-o1-dev median seconds per render, mixed resolutionssidentical mixed resolution-and-step sets on every rig · quantisation, runtime, power cap, context, concurrency and measurement window not stated
3.45 s
from A Rig Your Friend Already Owns § The table that answers the question read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the table’s own heading · read from a-rig-your-friend-already-owns · section id #the-table-that-answers-the-question
HiDream-O1-Image-Dev the page says: hidream-o1-dev seconds per render against the 96 GB card, at 512 pxat 512 · quantisation, runtime, power cap, context, concurrency and measurement window not stated
the laptop beats the 96 GB card outright (1.90 s vs 2.49 s)
from A Rig Your Friend Already Owns § Three findings we didn't order typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-rig-your-friend-already-owns · section id #three-findings-we-didnt-order
HiDream-O1-Image-Dev the page says: hidream-o1-dev the exhibit's own multiplier against the 96 GB cardidentical mixed resolution-and-step sets on every rig · quantisation, runtime, power cap, context, concurrency and measurement window not stated
†
the exhibit prints this ratio; the roster computes none from A Rig Your Friend Already Owns § The table that answers the question read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the same section · read from a-rig-your-friend-already-owns · section id #the-table-that-answers-the-question
Kandinsky 5.0 Image Lite the page says: kandinsky5-lite median seconds per render, mixed resolutionssidentical mixed resolution-and-step sets on every rig · quantisation, runtime, power cap, context, concurrency and measurement window not stated
24.46 s
from A Rig Your Friend Already Owns § The table that answers the question read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the table’s own heading · read from a-rig-your-friend-already-owns · section id #the-table-that-answers-the-question
Kandinsky 5.0 Image Lite the page says: kandinsky5-lite the exhibit's own multiplier against the 96 GB cardidentical mixed resolution-and-step sets on every rig · quantisation, runtime, power cap, context, concurrency and measurement window not stated
4.0×
the exhibit prints this ratio; the roster computes none from A Rig Your Friend Already Owns § The table that answers the question read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the same section · read from a-rig-your-friend-already-owns · section id #the-table-that-answers-the-question
Krea 2 Turbo the page says: krea2-turbo median seconds per render, mixed resolutionssidentical mixed resolution-and-step sets on every rig · quantisation, runtime, power cap, context, concurrency and measurement window not stated
5.86 s
from A Rig Your Friend Already Owns § The table that answers the question read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the table’s own heading · read from a-rig-your-friend-already-owns · section id #the-table-that-answers-the-question
Krea 2 Turbo the page says: krea2-turbo the exhibit's own multiplier against the 96 GB cardidentical mixed resolution-and-step sets on every rig · quantisation, runtime, power cap, context, concurrency and measurement window not stated
3.0×
the exhibit prints this ratio; the roster computes none from A Rig Your Friend Already Owns § The table that answers the question read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the same section · read from a-rig-your-friend-already-owns · section id #the-table-that-answers-the-question
Z-Image base the page says: z-image-base (undistilled, 768 only) median seconds per render, mixed resolutionssidentical mixed resolution-and-step sets on every rig · quantisation, runtime, power cap, context, concurrency and measurement window not stated
28.88 s
from A Rig Your Friend Already Owns § The table that answers the question read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the table’s own heading · read from a-rig-your-friend-already-owns · section id #the-table-that-answers-the-question
Z-Image base the page says: z-image-base (undistilled, 768 only) the exhibit's own multiplier against the 96 GB cardidentical mixed resolution-and-step sets on every rig · quantisation, runtime, power cap, context, concurrency and measurement window not stated
3.7×
the exhibit prints this ratio; the roster computes none from A Rig Your Friend Already Owns § The table that answers the question read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the same section · read from a-rig-your-friend-already-owns · section id #the-table-that-answers-the-question
Z-Image Turbo the page says: z-image-turbo median seconds per render, mixed resolutionssidentical mixed resolution-and-step sets on every rig · quantisation, runtime, power cap, context, concurrency and measurement window not stated
3.80 s
from A Rig Your Friend Already Owns § The table that answers the question read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the table’s own heading · read from a-rig-your-friend-already-owns · section id #the-table-that-answers-the-question
Z-Image Turbo the page says: z-image-turbo the exhibit's own multiplier against the 96 GB cardidentical mixed resolution-and-step sets on every rig · quantisation, runtime, power cap, context, concurrency and measurement window not stated
3.4×
the exhibit prints this ratio; the roster computes none from A Rig Your Friend Already Owns § The table that answers the question read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in the same section · read from a-rig-your-friend-already-owns · section id #the-table-that-answers-the-question

