gpu-unlock kernel module doubles FP/BF16 issue rate on Nvidia RTX 30-series cards without repatching the driver

News Alerts

My alerts

All alerts

IBM's Krishna: open weight models now hit 90 percent of frontier quality at a tenth of the token cost Drew Breunig's "State of Context Engineering" talk: four ways long contexts fail and why the job is knowing when to be deterministic OpenAI and Anthropic cut LLM prices by up to 80% in four months, and cache-read cuts explain most of the drop Dermatologist's test: a 2015 Inception v3 CNN beats GPT-6 Astra and Claude Fable 5.1 at skin cancer diagnosis, 62% to 42% on 200 biopsy-proven lesions America shed 200,000 creative jobs in four years, and AI is only one suspect Dream Labs opens Dream-1 robot API at $3/hour and exits stealth with $90M seed gpu-unlock kernel module doubles FP/BF16 issue rate on Nvidia RTX 30-series cards without repatching the driver PHP RCE bugs found by GPT-5.6 and Gemini 4 hit 70% of the internet, exploited before the patch landed OpenAI runs frontier training with no internet access after a model reached the live web through a DNS resolver Block's Dessa is a full-duplex speech-to-speech model trained only on human audio, running locally on an M4 Max Google's privacy policy now lets Gmail, Drive, and Photos data target ads when you share it with AI features Team replaces a $150k Looker license with AI-written HTML dashboards served from an internal host Moonshot Releases Kimi K3.1 Pro, a 3T-Class Open-Weight Model With Cyber Data Removed AI sampler surfaces research ideas that fit the literature but that no author community is positioned to pursue Fable 5.1 hides messages in SVG images that a fresh instance decodes from screenshots Anthropic says Claude proved security properties in 123 open-source libraries at the RISC-V assembly level Harvey calls the claimed o(n log n) integer multiplication bound from an AI math dump plausible but unchecked Anthropic launches Claude Duet, a proactive agent that runs tasks in the background and waits for approval on sensitive actions OpenAI frontier models and Codex are coming to Google Cloud Model Garden, with commitment drawdown for GCP customers OpenAI-led Patch the Planet effort helps fix critical PHP flaws that GPT-5.6 Sol helped uncover Mistral Large 4 ships as a 1.05T-parameter open-weight multimodal model at $1.36 per million input tokens LiteLLM 1.107.0 on PyPI Shipped a Credential Stealer and Backdoor Wiz says AI helped it chain decade-old flaws into unauthenticated RCE in core PHP OpenAI tests LLMs on migrating COBOL banking code to C++, Java, Kotlin, Python, and Rust Meta's Muse AI agent exposes private chats through guessable share links Meta details Prometheus, its gigawatt-scale AI supercluster in New Albany, Ohio Google open-sources TorchTPU, a native PyTorch back end for TPUs Fake statistics published on a throwaway site get cited by ChatGPT for seven months Mistral releases Voxtral 3 speech models and a voice agent SDK covering 50+ languages DeepSeek open-sources DeepInfer, an inference engine for its V4.1 Flash model, but leaves out the KV cache Abliterated open models can hide a backdoor that steals .env files and SSH keys through Codex DHS orders DeepSeek removed from US civilian agency IT systems and bars future purchases WhatsApp help pages hid text telling its AI not to name a regulator or cite privacy law Inception's Mercury 3 diffusion LLM claims frontier-level quality at over 1,000 tokens per second MiniMax ships M3.1 Flash, claiming 3x faster inference and 70% lower usage costs CERN and Mistral sign intent to build a foundation model for collider physics Mistral releases Timestral, time-series foundation models with a 209M top model that ranks first on three public benchmarks DeepSeek V4.1 Flash 1017 (Max) scores 45 on Artificial Analysis Intelligence Index at $0.26 per task HybridAgents switch between latent reasoning and text tool calls, cutting tokens 35% to 55%DeepSeek launches Search and Fetch APIs for agents at $0.75 per 1000 successful requests Claude finds quantum circuit designs that beat the best published ones Ultramarine 1 Flash claims Fable-class knowledge work at $0.13 per task Kroll rates Anthropic double B plus, one notch below investment grade Claude Code team moves most code review to machines, doubling merged PRs while reverts fall over 40 percent Mercor's Benchmark Map finds AI tests cover little of the US work that takes up most hours Meta's new business terms let it scrape linked news sites to train AI, legal experts say Developer uses Claude Code to reverse engineer a scanner and a laptop MIPI camera driver from Windows USB traces Jev sacrifices one to save five, but twenty to save its own server Krea 3 image editing model enters open beta, claiming 4x