gpu-unlock Linux module doubles FP/BF16 instruction rate on Nvidia 3000 series GPUs

·2026.10.25 06:38

Key point

The open-source tool unlocks hardware limits on GA102-GA106 chips, boosting BF16 GEMM throughput by up to 81% and LLM prefill speeds by 14-36%.

Linux kernel module unlocks FP/BF16 instruction issue rate limits on Nvidia GeForce 3000 series cards (GA102, GA103, GA104, GA106) to match professional equivalents.
Benchmarks show BF16 GEMM throughput increasing from 87.50 TFLOPS to 157.41 TFLOPS on RTX 3090 Ti, with user-reported gains ranging from 25% to 81% depending on power limits and card model.
LLM prefill speeds improve by 14-36%, while memory-bandwidth-limited LLM decode and INT8 workloads remain unchanged.
Implementation uses a pre-driver timing technique to write speed selectors before the standard NVIDIA driver binds, avoiding the need for manual driver patching on updates.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.