Power, temperature, and what it costs

19 rows — same rule as above.

value	what moves inside this group
FLUX.2 [klein] 4B the page says: FLUX.2 klein-4B core temperature plateauComfyUI · 0.21.1 · Ten minutes of continuous drawing — the 8-step recipe, one image at a time · quantisation, context and concurrency not stated
54 °C at 3.3 min	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
68 °C at 4.6 min	150 W, flat — every one of 1,140 samples at 2 Hz00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
68 °C at 5.5 min	150 W, flat — every one of 1,141 samples at 2 Hz21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
gemma4:26b over the wall clock, tokens/stok/sa q8_0 KV cache · ollama · 0.32.13 · context and batch or steps not stated
102.226tok/s	95 W1 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
165.321tok/s	95 W2 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
231.239tok/s	95 W4 streams18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
124.544tok/s	140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)1 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
202.821tok/s	140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)2 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
283.934tok/s	140 to 150 W over the arm (n=33 at five-second intervals; 150 W in 30 of them)4 streams21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
Qwen3.8-Flash-Next prompt rate, tokens a second, the engine's clocktok/sIQ3_S · a limit of 135-150 W · one request at a time · context not stated
626.7tok/s	llama.cpp · 3cf03257128K (128,037 tokens) · a 4,096-token micro-batch · 8 threads · 42 layers in system memory2026-10-05T14:43:47Z → 14:59:47Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe page's control, registered after the first readings: a control, no band; measured while two of our audit jobs ran on the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
234.9tok/s	llama.cpp · 3cf03257128K (128,037 tokens) · a 512-token micro-batch (its default) · 8 threads · 34 of 48 layers' experts in system memory2026-10-02T23:03:15Z → 23:40:55Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentences from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
2,705.0tok/s	the Strata engine · 0.1.25128K (128,037 tokens) · reading 8,192 tokens at a time · about a third of the experts on the card2026-10-03T00:46:41Z → 00:50:45Zon
“drafts several tokens a step from the model's own draft head” — § One model, two enginesmeasured while another job's dry run shared the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
640.9tok/s	llama.cpp · 3cf0325732K (32,038 tokens) · a 4,096-token micro-batch · 8 threads · 42 layers in system memory2026-10-05T14:43:47Z → 14:59:47Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe page's control, registered after the first readings: a control, no band; measured while two of our audit jobs ran on the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
255.1tok/s	llama.cpp · 3cf0325732K (32,038 tokens) · a 512-token micro-batch (its default) · 8 threads · 34 of 48 layers' experts in system memory2026-10-02T23:03:15Z → 23:40:55Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentences from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
2,655.1tok/s	the Strata engine · 0.1.2532K (32,038 tokens) · reading 8,192 tokens at a time · about a third of the experts on the card2026-10-03T00:46:41Z → 00:50:45Zon
“drafts several tokens a step from the model's own draft head” — § One model, two enginesmeasured while another job's dry run shared the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
533.7tok/s	llama.cpp · 3cf032574K (4,038 tokens) · a 4,096-token micro-batch · 8 threads · 42 layers in system memory2026-10-05T14:43:47Z → 14:59:47Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe page's control, registered after the first readings: a control, no band; measured while two of our audit jobs ran on the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
211.7tok/s	llama.cpp · 3cf032574K (4,038 tokens) · a 512-token micro-batch (its default) · 8 threads · 34 of 48 layers' experts in system memory2026-10-02T23:03:15Z → 23:40:55Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentences from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
2,147.5tok/s	the Strata engine · 0.1.254K (4,038 tokens) · reading 8,192 tokens at a time · about a third of the experts on the card2026-10-03T00:46:41Z → 00:50:45Zon
“drafts several tokens a step from the model's own draft head” — § One model, two enginesmeasured while another job's dry run shared the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
not stated price, the whole machine this part is soldered into, newordered 2026-07-03 · quantisation, runtime, power cap, context, batch or steps and concurrency not stated
$3,099.99
before tax, from Micro Center, and it is the price of the whole laptop rather than of this board: the page's own sentence is that this part is soldered into a laptop and is not sold as a card, and that a buyer is weighing that whole machine against a card plus the rest of a computer. The same section is where the page rules that no reading from this bench may fill a cell on this shelf's comparison page, its price tables included from A laptop, asked the desktop's questions § What this page cannot say typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #what-this-page-cannot-say

