Hardware · Updated 24 Oct, 03:28 pm IST

GPU-unlock kernel module doubles FP/BF16 issue rate on some Nvidia 3000 GPUs

Why it matters for readers: It shows a software way to change GPU behavior and gain performance on certain consumer cards.

GPU-unlock is a Linux kernel module that combines CMP unlocker V67/PLM technique with pre-driver timing used by UEFI approaches.1
The module writes speed selectors (SS0 0x88888888 and SS1 0x8 IMLA4) and performs FLR before the unmodified Nvidia driver binds the card.1
The unlock doubles the instruction issuing speed of FP/BF16 FMA into FP32 operations on GA102, GA103, GA104 and GA106 cards according to reports.1
User-reported benchmarks on an RTX 3090 Ti show large TFLOPS increases: BF16 from 87.50 to 157.41 TFLOPS and TF32 from 42.58 to 82.61 TFLOPS in a dense 8192^3 GEMM test.1
The module accepts multiple GeForce PCI IDs and relies on community compatibility reports; results vary by card model and report.1

Get a brief like this every morning

Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.