{
 "schema_version": "1.0",
 "page_kind": "article",
 "pack_kind": "generate",
 "slug": "two-3090s-out-one-3090-ti-in",
 "title": "Two 3090s out, one 3090 Ti in",
 "dek": "Two GeForce RTX 3090s came out of service and one EVGA GeForce RTX 3090 Ti FTW3 Ultra went in; a print lab and a beat lab moved with them on 2026-09-18 (UTC), with nothing closed. Every figure here is either a bench reading from the companion page or a line from a service's own journal, and every row says which.",
 "published": "2026-09-19",
 "series": [
  "notes"
 ],
 "licence": "CC BY 4.0",
 "status": "pending-judge",
 "pack_note": "no judge is seated, so every chip is held",
 "notice": "items with published:false are held and are not this page's published words",
 "source": {
  "url": "https://research.strata2signal.com/two-3090s-out-one-3090-ti-in/",
  "md_url": "https://research.strata2signal.com/two-3090s-out-one-3090-ti-in/index.md",
  "md_sha": "a8999bc6c673f6f8439976c6058540b1384bd0f85a16243ff570a95b8b54b1e7",
  "html_sha": "0400c07b30b3115cf2df472c9a342c57663956419573e971d5b76725fe2512e0",
  "anchors_sha": "0517b5fbab97335c54afd487ea961636bfc8e5c54d38677e2294e539f55d7a65"
 },
 "short": {
  "paragraph": "On 2026-09-18 the print lab's renderer and the beat lab's sleeve painter moved off a machine that held two GeForce RTX 3090s, one of which did all the drawing at a 280 W cap, and onto one that holds a single EVGA GeForce RTX 3090 Ti FTW3 Ultra at a 350 W cap. One render never splits across two cards, so those two were never a rendering pair; the second card served language until 2026-09-08 and drew nothing after. The first seven creations on the new card drew their plates in a median of 15.5 seconds, half a per cent over what the bench predicted and 11.5 to 18.1 per cent under what the old machine's cap bought; the first two sleeves painted in 2.1 seconds each, against a median of 2.6 over the previous week on the old card. The old machine now serves nothing and was powered off the same evening.",
  "from_dek": false,
  "counts": {
   "words": 1507,
   "minutes": 7,
   "tables": 2,
   "kit": false
  },
  "bullets": []
 },
 "sections": [
  {
   "id": "what-moved-and-from-where",
   "heading": "What moved, and from where",
   "level": 2,
   "span": [
    1607,
    2895
   ],
   "chunks": [
    [
     1607,
     2895
    ]
   ],
   "chars": 1288,
   "digest": "the print lab and the beat lab moved from a machine with two 3090s to a single machine with an EVGA GeForce RTX 3090 Ti FTW3 Ultra 24 GB on 2026-09-17. the new machine runs at a 350 W cap set at boot.",
   "digest_skipped": null
  },
  {
   "id": "why-this-card-and-why-350-w",
   "heading": "Why this card, and why 350 W",
   "level": 2,
   "span": [
    2895,
    4418
   ],
   "chunks": [
    [
     2895,
     4418
    ]
   ],
   "chars": 1523,
   "digest": "a 3090 Ti at 350 W draws the heavy plate 9.5 per cent faster than a 3090 at 350. ten minutes of continuous drawing at 350 W put this Ti at a 55 °C plateau and a 60 °C peak.",
   "digest_skipped": null
  },
  {
   "id": "before-and-after-on-the-same-instruments",
   "heading": "Before and after, on the same instruments",
   "level": 2,
   "span": [
    4418,
    6618
   ],
   "chunks": [
    [
     4418,
     6618
    ]
   ],
   "chars": 2200,
   "digest": "bench figures and live figures are kept apart. the print lab's live median is 15.47 seconds for a batch of three. the painter's after readings are 17 and 19 per cent under the old median.",
   "digest_skipped": null
  },
  {
   "id": "what-the-sign-says",
   "heading": "What the sign says",
   "level": 2,
   "span": [
    6618,
    7598
   ],
   "chunks": [
    [
     6618,
     7598
    ]
   ],
   "chars": 980,
   "digest": "the print lab's queue line tells a visitor \"about N seconds\" based on the median of the last twenty completed creations. the sign re-based itself on new hardware because it knows which card the work is going to.",
   "digest_skipped": null
  },
  {
   "id": "how-the-move-was-done",
   "heading": "How the move was done",
   "level": 2,
   "span": [
    7598,
    8701
   ],
   "chunks": [
    [
     7598,
     8701
    ]
   ],
   "chars": 1103,
   "digest": "the render leg was moved first with both machines alive. the painter leg followed by adding two network grants and re-pointing the beat lab's engine. finally, the old machine was torn down.",
   "digest_skipped": null
  },
  {
   "id": "what-this-does-not-measure",
   "heading": "What this does not measure",
   "level": 2,
   "span": [
    8701,
    10161
   ],
   "chunks": [
    [
     8701,
     10161
    ]
   ],
   "chars": 1460,
   "digest": "this section notes that whole-box power, memory-junction temperature, and the old machine's own live times are not measured.",
   "digest_skipped": null
  }
 ],
 "chips": [
  {
   "id": "c-ec4a5868",
   "q": "how do the print lab and beat lab services function?",
   "a": "The print lab turns words into three plates using a diffusion model, while the beat lab draws a two-JPEG sleeve for every groove posted to the board.",
   "cites": [
    "what-moved-and-from-where"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-9d209ce2",
   "q": "what factors determined the choice of the new power cap?",