Memory — what fits, and at which context

3 rows — same rule as above.

value	what moves inside this group
gemma4:26b held resident, idle W, and what residency costsollama · 0.32.13 · 131,072 · mean over 30 s · quantisation and concurrency not stated
9.03 W, so residency costs 1.85 W	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
11.34 W, so residency costs 4.20 W	140 to 150 W (n=19)21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
14.42 W, so residency costs 2.85 W	150 W, flat00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables

Everything else this shelf measured for it

85 rows — same rule as above.

value	what moves inside this group
FLUX.2 [klein] 4B the page says: FLUX.2 klein-4B board W median/maxWComfyUI · 0.21.1 · Ten minutes of continuous drawing — the 8-step recipe, one image at a time · quantisation, context and concurrency not stated
93.8 / 100.6W	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
146.4 / 155.2W	150 W, flat — every one of 1,140 samples at 2 Hz00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
145.9 / 156.9W	150 W, flat — every one of 1,141 samples at 2 Hz21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
FLUX.2 [klein] 4B the page says: FLUX.2 klein-4B core max °C°CComfyUI · 0.21.1 · Ten minutes of continuous drawing — the 8-step recipe, one image at a time · quantisation, context and concurrency not stated
58°C	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
70°C	150 W, flat — every one of 1,140 samples at 2 Hz00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
71°C	150 W, flat — every one of 1,141 samples at 2 Hz21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
FLUX.2 [klein] 4B the page says: FLUX.2 klein-4B images in ten minutesComfyUI · 0.21.1 · Ten minutes of continuous drawing — the 8-step recipe, one image at a time · quantisation, context and concurrency not stated
294 (600.9 s)	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
382 (600.5 s)	150 W, flat — every one of 1,140 samples at 2 Hz00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
378 (600.6 s)	150 W, flat — every one of 1,141 samples at 2 Hz21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
FLUX.2 [klein] 4B the page says: FLUX.2 klein-4B s per image (spread, n=2)sComfyUI · 0.21.1 · quantisation, context and concurrency not stated
6.086 (6.028–6.145)	95 W32 steps · two timed batches of three images each after a warm-up submit that is excluded18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. This board's render rows ran the same ComfyUI release as the desktop boards on a newer torch and driver, which a Blackwell board needs, so the page reads its render ratios as box against board rather than card against card. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
1.627 (1.579–1.675)	95 W8 steps + pixel pass · two timed batches of three images each after a warm-up submit that is excluded18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. This board's render rows ran the same ComfyUI release as the desktop boards on a newer torch and driver, which a Blackwell board needs, so the page reads its render ratios as box against board rather than card against card. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
1.640 (1.583–1.698)	95 W8 steps · two timed batches of three images each after a warm-up submit that is excluded18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. This board's render rows ran the same ComfyUI release as the desktop boards on a newer torch and driver, which a Blackwell board needs, so the page reads its render ratios as box against board rather than card against card. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
4.649 (4.584–4.715)	150 W, flat — 82 samples at 2 Hz across the three recipes' timed batches32 steps · two timed batches of three images each after a warm-up submit that is excluded21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. This board's render rows ran the same ComfyUI release as the desktop boards on a newer torch and driver, which a Blackwell board needs, so the page reads its render ratios as box against board rather than card against card. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
1.245 (1.194–1.296)	150 W, flat — 82 samples at 2 Hz across the three recipes' timed batches8 steps + pixel pass · two timed batches of three images each after a warm-up submit that is excluded21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. This board's render rows ran the same ComfyUI release as the desktop boards on a newer torch and driver, which a Blackwell board needs, so the page reads its render ratios as box against board rather than card against card. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
1.246 (1.202–1.290)	150 W, flat — 82 samples at 2 Hz across the three recipes' timed batches8 steps · two timed batches of three images each after a warm-up submit that is excluded21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. This board's render rows ran the same ComfyUI release as the desktop boards on a newer torch and driver, which a Blackwell board needs, so the page reads its render ratios as box against board rather than card against card. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
FLUX.2 [klein] 4B the page says: FLUX.2 klein-4B seconds per image (spread)sComfyUI · 0.21.1 · Ten minutes of continuous drawing — the 8-step recipe, one image at a time · quantisation, context and concurrency not stated
1.958 (1.918–2.019, n=294)	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
1.507 (1.475–1.590, n=382)	150 W, flat — every one of 1,140 samples at 2 Hz00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
1.521 (1.476–1.574, n=378)	150 W, flat — every one of 1,141 samples at 2 Hz21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
