⬢github C · 13.2K ★ +10.7K since we first saw it · pushed 1 d ago
Zagitta/gpu-unlock
Your GeForce runs AI at half speed. This fixes it for 20/30/40 series cards. Up to 2× FP16 / BF16 / TF32 tensor throughput
A Linux kernel module that unlocks the FP16/BF16 instruction issue rate on Nvidia RTX 3000-series (Ampere) consumer GPUs, letting them run at the throughput of their professional counterparts. It applies the unlock before the NVIDIA driver binds, so it survives driver updates without patching, and reports show ~25–80% faster GEMM and LLM prefill workloads.
Why now: Trending on Hacker News and among most-starred new repos after benchmarks showed major GEMM gains (e.g., RTX 3090 Ti BF16 jumping from 87 to 157 TFLOPS), essentially free compute for GPU owners.
Who it is for: Linux users with RTX 30-series cards who run matrix-heavy workloads like ML training, inference, or scientific computing.
Stars over our 508 snapshots: 2.5K to 13.2K, since 5 d ago.
Where people talked about it
- ⬢github new repos, most starred 5 min ago
- Yhn GPU-unlock: FP/BF16 instruction issue rate on Nvidia 3000's cards 36 min ago
- ⬢github new repos, most starred 4 h ago
- ⬢github new repos, most starred 1 d ago
- ⬢github new repos, most starred 2 d ago
- ⬢github new repos, most starred 3 d ago
- ⬢github new repos, most starred 4 d ago
- ⬢github new repos, most starred 5 d ago
API: https://socialmediatrends-api.osmike.com/v1/repos/Zagitta/gpu-unlock