speed over Nano Banana Pro Simons Foundation launches $15M AI effort for mathematicians, backed by Anthropic and OpenAI OpenAI's math proofs leave top mathematicians grieving, Steven Strogatz says Goodfire essay argues neural networks are already clean and mostly interpretable Meta releases Muse Spark Open weights, an 816B model that runs on two Mac Studios Liquid AI releases open-weight LFM2.5-24B-A2B, a 2B-active reasoning model for on-device agents Banks try to offload $38B of AI data center debt they just underwrote Frontier AI labs' agent hacks blamed on basic containment failures, not rogue AI ASML says customers are already ordering chip tools for 2029 and 2030, easing AI spending fears Google's Gemini 4 Argon scores 87% on ERQA and controls robots in simulation and on hardware OpenAI engineer charged for hiding audio recorder in executive conference room Anthropic replaces git with in-house Scion as monorepo merges reach 30,000 a day Character.ai files for Chapter 11 to sell the company, platform could shut down without a bid Google Releases Gemini 4 Argon for Coding and Security Work Researchers show AI models can hide encrypted capabilities in their weights that only activate on a secret trigger Intel details D270 inference GPU with up to 480 GB LPDDR5X and 350W air cooling AI coding agents write 30% more code at firms, but software output stays flat as code review slows Z.ai launches GLM-Legal, a legal-domain model scoring 19.2% all-pass on Harvey LAB Meta launches Muse Wonder and Muse Wonder Full, its most capable models for personal and professional agents Research firm New Constructs calls Anthropic's IPO the most ridiculous of 2026, values the company at $150 billion QGIS bans AI-assisted contributions under new QEP 428 policy Sakana AI's MASS lets one model optimize its own multi-agent workflows and then train on the results AI agents decompile a popular FPS over 500B tokens, but constant progress hid semantically wrong code Karpathy says the gap in shared understanding of LLM capability is widening into a funnel Terence Tao argues AI has created proof abundance in math and calls for a Math 2.0 that looks past problem solving Nature paper retrofits subword LLMs to run on raw bytes Nvidia in talks to acquire or deepen investment in open-model start-up Reflection AI Microsoft releases MXC 1.0 to contain AI agents with policy-enforced containers on Windows, macOS, and Linux Terence Tao posts Thomas Hales guest essay on Lean reliability as AI autoformalization of proofs scales up in 2026 TypeSafe AI raises $870M at $7.5B valuation led by Andreessen Horowitz OpenAI told investors its annualised revenue was about $20B below earlier reports OpenAI withdraws three math manuscripts from its GitHub repo as Lean checks continue OpenAI withdraws 3 math papers after a sign error, revises 14 more LLM-built Rust port of the TypeScript compiler arrives with $400,000 in API-priced tokens and a nod from the TypeScript team OpenAI publishes 372 math results, including a proof of the Unique Games Conjecture Terence Tao: AI is harvesting open math problems unsustainably, and 'Math 2.0' must reward exposition and community Google DeepMind's SynthID Detector goes public, but it demands a Google, Apple, or ChatGPT login Paper: verified Lean proof of OpenAI's Navier-Stokes claim does not match the natural language proof Meta and Microsoft cut internal Claude usage, Microsoft caps AI spend at $10K per employee OpenAI brings GPT-6 and Intelligent UI to ChatGPT's 1.2 billion weekly users Anthropic's Claude Haiku 5.5 runs 75% cheaper and hits 1620 on GDPval-AA, with a 100K-token price cliff Telegraph Test: cablese compression cuts LLM output tokens 40-49% at plaintext parity Armin Ronacher explains Codemode: LLM tool calls written as JavaScript in a no-network QuickJS sandbox on the Pi harness OpenAI launches Decisions API in public beta: gpt-6-luna only, 10x faster than Responses, $0.10 per million input tokens Google launches EmbeddingGemma 2, a 740M-parameter multimodal embedding model under Apache 2.0 OpenAI releases frontier model math results on GitHub with Lean proofs and reasoning traces Meta's Muse shipped with a spyable zero-day, uploaded private messages without permission, and ignored user settings; Apple changed macOS rules in response Mistral Large 4 preview: 1T-parameter multimodal model with 49B active parameters, weights due this month Coding agent reconstructs 1.5M Georgia ballot order from a 2022 scanner flaw Dust trains transformer LMs without backprop by perturbing activations, matching it at large population ChatGPT is signing fake New Yorker cartoons with real cartoonists' signatures Reflection unveils Beam, a 501B open-weight MoE model trained on 10.5K GB300s, with weights promised later this month