FLUX.2 [klein] 4B the page says: FLUX.2 klein-4B Wh per image, over the ten-minute windowWhComfyUI · 0.21.1 · Ten minutes of continuous drawing — the 8-step recipe, one image at a time · quantisation, context and concurrency not stated
0.0529Wh	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. Watt-hours an image is energy over the timed window divided by the images finished in it, from each run's own power trace — the page's own definition, and it warns that the ten-minute and the batch tables are therefore not the same quantity. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
0.0629Wh	150 W, flat — every one of 1,140 samples at 2 Hz00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. Watt-hours an image is energy over the timed window divided by the images finished in it, from each run's own power trace — the page's own definition, and it warns that the ten-minute and the batch tables are therefore not the same quantity. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
0.0633Wh	150 W, flat — every one of 1,141 samples at 2 Hz21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. Watt-hours an image is energy over the timed window divided by the images finished in it, from each run's own power trace — the page's own definition, and it warns that the ten-minute and the batch tables are therefore not the same quantity. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
FLUX.2 [klein] 4B the page says: FLUX.2 klein-4B Wh per image, over the timed batchesWhComfyUI · 0.21.1 · quantisation, context and concurrency not stated
0.1607Wh	95 W32 steps · two timed batches of three images each after a warm-up submit that is excluded18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. Watt-hours an image is energy over the timed window divided by the images finished in it, from each run's own power trace — the page's own definition, and it warns that the ten-minute and the batch tables are therefore not the same quantity. This board's render rows ran the same ComfyUI release as the desktop boards on a newer torch and driver, which a Blackwell board needs, so the page reads its render ratios as box against board rather than card against card. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
0.0429Wh	95 W8 steps + pixel pass · two timed batches of three images each after a warm-up submit that is excluded18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. Watt-hours an image is energy over the timed window divided by the images finished in it, from each run's own power trace — the page's own definition, and it warns that the ten-minute and the batch tables are therefore not the same quantity. This board's render rows ran the same ComfyUI release as the desktop boards on a newer torch and driver, which a Blackwell board needs, so the page reads its render ratios as box against board rather than card against card. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
0.0431Wh	95 W8 steps · two timed batches of three images each after a warm-up submit that is excluded18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. Watt-hours an image is energy over the timed window divided by the images finished in it, from each run's own power trace — the page's own definition, and it warns that the ten-minute and the batch tables are therefore not the same quantity. This board's render rows ran the same ComfyUI release as the desktop boards on a newer torch and driver, which a Blackwell board needs, so the page reads its render ratios as box against board rather than card against card. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
0.1934Wh	150 W, flat — 82 samples at 2 Hz across the three recipes' timed batches32 steps · two timed batches of three images each after a warm-up submit that is excluded21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. Watt-hours an image is energy over the timed window divided by the images finished in it, from each run's own power trace — the page's own definition, and it warns that the ten-minute and the batch tables are therefore not the same quantity. This board's render rows ran the same ComfyUI release as the desktop boards on a newer torch and driver, which a Blackwell board needs, so the page reads its render ratios as box against board rather than card against card. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
0.0513Wh	150 W, flat — 82 samples at 2 Hz across the three recipes' timed batches8 steps + pixel pass · two timed batches of three images each after a warm-up submit that is excluded21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. Watt-hours an image is energy over the timed window divided by the images finished in it, from each run's own power trace — the page's own definition, and it warns that the ten-minute and the batch tables are therefore not the same quantity. This board's render rows ran the same ComfyUI release as the desktop boards on a newer torch and driver, which a Blackwell board needs, so the page reads its render ratios as box against board rather than card against card. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
0.0500Wh	150 W, flat — 82 samples at 2 Hz across the three recipes' timed batches8 steps · two timed batches of three images each after a warm-up submit that is excluded21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. Watt-hours an image is energy over the timed window divided by the images finished in it, from each run's own power trace — the page's own definition, and it warns that the ten-minute and the batch tables are therefore not the same quantity. This board's render rows ran the same ComfyUI release as the desktop boards on a newer torch and driver, which a Blackwell board needs, so the page reads its render ratios as box against board rather than card against card. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
gemma4:26b board W (mean)Wollama · 0.32.13 · one parallel slot · batch or steps not stated