   "a": "A comparison of performance at four different power caps showed that a 350 W setting provided a significant speed increase over 300 W, while the next 50 W increment only yielded a 2 per cent gain.",
   "cites": [
    "why-this-card-and-why-350-w"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-a959fe27",
   "q": "how much faster is the new setup compared to the old one?",
   "a": "The new machine is 11.5 to 18.1 per cent faster than the old machine at its 280 W cap, and 8.5 to 9.5 per cent faster than a 3090 would have been at a matched cap.",
   "cites": [
    "before-and-after-on-the-same-instruments"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-2e0cf542",
   "q": "how does the print lab queue sign calculate wait times?",
   "a": "The sign multiplies the number of creations in the queue by the median of the last twenty completed creations on the specific card that will perform the drawing.",
   "cites": [
    "what-the-sign-says"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-35d7b0d0",
   "q": "what steps were taken to transition the rendering service?",
   "a": "The new instance was verified with render receipts, set up as a backup, and then promoted to primary after the old primary was moved to a backup role.",
   "cites": [
    "how-the-move-was-done"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  },
  {
   "id": "c-63c13ed8",
   "q": "what specific thermal data remains unavailable?",
   "a": "The memory-junction temperature for the GDDR6X sensor cannot be read through the current driver, so only the core sensor and fan data are provided.",
   "cites": [
    "what-this-does-not-measure"
   ],
   "published": false,
   "publish_state": "held",
   "quote": "",
   "quote_cite": ""
  }
 ],
 "related": [
  {
   "slug": "one-3090-against-one-3090-ti",
   "why": "shares ground with § Why this card, and why 350 W · § At a glance"
  },
  {
   "slug": "two-used-3080s-priced",
   "why": "shares ground with § What this does not measure · § What this page does not measure"
  },
  {
   "slug": "a-short-history-of-mistral",
   "why": "shares ground with § Why this card, and why 350 W · § What a power cap actually costs"
  }
 ],
 "thanks": "",
 "kit": null,
 "seat_class": {
  "writer": "gemma-class",
  "judge": null,
  "writer_runtime": "vllm"
 },
 "bench": {
  "bullets_written": 3,
  "bullets_kept": 0,
  "digests_written": 6,
  "digests_kept": 6,
  "chips_written": 6,
  "chips_kept": 6,
  "chips_grounded": 0,
  "chips_published": 0,
  "dropped_by": {
   "new_noun": 0,
   "figure": 0,
   "length": 0,
   "cite": 0,
   "judge": 0,
   "quote": 0,
   "directive": 0,
   "redaction": 0,
   "profanity": 0,
   "identifier": 0,
   "bare_figure": 1,
   "house_voice": 0,
   "hand-edit": 2
  },
  "new_noun_tokens_checked": [
   "11.5",
   "15.47",
   "15.5",
   "17",
   "18.1",
   "19",
   "2",
   "2026-09-17",
   "24",
   "280",
   "300",
   "3090",
   "3090s",
   "350",
   "50",
   "55",
   "60",
   "8.5",
   "9.5",
   "C",
   "EVGA",
   "FTW3",
   "GB",
   "GDDR6X",
   "GeForce",
   "N",
   "RTX",
   "Ti",
   "Ultra",
   "W"
  ],
  "new_noun_tokens_withheld": 0,
  "figure_definition": "v3",
  "drops": [
   {
    "kind": "bullet",
    "reason": "bare_figure",
    "item": "the sign's median is 15.5",
    "detail": "the cited section writes this figure as '15.5 seconds' and the bullet prints '15.5' alone"
   },
   {
    "kind": "bullet",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 31,
    "withheld": "a published figure left standing after an indefinite article with no noun",
    "detail": "Removed by hand 2026-09-19 at the pack cut. The bullet ended \"...runs at a 350 W\" where the article writes \"at a 350 W cap\": the figure kept its unit and lost the noun the unit describes, leaving an indefinite article with nothing after it. No gate owns that shape — gate 8 reads the cited section's prose for a unit and finds one, gate 7 reads it for an identifier and finds none — and it is the defect class of the two hand removals that became gate 8, one word further along. Recorded rather than cured in place: nothing here rewords a line the writer wrote."
   },
   {
    "kind": "bullet",
    "reason": "hand-edit",
    "item": null,
    "item_chars": 35,
    "withheld": "a published figure that lost the unit the article gave it",
    "detail": "Removed by hand 2026-09-19 at the pack cut. The bullet ended \"...readings are 17\" where the article writes \"two readings: 17 and 19 per cent under the old median\". gate8_bare_figure did not fire: it passes a figure the cited section's prose prints bare anywhere in it, and this section's prose prints \"17 and\" — the first of two readings sharing one trailing unit — so the escape that lets a genuinely unitless count through let this one through as well. The recorded span is the second half of it: the span search anchors on the digits and takes the first match in the section, which here is the \"17\" inside a table cell's \"17.01\", so the receipt's path back pointed at bytes that are not a figure on the page at all. The rule the bullet broke is D-20260918-72's — a bullet may not lose the unit the article gave its figure — so the bullet goes and both narrowings stay as ruled."
   }
  ]
 },
 "generated_utc": "2026-09-19T03:40:11Z"
}