the pair — two RTX 3080s in one 2021 desktop

whole machine·81 figures, from 1 exhibit · named in 3 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.

two GeForce RTX 3080 10 GB cards of 2021 in a desktop of the same vintage, one at PCIe x16 and one at x4 of an x16-capable slot, on one supply — 20 GB of graphics memory between them, and the rig exhibit fifty measures against a single card in the same box

It holds — RTX 3080 (10 GB), Intel Core i9-10850K.

Language 30
Pictures 2
Power, temperature, and what it costs 12
Memory 1
Everything else this shelf measured for it 36

Narrow these 81 figuresevery modelmodel: gemma4:26b · 29 of 81 figuresmodel: Mistral Small 3.2 · 24 of 81 figuresmodel: one 26-billion model split across both cards · 3 of 81 figuresmodel: two copies of a 12-billion model, one per card · 3 of 81 figuresmodel: not stated · 22 of 81 figuresevery power cappower cap: 250 W each — 500 W between them · 28 of 81 figurespower cap: 300 W each — 600 W between them · 30 of 81 figurespower cap: 250 W each — 500 W between them · 300 W each — 600 W between them · 2 of 81 figurespower cap: not stated · 21 of 81 figures

Narrow by model any gemma4:26b 29 Mistral Small 3.2 24 one 26-billion model split across both cards 3 two copies of a 12-billion model, one per card 3 not stated 22

Mistral Small 3.2, as this shelf’s pages spell itmistral-small3.2:24b

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 250 W each — 500 W between them 28 300 W each — 600 W between them 30 250 W each — 500 W between them · 300 W each — 600 W between them 2 not stated 21

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.

