Skip to content
GitHub·

🚀GPU Unlock 1.0 Doubles FP/BF16 FMA Speed on RTX Cards

RTX Cards Get a Major Speed Boost Without Manual Patches

TL;DR

GPU Unlock 1.0 module doubles FP/BF16 FMA speed on RTX cards, including RTX 3090 Ti, RTX 3080 Ti, RTX 3080, RTX 3070, RTX 3070 Ti, RTX 3060, and more. No manual patching needed, just install and opt-in.

GPU Unlock 1.0 module doubles FP/BF16 FMA speed on RTX cards, including RTX 3090 Ti, RTX 3080 Ti, RTX 3080, RTX 3070, RTX 3070 Ti, RTX 3060, and more. No manual patching needed, just install and opt-in. This is a big deal for anyone running matrix-heavy workloads like LLM prefill, diffusion, and FlashAttention, as it speeds up dense GEMM throughput by 14-36%. The module runs the unlock, writes speed selectors, and performs FLR before the unmodified NVIDIA driver binds the card. It's installed via `git clone https://github.com/Zagitta/gpu-unlock` and `make install`. Cards below 225 W see a +25% power limit increase.

GPU Unlock 1.0 Doubles FP/BF16 FMA Speed on RTX Cards — GitHub

Key Points

1

GPU Unlock 1.0 module doubles FP/BF16 FMA speed on RTX cards, including RTX 3090 Ti, RTX 3080 Ti, RTX 3080, RTX 3070, RTX 3070 Ti, RTX 3060, and more.

2

No manual patching needed, just install via `git clone https://github.com/Zagitta/gpu-unlock` and `make install`.

3

Module runs unlock, writes speed selectors (SS0 0x88888888 and SS1 0x8 IMLA4), and performs FLR before NVIDIA driver binds the card.

4

Cards below 225 W see a +25% power limit increase, speeding up matrix-heavy workloads like LLM prefill, diffusion, and FlashAttention.

5

LLM decode and INT8 work stay about the same, with bit-identical outputs before and after.

Why It Matters

If you're running matrix-heavy workloads like LLM prefill, diffusion, and FlashAttention, GPU Unlock 1.0 doubles FP/BF16 FMA speed on RTX cards, including RTX 3090 Ti, RTX 3080 Ti, RTX 3080, RTX 3070, RTX 3070 Ti, RTX 3060, and more. No manual patching needed, just install and opt-in. This speeds up dense GEMM throughput by 14-36%, making it a must-have for anyone pushing the limits of GPU performance.

gpu-unlockrtx-cardsperformance-boostmatrix-heavy-workloadsopen-source

Comments

Subscribe to join the conversation...

Be the first to comment

Enjoyed this article?

Get it daily. 7am. Free. Reads in 5 minutes.

Join 3,567 builders reading daily.

Also get