76.53W	KV cache f1695 W131,07218:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
79.64W†	KV cache q8_0 (the default)95 W131,07218:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page labels this row the ladder's own rung, for reference: the comparison table at the top of this same section prints the same reading with the posture's own cap in its cell. Two printings of one measurement are two rows on this roster, and each names the other's section — here its own, since both printings sit inside this page's one cited section from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
86.87W	KV cache q4_095 W131,07218:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
88.38W	a q8_0 KV cache95 Wthe window filled to 131,07218:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
122.57W†	KV cache q8_0 (the default)140 to 150 W131,07221:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page labels this row the ladder's own rung, for reference: the comparison table at the top of this same section prints the same reading with the posture's own cap in its cell. Two printings of one measurement are two rows on this roster, and each names the other's section — here its own, since both printings sit inside this page's one cited section. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
123.99W	a q8_0 KV cache140 to 150 Wthe window filled to 131,07221:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
148.71W	KV cache q4_0140 to 150 W131,07221:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
116.01W	KV cache f16150 W in every one of the 24 samples of its own 2 Hz trace131,07221:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
gemma4:26b board W (the three runs' means)Wa q8_0 KV cache · ollama · 0.32.13 · 131,072 · one parallel slot · batch or steps not stated
79.64 (76.95 · 79.64 · 94.22)†	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. The page prints the three runs' means beside the median so the coarseness is visible: a mixture-model run here is about two seconds long, eight to ten samples at 2 Hz, and the three means spread widely for the same speed — the joule and per-100-W cells inherit it. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
122.57 (98.58 · 146.10 · 122.57)†	140–150 W (n=374; 150 W in 335)21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. The page prints the three runs' means beside the median so the coarseness is visible: a mixture-model run here is about two seconds long, eight to ten samples at 2 Hz, and the three means spread widely for the same speed — the joule and per-100-W cells inherit it. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
116.12 (116.12 · 115.76 · 146.42)	150 W, flat (n=385)00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. The page prints the three runs' means beside the median so the coarseness is visible: a mixture-model run here is about two seconds long, eight to ten samples at 2 Hz, and the three means spread widely for the same speed — the joule and per-100-W cells inherit it. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
gemma4:26b core max °C°Collama · 0.32.13 · one parallel slot · batch or steps not stated
42°C	KV cache q8_0 (the default)95 W131,07218:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page labels this row the ladder's own rung, for reference: the comparison table at the top of this same section prints the same reading with the posture's own cap in its cell. Two printings of one measurement are two rows on this roster, and each names the other's section — here its own, since both printings sit inside this page's one cited section from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
48°C	KV cache f1695 W131,07218:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
56°C	KV cache q4_095 W131,07218:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
64°C	a q8_0 KV cache95 Wthe window filled to 131,07218:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
44°C	KV cache q8_0 (the default)140 to 150 W131,07221:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page labels this row the ladder's own rung, for reference: the comparison table at the top of this same section prints the same reading with the posture's own cap in its cell. Two printings of one measurement are two rows on this roster, and each names the other's section — here its own, since both printings sit inside this page's one cited section. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
44°C	KV cache q4_0140 to 150 W131,07221:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
48°C	a q8_0 KV cache140 to 150 Wthe window filled to 131,07221:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
44°C	KV cache f16150 W in every one of the 24 samples of its own 2 Hz trace131,07221:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
gemma4:26b largest window held wholea q8_0 KV cache · ollama · 0.32.13 · at every posture · one parallel slot · context, batch or steps and measurement window not stated
131,072 — the ladder ran out of rungs first
from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
MiniCPM-V 4.5 the page says: minicpm-v4.5 board W (mean)Wa q8_0 KV cache · ollama · 0.32.13 · one parallel slot · context not stated
79.27W	95 Wmeasured here on its text path only18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
144.74W	140 to 150 Wtext path only21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
MiniCPM-V 4.5 the page says: minicpm-v4.5 core max °C°Ca q8_0 KV cache · ollama · 0.32.13 · one parallel slot · context not stated
41°C	95 Wmeasured here on its text path only18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
43°C	140 to 150 Wtext path only21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