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 decode and residency out of the box, the server's own placementQ4_K_M · ollama · 0.32.13 · a 4,096 window · 256 tokens out · 2026-09-17 · power cap and concurrency not stated
59.5 tok/s, 75.6 % on the cards
from Two 3080s against one 3090, and the cap decides § How much can each one hold? 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 two-used-3080s-priced · section id #how-much-can-each-one-hold
gemma4:26b decode at the full window, and the memory it usedollama · 0.32.13 · its full 131,072-token window · 256 tokens out · 2026-09-17 · power cap and concurrency not stated
out of memory — holds 65,536 instead, at 132.6 tok/s†	Q4_K_M · f16, unquantised
from Two 3080s against one 3090, and the cap decides § What the cache type costs 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 two-used-3080s-priced · section id #what-the-cache-type-costs
113.8 tok/s, 18,815 MiB	Q4_K_M · q4_0
from Two 3080s against one 3090, and the cap decides § What the cache type costs 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 two-used-3080s-priced · section id #what-the-cache-type-costs
115.4 tok/s, 19,531 MiB†	Q4_K_M · q8_0, the setting both rigs are measured at
from Two 3080s against one 3090, and the cap decides § What the cache type costs 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 two-used-3080s-priced · section id #what-the-cache-type-costs
gemma4:26b decode out of the box once the layer count is namedtok/sQ4_K_M · ollama · 0.32.13 · a 4,096 window · 256 tokens out · power cap, concurrency and measurement window not stated
124.7tok/s
the pair's own doubling, printed in the winner cell and nowhere else on the page; the same cell says the two rigs' out-of-the-box arms did not run at the same cap from Two 3080s against one 3090, and the cap decides § At a glance typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #at-a-glance
gemma4:26b decode per conversation, tokens a secondtok/sQ4_K_M · ollama · 0.32.13 · a 4,096 window · 256 tokens out · 4 · 2026-09-17
67.0tok/s	250 W each — 500 W between them
from Two 3080s against one 3090, and the cap decides § The full concurrency ladder 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 two-used-3080s-priced · section id #the-full-concurrency-ladder
66.9tok/s	300 W each — 600 W between them
from Two 3080s against one 3090, and the cap decides § The full concurrency ladder 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 two-used-3080s-priced · section id #the-full-concurrency-ladder
gemma4:26b decode summed across conversations, tokens a secondtok/sQ4_K_M · ollama · 0.32.13 · a 4,096 window · 256 tokens out · 2026-09-17
124.4tok/s	250 W each — 500 W between them1
from Two 3080s against one 3090, and the cap decides § The full concurrency ladder 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 two-used-3080s-priced · section id #the-full-concurrency-ladder
188.9tok/s	250 W each — 500 W between them2
from Two 3080s against one 3090, and the cap decides § The full concurrency ladder 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 two-used-3080s-priced · section id #the-full-concurrency-ladder
268.0tok/s	250 W each — 500 W between them4
from Two 3080s against one 3090, and the cap decides § The full concurrency ladder 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 two-used-3080s-priced · section id #the-full-concurrency-ladder
124.0tok/s	300 W each — 600 W between them1
from Two 3080s against one 3090, and the cap decides § The full concurrency ladder 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 two-used-3080s-priced · section id #the-full-concurrency-ladder
189.5tok/s	300 W each — 600 W between them2
from Two 3080s against one 3090, and the cap decides § The full concurrency ladder 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 two-used-3080s-priced · section id #the-full-concurrency-ladder
267.6tok/s	300 W each — 600 W between them4
from Two 3080s against one 3090, and the cap decides § The full concurrency ladder 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 two-used-3080s-priced · section id #the-full-concurrency-ladder
gemma4:26b decode, tokens a secondtok/sQ4_K_M · ollama · 0.32.13 · 131,072 · 256 tokens out · 2026-09-17 · concurrency not stated
114.8tok/s†	250 W each — 500 W between them
from Two 3080s against one 3090, and the cap decides § How fast does each one write? 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 two-used-3080s-priced · section id #how-fast-does-each-one-write
115.4tok/s†	300 W each — 600 W between them
from Two 3080s against one 3090, and the cap decides § How fast does each one write? 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 two-used-3080s-priced · section id #how-fast-does-each-one-write
gemma4:26b joules per 1,000 tokensJ / 1k tokQ4_K_M · ollama · 0.32.13 · each rig's headline window · 256 tokens out · 2026-09-17 · concurrency not stated
2,531J / 1k tok	250 W each — 500 W between them
