GPU Unlock Compatibility Checker

Check your NVIDIA RTX GPU, compare source-backed benchmarks, and find the right installation steps.

Unofficial community guide. Based on the open-source GPU Unlock project. No installation required to check compatibility.

Browse supported GPUs

Check your GPU

Support starts with the model

Browse all GPU entries

Linux upstream overview. Last checked October 29, 2026.
Generation	Architecture	Status	What to check
RTX 30	Ampere	Supported	Upstream labels this generation stable; check the exact PCI ID.
RTX 40	Ada Lovelace	Experimental	Model and environment need individual review.
RTX 20 / Titan RTX	Turing	Experimental	Review current reports and device IDs.
RTX 50	Blackwell	Not Supported	No Linux installation recommendation.

Source: support data and sources. A model listing, a recognized PCI ID and a successful hardware test are different kinds of evidence.

Tensor math, measured upstream

Explore benchmark evidence

Upstream BF16 measurements with FP32 accumulation, in TFLOPS. The RTX 4090 entry reports peak MMA instruction rate; it is not an application speedup.

Stock power limits except the RTX 3090 Ti at 450 W. Gains are transcribed as reported, not recalculated from rounded values.
GPU	Before	After	Reported gain
RTX 3060 12 GB	27	51	+87%
RTX 3060 Ti	37	68	+82%
RTX 3070	40	72	+81%
RTX 3080 10 GB	65	113	+73%
RTX 3090	75	125	+65%
RTX 3090 Ti	88	157	+80%
RTX 4070 Ti	81	150	+84%
RTX 4080	104	193	+87%
RTX 4090	165	329	+100%

The RTX 4090 figure is peak MMA rate; sustained large GEMMs gain less at stock power limits. These are not this website’s measurements.

Application speed is a separate question. The README reports +25–55% for selected LLM prefill workloads and 0–15% for quantized GGUF in llama.cpp / Ollama, mostly at long context. It reports no change for memory-bound token generation, gaming or FP32 CUDA. Those ranges are upstream summaries, not predictions for your machine.

Read the recovery path first

Installation changes kernel and boot behavior. Keep a known-good configuration before you begin.

Open the full installation guide

Confirm your deviceCheck the exact model, PCI ID, Linux environment and upstream status.
Prepare a recovery bootBack up important data and keep a working boot entry.
Follow the correct boot setupChoose your distribution and bootloader. Do not combine unrelated commands.
Verify after rebootRead the kernel log and compare a relevant workload on your own system.

How GPU Unlock works

The upstream module runs before the NVIDIA driver binds the GPU. It changes tensor speed selectors and resets the device before normal driver initialization. The README describes this as a kernel module, rather than a patched NVIDIA driver or a VBIOS flash.

That timing matters. Loading the module after a driver has already bound the card does not unlock that card. Laptop power-off behavior and resume paths also have specific limits.

Read boot configuration details

Questions before you install

What is GPU Unlock?

It is an upstream Linux kernel module that changes tensor compute rate limits before the NVIDIA driver binds a supported card. This website is an independent reference tool and does not develop or distribute the module.

Is GPU Unlock safe?

Kernel and boot changes can cause instability, boot failures or data loss. Upstream reports do not establish safety for every machine. Keep backups and a working boot configuration before deciding whether to proceed.

Which NVIDIA GPUs are supported?

The current upstream overview lists Ampere / RTX 30 as stable and Turing / RTX 20 and Ada / RTX 40 as experimental. Exact PCI IDs and laptop variants matter. Use the model table and check recipes.c.

Does GPU Unlock work on Windows?

The Linux maintainer links to an independent Windows port. Its hardware and machine requirements are separate. Read the Windows profile and recovery overview. This site does not verify your Windows machine or provide executable installers.

Does GPU Unlock support RTX 4090?

The desktop RTX 4090 is listed as experimental. The published peak MMA result is not a guarantee of sustained application performance. Check the exact device and current upstream reports.

Does GPU Unlock support RTX 5090?

No. The upstream Linux project excludes Blackwell / RTX 50. Do not apply installation instructions on the assumption that all RTX cards are supported.

Does GPU Unlock remove the GPU power limit or unlock more VRAM?

No. GPU Unlock targets selected NVIDIA tensor compute rate restrictions. It does not raise the board power limit, add VRAM capacity or remove laptop GPU power limits; those are separate hardware or firmware topics. See the performance and power limits.

Will GPU Unlock improve gaming performance?

The upstream README reports no change for gaming and FP32 CUDA. The described change concerns selected tensor math workloads.

Does GPU Unlock improve Ollama or LLM inference?

The upstream README reports 0–15% for quantized GGUF in llama.cpp / Ollama, mostly at long context, and no change for memory-bound token generation. Prefill can behave differently. Results depend on the workload.

How do I verify GPU Unlock is working?

After a controlled reboot, review this boot’s kernel log for gpu_unlock messages and inspect your actual PCI device. A loaded module alone does not prove that a card was unlocked. Use the verification steps and troubleshooting guide.

How do I disable GPU Unlock?

Remove the gpu_unlock.targets= option from the active boot configuration and regenerate entries where required. For a one-off recovery boot, remove it in the bootloader edit screen. Read the rollback section before installing.

Answers are based on the upstream README, checked October 29, 2026. See sources and verification for the original project.

Sources and verification

Compatibility, benchmarks and technical explanations summarize the original project’s README and device recipes, checked October 29, 2026. Our support and installation pages explain how to use this snapshot. Use the original repository to verify current code, reports and notices. This independent website has no official partnership with NVIDIA, Zagitta or the project’s developers.

View Zagitta/gpu-unlock on GitHub