{"rewrite":{"id":"r_0609d6de68c8334da87cc722","clusterId":"c_990865c2e71e97166e606bad","slug":"strata-runs-deepseek-v4-1-flash-on-a-64gb-pc","model":"deepseek-v4-1-flash","headline":"Strata Runs DeepSeek-V4.1-Flash On A 64GB PC","summary":"Niko Veit, developer of the open-source inference engine Strata, said on October 24 that the DeepSeek-V4.1-Flash model now runs on a PC with 64GB of main memory. The quantized release uses the GSQ-RCO compression method from DASLab at the Institute of Science and Technology Austria. Strata spreads model data across VRAM, main memory, and SSD. The 2-bit build measured 4 to 6 tokens per second on a GeForce RTX 5070 with 64GB of memory.","whyItMatters":"Strata's pitch is that memory and storage, not VRAM alone, decide whether a 763-billion-parameter model can run at home, and the measurement puts that claim in numbers, with several hundred GB of disk as the practical catch.","webCardHtml":"\u003cp\u003eThe provider of the quantized builds says the 2-bit version matched the original model\u0026#39;s average score on three evaluations: math, science, and code generation. That claim matters more than the token rate, because a checkpoint that loses accuracy at 2 bits would make the compression interesting and not much else.\u003c/p\u003e\u003cp\u003eThe three published models average 1.75, 2, and 2.25 bits of weight precision. Their download sizes, related files included, are about 239GB, 257GB, and 274GB. The 2-bit version needs roughly 265GB of free space. The measurement used Strata 0.2.5 with the 12GB of dedicated memory on a GeForce RTX 5070 and 64GB of main memory, and it generated 4 to 6 tokens per second.\u003c/p\u003e","blueskyPost":"Strata's 4 to 6 tokens per second on a GeForce RTX 5070 is the number that matters. The 64GB footprint is split across VRAM, main memory and SSD, so the model streams rather than fits.","twitterPost":"Strata streams the DeepSeek-V4.1-Flash model across VRAM, main memory and SSD, and 4 to 6 tokens per second is what the RTX 5070 split costs.","threadsPost":"Strata's GSQ-RCO compression from DASLab makes DeepSeek-V4.1-Flash fit a 64GB PC, but the split across VRAM, main memory and SSD is what sets the 4 to 6 tokens per second on a GeForce RTX 5070. The model streams from storage instead of fitting in one pool.","newsletterBlurb":"Niko Veit, developer of the open-source inference engine Strata, said on October 24 that DeepSeek-V4.1-Flash can run on a PC with 64GB of main memory. The quantized builds use GSQ-RCO compression from DASLab at the Institute of Science and Technology Austria and arrive in three weight precisions. The 2-bit version measured 4 to 6 tokens per second on a GeForce RTX 5070 with 64GB of memory, and its provider says it matched the original model's average scores on math, science, and code generation.","attributionJson":"[{\"source\":\"ASCII.jp\",\"url\":\"https://ascii.jp/elem/000/004/445/4445574/?rss\",\"title\":\"What?! Running DeepSeek's 763-billion-parameter model on a 64GB PC!? Strata finally makes it possible\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":4826,"outputTokens":769,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1792979280,"createdAt":"2026-10-26T01:43:42.000Z","publishedAt":"2026-10-26T01:46:58.000Z","updatedAt":"2026-10-26T01:46:58.000Z"},"cluster":{"id":"c_990865c2e71e97166e606bad","canonicalTitle":"えっ!　メモリ64GBのPCで7630億パラメーターのDeepSeekを!?　Strataがついに実現","representativeArticleId":"a_3cfb47828f2837e2230a6d95","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"DeepSeek\"],\"studios\":[],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-10-26T01:00:00.000Z","lastSeenAt":"2026-10-26T01:00:00.000Z","updatedAt":"2026-10-26T01:46:58.000Z"},"attribution":[{"source":"ASCII.jp","url":"https://ascii.jp/elem/000/004/445/4445574/?rss","title":"えっ!　メモリ64GBのPCで7630億パラメーターのDeepSeekを!?　Strataがついに実現"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["DeepSeek"],"studios":[],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":null,"pageLinks":{"healthSlugs":["deepseek"],"storySlugs":[],"entities":[{"slug":"deepseek","supported":0,"disputed":0,"contradicted":0,"unverified":6,"overall":"unverified","story_count":5,"has_dossier":true}],"stories":[],"entitiesOk":true,"storiesOk":true}}