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
2,994J / 1k tok	300 W each — 600 W between them
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
Mistral Small 3.2 the page says: mistral-small3.2:24b decode, tokens a secondtok/sQ4_K_M · ollama · 0.32.13 · 256 tokens out · 2026-09-17 · concurrency not stated
44.6tok/s†	250 W each — 500 W between them49,152
from Two 3080s against one 3090, and the cap decides § How fast does each one write? 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 two-used-3080s-priced · section id #how-fast-does-each-one-write
Does not fit	250 W each — 500 W between them65,536
20 GB between two cards does not hold the dense model at this window; the pair's own ceiling is the 49,152 row above from Two 3080s against one 3090, and the cap decides § How fast does each one write? 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 two-used-3080s-priced · section id #how-fast-does-each-one-write
44.7tok/s†	300 W each — 600 W between them49,152
from Two 3080s against one 3090, and the cap decides § How fast does each one write? 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 two-used-3080s-priced · section id #how-fast-does-each-one-write
Does not fit	300 W each — 600 W between them65,536
from Two 3080s against one 3090, and the cap decides § How fast does each one write? 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 two-used-3080s-priced · section id #how-fast-does-each-one-write
Mistral Small 3.2 the page says: mistral-small3.2:24b joules per 1,000 tokensJ / 1k tokQ4_K_M · ollama · 0.32.13 · each rig's headline window · 256 tokens out · 2026-09-17 · concurrency not stated
10,442J / 1k tok	250 W each — 500 W between them
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
11,385J / 1k tok	300 W each — 600 W between them
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
not stated decode once the window is full, tokens a secondtok/sQ4_K_M · ollama · 0.32.13 · 300 W each — 600 W between them · One 131,072-token window with about 98,000 tokens really in it · 256 tokens out · 2026-09-17 · concurrency not stated
61.5 tok/s
from Two 3080s against one 3090, and the cap decides § What about a long document? 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 two-used-3080s-priced · section id #what-about-a-long-document
not stated reading 98,024 tokens, secondssQ4_K_M · ollama · 0.32.13 · 300 W each — 600 W between them · One 131,072-token window with about 98,000 tokens really in it · 256 tokens out · 2026-09-17 · concurrency not stated
28.36 s
from Two 3080s against one 3090, and the cap decides § What about a long document? 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 two-used-3080s-priced · section id #what-about-a-long-document
not stated time to the first token, secondssQ4_K_M · ollama · 0.32.13 · 300 W each — 600 W between them · One 131,072-token window with about 98,000 tokens really in it · 256 tokens out · 2026-09-17 · concurrency not stated
30.04 s
from Two 3080s against one 3090, and the cap decides § What about a long document? 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 two-used-3080s-priced · section id #what-about-a-long-document
one 26-billion model split across both cards tokens a second each caller, two callerstok/sQ4_K_M · ollama · 0.32.13 · 300 W each — 600 W between them · 256 tokens out · two callers · Measured 2026-09-17 01:17 to 01:22 UTC · context not stated
94.8tok/s
from Two 3080s against one 3090, and the cap decides § Two copies, or one bigger model? 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 two-used-3080s-priced · section id #two-copies-or-one-bigger-model
one 26-billion model split across both cards total tokens a second, two callerstok/sQ4_K_M · ollama · 0.32.13 · 300 W each — 600 W between them · 256 tokens out · two callers · Measured 2026-09-17 01:17 to 01:22 UTC · context not stated
189.5tok/s
from Two 3080s against one 3090, and the cap decides § Two copies, or one bigger model? 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 two-used-3080s-priced · section id #two-copies-or-one-bigger-model
two copies of a 12-billion model, one per card tokens a second each caller, two callerstok/sQ4_K_M · ollama · 0.32.13 · 300 W each — 600 W between them · 256 tokens out · two callers · Measured 2026-09-17 01:17 to 01:22 UTC · context not stated
59.1tok/s
from Two 3080s against one 3090, and the cap decides § Two copies, or one bigger model? 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 two-used-3080s-priced · section id #two-copies-or-one-bigger-model
two copies of a 12-billion model, one per card total tokens a second, two callerstok/sQ4_K_M · ollama · 0.32.13 · 300 W each — 600 W between them · 256 tokens out · two callers · Measured 2026-09-17 01:17 to 01:22 UTC · context not stated
118.2tok/s
from Two 3080s against one 3090, and the cap decides § Two copies, or one bigger model? 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 two-used-3080s-priced · section id #two-copies-or-one-bigger-model

