[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-modders-unlock-half-speed-ai-cap-on-nvidia-rtx-20-30-40-gpus":10,"sections":35},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},12385,"modders-unlock-half-speed-ai-cap-on-nvidia-rtx-20-30-40-gpus","Modders Unlock Half-Speed AI Cap on Nvidia RTX 20, 30, 40 GPUs","A firmware exploit restores full-rate matrix math on GeForce cards, though real workloads gain far less than the headline doubling.","Modders have found a firmware exploit that removes a half-speed limit on AI math for Nvidia's RTX 20, 30 and 40-series GPUs. On paper, an RTX 4090 doubles its peak tensor rate. In practice, the gains are much smaller.\n\nNvidia's own published specs show GeForce cards running FP32-accumulate matrix math at half their FP16-accumulate rate. That cap does not apply to the Turing and Ampere pro cards. Three projects now lift it, each with an unofficial Linux module: gpu-unlock for Ampere, gpu-unlock-ada for the RTX 4090 and a few experimental Ada cards, and turing-unlock for the RTX 2080 Ti.\n\nThe peak numbers are striking. The Ada fork developer's RTX 4090 went from 165.6 to 331.6 TFLOPS on FP16. Real benchmarks are messier. One owner's RTX 4090 gained 12 to 34% on FP16 matrix multiplication, and a tested RTX 3090 Ti gained 35% on FP16 and 80% on BF16. User reports for other Ampere cards range from 25% to about 81%, and the lowest figure came from a card capped at 225 W. Power limits matter a lot here. Ampere prompt processing improved 14 to 36%, while chatbot token speeds stayed about the same because they depend on memory bandwidth. One project's own documentation disclaims any 2x speedup in games or applications.\n\nThe mechanism is the interesting part. The rate limits sit in two registers the host driver cannot write. Nvidia's signed SEC2 Booter firmware contains a stack buffer overflow, which the modders use to run privileged code with a four-byte write. The flaw surfaced after Nvidia published a debug Booter encrypted with a known public test key. Researcher Jon Pry described the underlying bug in a June paper, found on the CMP 170HX mining card, and that card's unlocker is the best-known use of the exploit.\n\nThe pro-card exemption suggests the cap is a product-tier decision rather than a hardware limit. That also explains why the fix lives in firmware rather than silicon.\n\nThe risks are real. Two of the projects warn the unlock can hang a GPU, block booting, corrupt data or damage hardware. Blackwell is unsupported, and gpu-unlock's developer calls Hopper and Blackwell out of scope for this approach. No project has published a roadmap, but gpu-unlock is collecting compatibility reports from owners.","[\"nvidia\",\"gpu\",\"firmware\",\"ai-hardware\"]","2026-10-26T11:00:00.000Z","2026-10-26T11:28:14.004Z","2026-10-26T11:28:14.224Z","published",null,[],"hardware",[26,27,28,29],"nvidia","gpu","firmware","ai-hardware",[31],{"name":32,"url":33},"Tom's Hardware","https:\u002F\u002Fwww.tomshardware.com\u002Fpc-components\u002Fgpus\u002Fnvidia-rtx-20-30-and-40-series-gpus-hacked-to-remove-a-half-speed-ai-throttle-firmware-exploit-doubles-rtx-4090s-peak-tensor-rate-lifts-rtx-3090-ti-matrix-math-up-to-80-percent",0,{"sections":36},[37,42,47,52,57,62,66,71,76,81,86,91,96,101],{"name":38,"slug":39,"count":40,"latest_published_at":41},"AI","ai",7083,"2026-10-25T15:25:40.000Z",{"name":43,"slug":44,"count":45,"latest_published_at":46},"Security","security",1069,"2026-10-26T12:00:52.000Z",{"name":48,"slug":49,"count":50,"latest_published_at":51},"Policy","policy",589,"2026-10-26T11:05:02.000Z",{"name":53,"slug":54,"count":55,"latest_published_at":56},"Deals","deals",575,"2026-10-26T10:23:06.000Z",{"name":58,"slug":59,"count":60,"latest_published_at":61},"Consumer Tech","consumer-tech",303,"2026-10-26T11:31:21.000Z",{"name":63,"slug":24,"count":64,"latest_published_at":65},"Hardware",297,"2026-10-26T11:24:42.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":70},"Science","science",218,"2026-10-23T22:48:25.000Z",{"name":72,"slug":73,"count":74,"latest_published_at":75},"Startups","startups",148,"2026-10-23T16:26:29.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Software","software",145,"2026-10-26T09:30:00.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"Dev Tools","dev-tools",122,"2026-10-21T16:00:00.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"General","general",104,"2026-10-26T09:26:49.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Gaming","gaming",88,"2026-10-22T22:25:33.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"Reviews","reviews",38,"2026-10-20T12:00:00.000Z",{"name":102,"slug":103,"count":104,"latest_published_at":105},"How-To","how-to",8,"2026-10-05T09:00:00.000Z"]