Mistral Small 3.2 the page says: mistral-small3.2:24b a gate-shaped call, wall latency, median of eighta q8_0 KV cache · ollama · 0.32.13 · four callers at once · context and batch or steps not stated
928.54 ms (p95 4,671.24; first token 465.75 ms; 66.59 W mean)	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
948.56 ms (p95 4,567.46; first token 569.39 ms; 73.56 W mean)	150 W over their short stages (n=11 and n=5 at five-second intervals)21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
885.73 ms (p95 4,452.80; first token 470.94 ms; 70.93 W mean)	150 W, flat00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
Mistral Small 3.2 the page says: mistral-small3.2:24b a gate-shaped call, wall latency, median of tena q8_0 KV cache · ollama · 0.32.13 · one caller · context and batch or steps not stated
461.64 ms (p95 493.52; first token 314.52 ms; 27.81 W mean)	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
438.98 ms (p95 468.26; first token 298.34 ms; 32.11 W mean)	150 W over their short stages (n=11 and n=5 at five-second intervals)21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
430.85 ms (p95 448.20; first token 293.91 ms; 32.37 W mean)	150 W, flat00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
Mistral Small 3.2 the page says: mistral-small3.2:24b board W (mean)Wa q8_0 KV cache · ollama · 0.32.13 · one parallel slot
93.13W	95 W98,304every layer forced onto the card18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
90.31W	95 W65,536not stated18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. On the dense model the runs are twenty seconds long and the three means agree within a few per cent, which is the page's own reason for printing one mean here and three above. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
140.48W	140 to 150 W98,304every layer forced onto the card21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
135.76W	140–150 W (n=374; 150 W in 335)65,536not stated21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. On the dense model the runs are twenty seconds long and the three means agree within a few per cent, which is the page's own reason for printing one mean here and three above. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
129.90W	150 W, flat (n=385)65,536not stated00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. On the dense model the runs are twenty seconds long and the three means agree within a few per cent, which is the page's own reason for printing one mean here and three above. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
Mistral Small 3.2 the page says: mistral-small3.2:24b core max °C°Ca q8_0 KV cache · ollama · 0.32.13 · 98,304 · every layer forced onto the card · one parallel slot
45°C	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
48°C	140 to 150 W21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. the page's own witness for this arm: the other arms touched 140 W, against the f16 arm's flat 150 W from A laptop, asked the desktop's questions § The tables typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
Mistral Small 3.2 the page says: mistral-small3.2:24b largest window held wholea q8_0 KV cache · ollama · 0.32.13 · at every posture · one parallel slot · context, batch or steps and measurement window not stated
65,536 — spilled at 98,304
from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
nomic-embed-text the embedding arm — texts a second, milliseconds a batch, and J per 1,000 textsollama · 0.32.13 · batches of 64 texts, 768 dimensions · quantisation, context and concurrency not stated
252.570 texts a second · 253.40 ms a batch · 65.67 J per 1,000 texts	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
268.988 texts a second · 237.93 ms a batch · 83.79 J per 1,000 texts	150 W over their short stages (n=11 and n=5 at five-second intervals)21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
250.655 texts a second · 255.33 ms a batch · 90.19 J per 1,000 texts	150 W, flat00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
Qwen3.8-Flash-Next first word, secondssa limit of 135-150 W · one request at a time · context not stated
321.8s	Q2_0llama.cpp · 3cf03257128K (128,037 tokens) · 8 threads · 24 of 48 layers of experts in system memory2026-10-03T06:07:35Z → 06:31:09Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe clean re-read, the figure of record; llama.cpp's memory limit reached, as the page's caption says, and the two llama.cpp attempts agree within about 6 % on every figure from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The Q2_0 pair (the laptop) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-q2-0-pair-the-laptop
204.4s	IQ3_Sllama.cpp · 3cf03257128K (128,037 tokens) · a 4,096-token micro-batch · 8 threads · 42 layers in system memory2026-10-05T14:43:47Z → 14:59:47Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe page's control, registered after the first readings: a control, no band; measured while two of our audit jobs ran on the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
545.1s	IQ3_Sllama.cpp · 3cf03257128K (128,037 tokens) · a 512-token micro-batch (its default) · 8 threads · 34 of 48 layers' experts in system memory2026-10-02T23:03:15Z → 23:40:55Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentences from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
45.2s	Q2_0the Strata engine · 0.1.25128K (128,037 tokens) · reading 8,192 tokens at a time2026-10-03T05:09:47Z → 05:13:25Zon
“the draft head on” — § The Q2_0 pair (the laptop) from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The Q2_0 pair (the laptop) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-q2-0-pair-the-laptop
47.6s	IQ3_Sthe Strata engine · 0.1.25128K (128,037 tokens) · reading 8,192 tokens at a time · about a third of the experts on the card2026-10-03T00:46:41Z → 00:50:45Zon
“drafts several tokens a step from the model's own draft head” — § One model, two enginesmeasured while another job's dry run shared the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