Where one condition moved, and nothing else

a gate-shaped call, median / p95 — concurrency apart

The same mistral-small3.2:24b

450 ms / 486 ms — 1 caller · 1 parallel slot The slot trap, if you serve other people
709 ms / 712 ms — 1 caller · 4 parallel slots The slot trap, if you serve other people
782 ms / 1,098 ms — 4 callers · 1 parallel slot The slot trap, if you serve other people
1,076 ms / 7,599 ms — 4 callers · 4 parallel slots The slot trap, if you serve other people

Every row above is the same card and the same mistral-small3.2:24b. The only thing that moved is the concurrency. That is why nothing on this page is ranked, and why a condition is part of a row rather than a heading above one.

core temp max — power cap apart

The same gemma4:26b

46 · 42 °C — 250 W each — 500 W between them Power, heat, and the fifty watts
59 · 58 °C — 300 W each — 600 W between them Power, heat, and the fifty watts

Pictures — seconds per render

2 rows — same rule as above.

value	what moves inside this group
not stated one picture with the text encoder moved to the second cardComfyUI · 0.21.1 · 250 W each — 500 W between them · 2026-09-17 · quantisation, context, batch or steps and concurrency not stated
from 3.462 s to 2.261 s — 1.53×
the same bytes and the same picture — the page's own words, and it says this is more than the whole generational gap above from Two 3080s against one 3090, and the cap decides § Renders: one 3080 against one 3090, seconds per image typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #renders-one-3080-against-one-3090-seconds-per-image
not stated two pictures at once, images a minuteComfyUI · 0.21.1 · 250 W each — 500 W between them · 2026-09-17 · quantisation, context, batch or steps and concurrency not stated
24.96 images a minute against 16.37 for one of them alone, 1.52×
from Two 3080s against one 3090, and the cap decides § Renders: one 3080 against one 3090, seconds per image typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #renders-one-3080-against-one-3090-seconds-per-image

Power, temperature, and what it costs

12 rows — same rule as above.

value	what moves inside this group
gemma4:26b mean drawWQ4_K_M · ollama · 0.32.13 · each rig's headline window · 256 tokens out · 2026-09-17 · concurrency not stated
290 W	250 W each — 500 W between them
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
346 W	300 W each — 600 W between them
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
gemma4:26b temperature on the quietest arm, with neither fan turning°CQ4_K_M · ollama · 0.32.13 · 250 W each — 500 W between them · each rig's headline window · 256 tokens out · 2026-09-17 · concurrency not stated
40 to 46 °C
the page calls this the most surprising measurement on it: two five-year-old gaming cards writing in silence from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
Mistral Small 3.2 the page says: mistral-small3.2:24b mean drawWQ4_K_M · ollama · 0.32.13 · each rig's headline window · 256 tokens out · 2026-09-17 · concurrency not stated
466 W	250 W each — 500 W between them
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
509 W	300 W each — 600 W between them
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
Mistral Small 3.2 the page says: mistral-small3.2:24b what fifty watts are worth, denseQ4_K_M · ollama · 0.32.13 · 250 W each — 500 W between them · 256 tokens out · 2026-09-17 · context and concurrency not stated
0.2 per cent†
from Two 3080s against one 3090, and the cap decides § How fast does each one write? typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #how-fast-does-each-one-write
not stated energy at depth, joules per 1,000 tokensJ / 1k tokQ4_K_M · ollama · 0.32.13 · 300 W each — 600 W between them · One 131,072-token window with about 98,000 tokens really in it · 256 tokens out · 2026-09-17 · concurrency not stated
5,837J / 1k tok
from Two 3080s against one 3090, and the cap decides § What about a long document? 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 two-used-3080s-priced · section id #what-about-a-long-document
not stated the exhibit's own energy cost against the single cardQ4_K_M · ollama · 0.32.13 · 300 W each — 600 W between them · each rig's headline window · 256 tokens out · 2026-09-17 · concurrency not stated
the pair spends 69.0 and 87.9 per cent more for the same thousand tokens
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
not stated what fifty watts buy on the pairQ4_K_M · ollama · 0.32.13 · 250 W each — 500 W between them · 300 W each — 600 W between them · each rig's headline window · 256 tokens out · 2026-09-17 · concurrency not stated
0.5 per cent on the mixture and 0.2 on the dense model†
the page's own heading for this reading is that on the pair, fifty watts buy nothing from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
not stated what fifty watts cost the pair in energyQ4_K_M · ollama · 0.32.13 · 250 W each — 500 W between them · 300 W each — 600 W between them · each rig's headline window · 256 tokens out · 2026-09-17 · concurrency not stated
18 and 9 per cent more energy per thousand tokens
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
one 26-billion model split across both cards mean drawWQ4_K_M · ollama · 0.32.13 · 300 W each — 600 W between them · 256 tokens out · two callers · Measured 2026-09-17 01:17 to 01:22 UTC · context not stated
356 W
from Two 3080s against one 3090, and the cap decides § Two copies, or one bigger model? 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 two-used-3080s-priced · section id #two-copies-or-one-bigger-model
two copies of a 12-billion model, one per card mean drawWQ4_K_M · ollama · 0.32.13 · 300 W each — 600 W between them · 256 tokens out · two callers · Measured 2026-09-17 01:17 to 01:22 UTC · context not stated
542 W
from Two 3080s against one 3090, and the cap decides § Two copies, or one bigger model? 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 two-used-3080s-priced · section id #two-copies-or-one-bigger-model

