gpu-unlock kmod lifts RTX 30-series BF16 to 157 TFLOPS, matching pro cards
StasBekman · x · 2026-10-27
Stas Bekman highlights gpu-unlock (9.1k GitHub stars), a Linux kernel module that unlocks FP/BF16 instruction issue rate on RTX 30-series cards to their professional equivalents. It merges the CMP unlocker's V67/PLM technique with pre-driver UEFI-style timing, writing speed selectors (SS0 0x88888888 for full BF16 rate) and performing FLR before the stock NVIDIA driver binds — no manual patching per driver release. Reported results: FP16 87.46→118.39 TFLOPS, BF16 87.50→157.41 TFLOPS (+80%), TF32 42.58→82.61 TFLOPS. Bekman notes Nvidia selling the same silicon cheaply yet profitably implies huge datacenter GPU margins.
Related event: gpu-unlock Unlocks Hidden Tensor Core Performance on RTX 20/30/40 Cards(7 posts)→
More from Infra
- RTX 5080 price jumps from $1658 to $2318 in ten days on Amazon — Frizzy-MacDrizzle · 2026-10-27
- ComfyUI on 12GB VRAM: measured peaks show offloading turns 24s into 11 minutes — Impossible-Mail-4619 · 2026-10-27
- Neon rebuilds its docs search entirely on its own stack: Lakebase Search, Functions, AI Gateway — matei_zaharia · 2026-10-27
- PyTorch details 2.12x faster on-device streaming ASR on Arm SME2 via ExecuTorch — PyTorch · 2026-10-27
- NVIDIA shows speculative decoding speeding up LLM inference on Jetson with Qwen3.5 9B — NVIDIA Developer · 2026-10-27
- Revenue per GPU can rise as token prices fall, argues Gavin Baker — GavinSBaker · 2026-10-27