73.3s	Q2_0llama.cpp · 3cf0325732K (32,038 tokens) · 8 threads · 24 of 48 layers of experts in system memory2026-10-03T06:07:35Z → 06:31:09Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe clean re-read, the figure of record; llama.cpp's memory limit reached, as the page's caption says, and the two llama.cpp attempts agree within about 6 % on every figure from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The Q2_0 pair (the laptop) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-q2-0-pair-the-laptop
50.0s	IQ3_Sllama.cpp · 3cf0325732K (32,038 tokens) · a 4,096-token micro-batch · 8 threads · 42 layers in system memory2026-10-05T14:43:47Z → 14:59:47Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe page's control, registered after the first readings: a control, no band; measured while two of our audit jobs ran on the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
125.6s	IQ3_Sllama.cpp · 3cf0325732K (32,038 tokens) · a 512-token micro-batch (its default) · 8 threads · 34 of 48 layers' experts in system memory2026-10-02T23:03:15Z → 23:40:55Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentences from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
11.5s	Q2_0the Strata engine · 0.1.2532K (32,038 tokens) · reading 8,192 tokens at a time2026-10-03T05:09:47Z → 05:13:25Zon
“the draft head on” — § The Q2_0 pair (the laptop) from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The Q2_0 pair (the laptop) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-q2-0-pair-the-laptop
12.1s	IQ3_Sthe Strata engine · 0.1.2532K (32,038 tokens) · reading 8,192 tokens at a time · about a third of the experts on the card2026-10-03T00:46:41Z → 00:50:45Zon
“drafts several tokens a step from the model's own draft head” — § One model, two enginesmeasured while another job's dry run shared the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
9.5s	Q2_0llama.cpp · 3cf032574K (4,038 tokens) · 8 threads · 24 of 48 layers of experts in system memory2026-10-03T06:07:35Z → 06:31:09Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe clean re-read, the figure of record; llama.cpp's memory limit reached, as the page's caption says, and the two llama.cpp attempts agree within about 6 % on every figure from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The Q2_0 pair (the laptop) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-q2-0-pair-the-laptop
7.6s	IQ3_Sllama.cpp · 3cf032574K (4,038 tokens) · a 4,096-token micro-batch · 8 threads · 42 layers in system memory2026-10-05T14:43:47Z → 14:59:47Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentencesthe page's control, registered after the first readings: a control, no band; measured while two of our audit jobs ran on the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
19.1s	IQ3_Sllama.cpp · 3cf032574K (4,038 tokens) · a 512-token micro-batch (its default) · 8 threads · 34 of 48 layers' experts in system memory2026-10-02T23:03:15Z → 23:40:55Zoff
“llama.cpp at commit 3cf03257 writes one token a step” — § Why the first word comes sooner, in three sentences from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
1.9s	IQ3_Sthe Strata engine · 0.1.254K (4,038 tokens) · reading 8,192 tokens at a time · about a third of the experts on the card2026-10-03T00:46:41Z → 00:50:45Zon
“drafts several tokens a step from the model's own draft head” — § One model, two enginesmeasured while another job's dry run shared the laptop from Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench § The first word (the laptop, the IQ3_S file) read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from flash-next-why-the-first-word-comes-sooner · section id #the-first-word-the-laptop-the-iq3-s-file
not stated the card empty, idle WWmean over 30 s · quantisation, runtime, context and concurrency not stated
7.18 W (57 samples)	95 W18:43 to 19:34 UTC on 2026-09-21
the fixed posture: Dynamic Boost stopped and the board enforcing its own default, 95 W of a possible 175, with nothing on this board able to set a limit at all — the limit was read back from the card at the open and the close of every stage and at 2 Hz inside every scored run, and a change would have voided the stage. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
7.14 W (57 samples)	140 to 150 W (n=19)21:01 to 21:50 UTC the same day
the dynamic-boost posture, the state a laptop like this arrives in: Dynamic Boost running, no limit set because none can be, and the board floating its budget between its 95 W default and its 175 W maximum against the processor's draw — so the cap cell is the budget this pass sampled and not a setting anybody made. The driver answers [N/A] for fan speed on this board, so no result file carries a fan column; an operator set the laptop's fan to maximum by hand before the first pass, and the page asks that every temperature be read as a temperature under that condition and every figure that might have throttled on an ordinary fan curve as one that did not get the chance. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables
11.57 W (58 samples)	150 W, flat00:10 to 00:56 UTC on 2026-09-22
this machine's boot state: Dynamic Boost running and the boot-time clock lock an operator added to this laptop, 1,200 to 2,550 MHz, read back from the card as in force at every gate — not a state a buyer's machine is in unless they set it. The driver answers [N/A] for fan speed on this board, and the page records that an operator, asked afterwards whether the fan was still at maximum for this third pass, was not sure. from A laptop, asked the desktop's questions § The tables read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named elsewhere on that page · read from a-laptop-asked-the-desktops-questions · section id #the-tables