Memory — what fits, and at which context

1 row — same rule as above.

value	what moves inside this group
not stated card board power, the model residentW250 W each — 500 W between them · nothing generating · 2026-09-17, 01:43 to 01:50 · quantisation, runtime, context and concurrency not stated
85.7 W
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts 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 two-used-3080s-priced · section id #power-heat-and-the-fifty-watts

Everything else this shelf measured for it

36 rows — same rule as above.

value	what moves inside this group
gemma4:26b core temp max°CQ4_K_M · ollama · 0.32.13 · each rig's headline window · 256 tokens out · 2026-09-17 · concurrency not stated
46 · 42 °C	250 W each — 500 W between them
per card, the re-padded Gigabyte board first and the stock EVGA board second — the page's own column note from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
59 · 58 °C	300 W each — 600 W between them
per card, the re-padded Gigabyte board first and the stock EVGA board second — the page's own column note from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
gemma4:26b fan max%Q4_K_M · ollama · 0.32.13 · each rig's headline window · 256 tokens out · 2026-09-17 · concurrency not stated
0 % · 0 %	250 W each — 500 W between them
per card, the re-padded Gigabyte board first and the stock EVGA board second — the page's own column note from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
64 % · 0 %	300 W each — 600 W between them
per card, the re-padded Gigabyte board first and the stock EVGA board second — the page's own column note from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
gemma4:26b largest window an unquantised cache leaves the pairQ4_K_M · f16 · ollama · 0.32.13 · 2026-09-17 · power cap, context, batch or steps and concurrency not stated
holds 65,536†
from Two 3080s against one 3090, and the cap decides § How much can each one hold? typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #how-much-can-each-one-hold
gemma4:26b largest window held whole, and the memory it used, the layer count namedQ4_K_M · ollama · 0.32.13 · 256 tokens out · 2026-09-17 · power cap, context and concurrency not stated
131,072 whole, 19,531 MiB used
from Two 3080s against one 3090, and the cap decides § How much can each one hold? 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 two-used-3080s-priced · section id #how-much-can-each-one-hold
gemma4:26b memory left unused under the server's own placementQ4_K_M · ollama · 0.32.13 · 256 tokens out · 2026-09-17 · power cap, context and concurrency not stated
about 5 GB of the pair's 20 unused
from Two 3080s against one 3090, and the cap decides § How much can each one hold? typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #how-much-can-each-one-hold
gemma4:26b tok/stok/sQ4_K_M · ollama · 0.32.13 · each rig's headline window · 256 tokens out · 2026-09-17 · concurrency not stated
114.8tok/s†	250 W each — 500 W between them
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
115.4tok/s†	300 W each — 600 W between them
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
Mistral Small 3.2 the page says: mistral-small3.2:24b a gate-shaped call, median / p95Q4_K_M · ollama · 0.32.13 · eight tokens out · 2026-09-17 · power cap and context not stated
450 ms / 486 ms	1 caller · 1 parallel slot
from Two 3080s against one 3090, and the cap decides § The slot trap, if you serve other people read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #the-slot-trap-if-you-serve-other-people
709 ms / 712 ms	1 caller · 4 parallel slots
from Two 3080s against one 3090, and the cap decides § The slot trap, if you serve other people read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #the-slot-trap-if-you-serve-other-people
782 ms / 1,098 ms	4 callers · 1 parallel slot
from Two 3080s against one 3090, and the cap decides § The slot trap, if you serve other people read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #the-slot-trap-if-you-serve-other-people
1,076 ms / 7,599 ms	4 callers · 4 parallel slots
from Two 3080s against one 3090, and the cap decides § The slot trap, if you serve other people read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #the-slot-trap-if-you-serve-other-people
Mistral Small 3.2 the page says: mistral-small3.2:24b core temp max°CQ4_K_M · ollama · 0.32.13 · each rig's headline window · 256 tokens out · 2026-09-17 · concurrency not stated