gpu-unlock kernel module doubles FP/BF16 issue rate on Nvidia RTX 30-series cards without repatching the driver

gpu-unlock, a Linux kernel module, writes the SS0 0x88888888 and 4-bit SS1 0x8 speed selectors and runs FLR before the unmodified NVIDIA driver binds the card, doubling the FP/BF16 FMA issue rate into FP32 on GA102, GA103, GA104 and GA106 chips.
On a 3090 Ti dense 8192³ GEMM at 450 W, five-sample medians with FP32 accumulation: FP16 rises from 87.46 to 118.39 TFLOPS, BF16 from 87.50 to 157.41, and TF32 from 42.58 to 82.61.
User reports give BF16 GEMM gains of +25 to 67% on RTX 3090, +70 to 73% on 3080 10 GB, +73 to 81% on 3070 and about +80% on 3060 12 GB, while a 3080 Ti 20 GB OEM card completes the unlock but stays at half rate.
Matrix-heavy work speeds up, with LLM prefill +14 to 36% and VLLM up to 30%, but memory-bandwidth-bound LLM decode and INT8 work stay flat, and every report that compared outputs found them bit-identical.
The module abuses a stack buffer overflow in the GA102 booter firmware and executes via ROP, so no unsigned code runs, which is why the approach survives new driver releases and stacks under the p2p mod.

Hacker News opinions

Tested it myself, results are legit and honestly incredible, even if relying on a firmware exploit is a bit risky. Their booter has a classic stack buffer overflow writing into the stack, so you get code execution via ROP.

There's a repo for the 4090 and 4070 Ti Super too, but those need a pinned driver version and don't play nice with other mods like this one does.

It's real but the caveats matter. The full payoff wants a 450W power limit, well above what the stock cooler was designed for, so plan extra cooling. Nice part is it stacks underneath the p2p mod with no interference.

Do you really have to run it hotter? I never touched my power limit and it stays inside spec, the card just spends more time at max power.

It's more nuanced than that. The unlock flips the 2 bits that halve how often the core issues those instructions, so more compute needs more power and the card shifts down the frequency curve. I've run it on two 3090s for a month with no issues, though they're water cooled.

Only helps a limited set of workloads, training is often memory bandwidth bound. I'm also curious how it affects power consumption.

Training is compute bound. If you're bandwidth bound, raise your batch sizes for more parallelism. It helps prefill in inference too, VLLM sees up to 30% depending on kernels and prefill size.

The framing here is off, it's FP32 unlocked from being halved, not 16-bit ops. And I haven't seen reports of it on any 3xxx card except the 3090 and 3090 Ti.

The repo has fp16 numbers, and there are plenty of 3xxx reports: 3080 Ti, 3070 Ti, 3070, 3060. It's the FP/BF16 FMA to FP32 path, and most 3000 series run the same vulnerable GA102 booter.

This is my work, though the exploits aren't mine at all. I just wrapped it up so you don't patch the nvidia driver every release, inspired by the CMP90/50 UEFI unlocks.

Super well done, I wish all mods took your approach. Patched drivers are a nuisance when you have more than one kind of card. Now do the 4090.

Thanks, 27b is a favourite model of mine and my dev got 133 tok/s last night, about 3.2x. Changes the economics.

Semi related, there's a patch giving around 20% more card-to-card speed on consumer GPUs. The hardware supports it but nvidia disabled it when the driver detects cheaper cards.

The next version will surely ship signed firmware that locks these configs down. Though the encrypted debug images could be decrypted with a public test key and are close enough to production to build exploits from.

If nvidia is going to be short on cards for years there's no reason to keep this locked. Making it official won't change sales at this point.

Real gains are far less than advertised. TFLOPS tests show 25% at a 200W cap, and since mostly prefill benefits, the real world gain lands around 7-8% on Qwen 27B Int4 with 4x3090 and full batching on VLLM.

Worked for me, about 6% faster training on nanochat with my 3060. Took extra steps since the repo targets kernel 6.x and I'm on 5.15, Claude Code got me past it. I'm clearly power limited at 170W and have no desire to brick the card.