Where it appears

Every released exhibit that names this identity, in the shelf’s own order. A count beside an exhibit is how many of its figures are carried above, and it links the control that narrows the table to them — a number with no way back to what it counts is the thing this shelf exists against.

A laptop, asked the desktop's questions — 168 figures on this page
A Rig Your Friend Already Owns — 16 figures on this page
A short history of Qwen — named, and no figure carried from it — yet
Electricity per million tokens, card by card — named, and no figure carried from it — yet
Qwen3.8-Flash-Next on an RTX 5090 laptop, an RTX PRO 6000 Blackwell and an RTX 3090 Ti: the Strata engine and llama.cpp, the numbers — named, and no figure carried from it — yet
Why the first word comes sooner: the Strata engine and llama.cpp on Qwen3.8-Flash-Next, the full bench — 37 figures on this page
How the speed tool works — named, and no figure carried from it — yet
RTX 5080 vs RTX 3080 Ti — named, and no figure carried from it — yet
Six graphics cards, the same questions — named, and no figure carried from it — yet
The cost of intelligence, as a utility — named, and no figure carried from it — yet
The same eighteen pictures, card by card — named, and no figure carried from it — yet
What it costs to run a high-end AI rig at home — named, and no figure carried from it — yet

How this one is named here

The canonical name above is the full product name. Everything below is a spelling a published page or kit has actually used for it, kept so a reader who searches an older page’s words lands on this row. The set is closed and historical: no page cut after 2026-09-17 writes a memory class where it can write a card.

RTX 5090 Laptop
RTX 5090 Laptop 24G
RTX 5090 Laptop GPU 24G (gaming laptop)
an RTX 5090 Laptop GPU (24 GB — the laptop part, not the desktop card) in a gaming laptop