58 · 52 °C	250 W each — 500 W between them
per card, the re-padded Gigabyte board first and the stock EVGA board second — the page's own column note from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
69 · 58 °C	300 W each — 600 W between them
per card, the re-padded Gigabyte board first and the stock EVGA board second — the page's own column note from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
Mistral Small 3.2 the page says: mistral-small3.2:24b fan max%Q4_K_M · ollama · 0.32.13 · each rig's headline window · 256 tokens out · 2026-09-17 · concurrency not stated
62 % · 0 %	250 W each — 500 W between them
per card, the re-padded Gigabyte board first and the stock EVGA board second — the page's own column note from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
74 % · 0 %	300 W each — 600 W between them
per card, the re-padded Gigabyte board first and the stock EVGA board second — the page's own column note from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
Mistral Small 3.2 the page says: mistral-small3.2:24b largest window held whole, and the memory it used, the layer count namedQ4_K_M · ollama · 0.32.13 · 256 tokens out · 2026-09-17 · power cap, context and concurrency not stated
49,152 whole, 18,909 MiB used
from Two 3080s against one 3090, and the cap decides § How much can each one hold? 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 two-used-3080s-priced · section id #how-much-can-each-one-hold
Mistral Small 3.2 the page says: mistral-small3.2:24b residency under the server's own placementQ4_K_M · ollama · 0.32.13 · 2026-09-17 · power cap and context not stated
73.7 %	eight tokens out1 parallel slot
from Two 3080s against one 3090, and the cap decides § The slot trap, if you serve other people read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #the-slot-trap-if-you-serve-other-people
41.7 %	eight tokens out4 parallel slots
from Two 3080s against one 3090, and the cap decides § The slot trap, if you serve other people read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #the-slot-trap-if-you-serve-other-people
Never measured as a ladder; at a 49,152 window the planner kept 73.7 % on the cards	256 tokens outnot stated
from Two 3080s against one 3090, and the cap decides § How much can each one hold? 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 two-used-3080s-priced · section id #how-much-can-each-one-hold
Mistral Small 3.2 the page says: mistral-small3.2:24b the exhibit's own margin over the single card, denseQ4_K_M · ollama · 0.32.13 · 250 W each — 500 W between them · 49,152 · 256 tokens out · 2026-09-17 · concurrency not stated
34.9 per cent
the page states this as the single card's loss; it is the pair's one decode win and the page calls it its turning point from Two 3080s against one 3090, and the cap decides § How fast does each one write? typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #how-fast-does-each-one-write
Mistral Small 3.2 the page says: mistral-small3.2:24b tok/stok/sQ4_K_M · ollama · 0.32.13 · each rig's headline window · 256 tokens out · 2026-09-17 · concurrency not stated
44.6tok/s†	250 W each — 500 W between them
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
44.7tok/s†	300 W each — 600 W between them
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts read off that page’s own table by the roster’s reader, then checked against it word for word · the card is named in this row · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
not stated card board power, cards emptyW250 W each — 500 W between them · nothing generating · 2026-09-17, 01:43 to 01:50 · quantisation, runtime, context and concurrency not stated
44.0 W
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts 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 two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
not stated left on continuously, a day and a year250 W each — 500 W between them · nothing generating · 2026-09-17, 01:43 to 01:50 · quantisation, runtime, context and concurrency not stated
0.78 kWh a day and 286 kWh a year
the difference between the two rigs held idle, not either rig's own consumption from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
not stated left on continuously, in money250 W each — 500 W between them · nothing generating · 2026-09-17, 01:43 to 01:50 · quantisation, runtime, context and concurrency not stated
about $52 a year, or $4.37 a month
