# GPU Unlock

> Independent guide to gpu-unlock, the open-source Linux kernel module (Zagitta/gpu-unlock, October 2026) that lifts the half-rate FP16/BF16/TF32 tensor-core cap (FP32 accumulation) on GeForce RTX cards before the NVIDIA driver loads. Not affiliated with NVIDIA or the project authors; no binaries are distributed.

## Key facts

- Up to 2x FP16/BF16/TF32 tensor issue rate; LLM prefill +25-55%, BF16 fine-tuning +25-50%, image generation +30-55%.
- No change for token generation (memory bound), INT8/FP8/INT4 models, gaming or FP32 CUDA.
- Linux only upstream; opt in with the kernel option gpu_unlock.targets=all; cold boot recommended.
- Stable: RTX 3090 Ti, RTX 3090, RTX 3080 Ti, RTX 3080, RTX 3070 Ti, RTX 3070, RTX 3060 Ti, RTX 3060, RTX 3050, RTX 3080 Laptop GPU, RTX 3070 Ti Laptop GPU, RTX 3070 Laptop GPU.
- Experimental: RTX 4090, RTX 4090 D, RTX 4080 SUPER, RTX 4080, RTX 4070 Ti SUPER, RTX 4070 Ti, RTX 4070 SUPER, RTX 4070, RTX 4060 Ti, RTX 4060, RTX 4090 Laptop GPU, RTX 4060 Laptop GPU, RTX 4050 Laptop GPU, RTX 2080 Ti, RTX 2080 SUPER, RTX 2080, RTX 2070 SUPER, RTX 2070, RTX 2060 SUPER, RTX 2060.
- Not supported: RTX 5090, RTX 5080, RTX 5070 Ti, RTX 5070, RTX 5060 Ti, RTX 5060.

## Pages

- [GPU Unlock](https://gpu-unlock.net/): Compatibility checker, measured gains, safety notes and FAQ.
- [Supported GPUs](https://gpu-unlock.net/gpus): Every covered GeForce RTX card with PCI IDs, status and BF16 gains.
- [Install on Linux](https://gpu-unlock.net/install): Build from source, kernel option per distribution, verify and undo.
- [Windows](https://gpu-unlock.net/windows): Community UEFI ports, reported setups and risks.
- [Benchmarks](https://gpu-unlock.net/benchmarks): Stock vs unlocked tensor throughput and workload speedups.
- [Projects](https://gpu-unlock.net/projects): Open-source gpu-unlock projects and forks compared.
- [vGPU unlock vs GPU unlock](https://gpu-unlock.net/vgpu-unlock): How vgpu_unlock differs from the tensor-core unlock.
- [Guides](https://gpu-unlock.net/blog): Answers about safety, local LLM speed and setup.
- [Privacy policy](https://gpu-unlock.net/privacy-policy): Privacy policy.
- [Terms of service](https://gpu-unlock.net/terms-of-service): Terms of service.

## Source repositories

- [Zagitta/gpu-unlock](https://github.com/Zagitta/gpu-unlock)
- [JustAResearcher/gpu-unlock-ada](https://github.com/JustAResearcher/gpu-unlock-ada)
- [impacifist/Windows-GPU-UNLOCK](https://github.com/impacifist/Windows-GPU-UNLOCK)
- [nlzy/turing-unlock](https://github.com/nlzy/turing-unlock)
- [physicsdolphin/nv-uefi-unlocker](https://github.com/physicsdolphin/nv-uefi-unlocker)
- [phiraml/gpu-unlock-blackwell](https://github.com/phiraml/gpu-unlock-blackwell)
- [miounet11/gpuunlock](https://github.com/miounet11/gpuunlock)
- [PantACRO4life/gpu-unlock-4090-windows](https://github.com/PantACRO4life/gpu-unlock-4090-windows)
- [NeoChen1024/gpu-unlock-ad103](https://github.com/NeoChen1024/gpu-unlock-ad103)
- [Sticzu/nixos-gpu-unlock](https://github.com/Sticzu/nixos-gpu-unlock)
- [DualCoder/vgpu_unlock](https://github.com/DualCoder/vgpu_unlock)
- [mbilker/vgpu_unlock-rs](https://github.com/mbilker/vgpu_unlock-rs)
- [VGPU-Community-Drivers/vGPU-Unlock-patcher](https://github.com/VGPU-Community-Drivers/vGPU-Unlock-patcher)

Full list with every GPU page: https://gpu-unlock.net/llms-full.txt