at 18.34 cents a kilowatt-hour, the United States residential average for June 2026 as the page names it — and it is the DIFFERENCE between the two rigs held idle, card board power only, so a wall meter would read more from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
not stated the build, totalread between 15:00 and 15:30 UTC on 17 September 2026 · quantisation, runtime, power cap, context, batch or steps and concurrency not stated
$1,483.80
the record behind this figure is public — the listing its page prints in full, read in the window that page names from Two 3080s against one 3090, and the cap decides § What it costs, priced 17 September 2026 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 two-used-3080s-priced · section id #what-it-costs-priced-17-september-2026
$1,488.81
the record behind this figure is public — the listing its page prints in full, read in the window that page names from Two 3080s against one 3090, and the cap decides § What it costs, priced 17 September 2026 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 two-used-3080s-priced · section id #what-it-costs-priced-17-september-2026
$1,578.80
the record behind this figure is public — the listing its page prints in full, read in the window that page names from Two 3080s against one 3090, and the cap decides § What it costs, priced 17 September 2026 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 two-used-3080s-priced · section id #what-it-costs-priced-17-september-2026
$1,080.83
the record behind this figure is public — the listing its page prints in full, read in the window that page nameswhat a thousand dollars buys, and nothing else: no case, no drive, no operating system, and no shipping — the page's own list of what is still to add from Two 3080s against one 3090, and the cap decides § What it costs, priced 17 September 2026 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 two-used-3080s-priced · section id #what-it-costs-priced-17-september-2026
$1,393.81
the record behind this figure is public — the listing its page prints in full, read in the window that page namesthe cheapest combination the page totals from Two 3080s against one 3090, and the cap decides § What it costs, priced 17 September 2026 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 two-used-3080s-priced · section id #what-it-costs-priced-17-september-2026
not stated the exhibit's own ratio on the readQ4_K_M · ollama · 0.32.13 · 300 W each — 600 W between them · One 131,072-token window with about 98,000 tokens really in it · 256 tokens out · 2026-09-17 · concurrency not stated
1.59×
from Two 3080s against one 3090, and the cap decides § What about a long document? typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #what-about-a-long-document
not stated the exhibit's own ratio, one big model against two copiesQ4_K_M · ollama · 0.32.13 · 300 W each — 600 W between them · 256 tokens out · two callers · Measured 2026-09-17 01:17 to 01:22 UTC · context not stated
60.3 per cent
from Two 3080s against one 3090, and the cap decides § Two copies, or one bigger model? typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #two-copies-or-one-bigger-model
not stated what the model itself costs to holdW250 W each — 500 W between them · nothing generating · 2026-09-17, 01:43 to 01:50 · quantisation, runtime, context and concurrency not stated
41.7 W
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts 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 two-used-3080s-priced · section id #power-heat-and-the-fifty-watts
not stated what the pair costs over the single card to sit there holding a modelW250 W each — 500 W between them · nothing generating · 2026-09-17, 01:43 to 01:50 · quantisation, runtime, context and concurrency not stated
32.6 W more than the single card
from Two 3080s against one 3090, and the cap decides § Power, heat, and the fifty watts typed in from that page by hand, then checked against it word for word · the card is named elsewhere on that page · read from two-used-3080s-priced · section id #power-heat-and-the-fifty-watts

Where it appears

Every released exhibit that names this identity, in the shelf’s own order.

Electricity per million tokens, card by card — named, and no figure carried from it — yet
Six graphics cards, the same questions — named, and no figure carried from it — yet
Two 3080s against one 3090, and the cap decides — 81 figures on this page

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.

two 3080s