GPU Unlock supported GPUs
GPU Unlock works on NVIDIA GeForce RTX 3000 cards built on the GA102, GA103, GA104 and GA106 chips. This is the list of cards owners have reported on so far, with the gains they measured.
RTX 3000 (Ampere) cards
| Chip | Card | PCI ID | Result | BF16 GEMM gain |
|---|---|---|---|---|
| GA102 | RTX 3090 Ti | 10de:2203 | Unlocked | +68–80% |
| GA102 | RTX 3090 | 10de:2204 | Unlocked | +25–67% |
| GA102 | RTX 3080 Ti 20 GB (OEM, GA102-250) | 10de:2205 | Unlock completes, BF16 stays at half rate | — |
| GA102 | RTX 3080 10 GB | 10de:2206 | Unlocked | +70–73% |
| GA102 | RTX 3080 20 GB (memory upgrade) | 10de:2206 | Unlocked | +56–59% |
| GA104 | RTX 3070 | 10de:2484 | Unlocked | +73–81% |
| GA104 | RTX 3070 Ti | 10de:2482 | Not tested with this module (worked from a UEFI port on Windows) | — |
| GA106 | RTX 3060 12 GB | 10de:2504 | Unlocked | about +80% |
| GA103 | any GA103 card | 10de:2414 | No reports yet | — |
From the reports table in the official README, checked October 25, 2026. The module accepts every GeForce ID in its source code; the cards above are the ones someone has reported. Gains come from each owner’s own benchmark, so they are not directly comparable, and power limit matters a lot.
Tested a card that is not listed, or got a different result? Add a compatibility report on GitHub, whether it worked or not.
RTX 4000 (Ada) cards
These need the separate GPU Unlock Ada fork, not the original module.
| Chip | Card | PCI ID | Status | Boot option |
|---|---|---|---|---|
| AD102 | RTX 4090 | 10de:2684 | Supported; about 2× FP16/BF16 tensor issue rate measured | gpu_unlock.targets=<address> |
| AD102 | RTX 4090 D | 10de:2685 | Experimental | add gpu_unlock.rtx4090d_experimental=1 |
| AD103 | RTX 4070 Ti SUPER | 10de:2705 | Experimental, validated on one card | add gpu_unlock.ad103_experimental=1 |
| AD106 | RTX 4060 Ti 16GB | 10de:2805 | Experimental, unlocked on one card, no throughput run yet | add gpu_unlock.ad106_experimental=1 |
Other Ada cards are not enabled. Details are on the RTX 4090 and 40 series page.
What gets faster
| Workload | Effect reported |
|---|---|
| FP16 / BF16 / TF32 matrix math with FP32 accumulation | Up to about 2× issue rate |
| LLM prefill (prompt processing) | +14–36% in four reports |
| Diffusion models, FlashAttention | Faster |
| LLM decode (token generation) | About the same, it is memory-bound |
| INT8 | About the same |
| Games | No gain reported |
How to find your card’s PCI ID
lspci -Dnn | grep -i nvidiaThe four-digit code after 10de: is the device ID in the tables above. The first field, such as 0000:21:00.0, is the address you can give to gpu_unlock.targets=.
Supported GPU questions
Does GPU Unlock work on the RTX 3090?
Yes. The RTX 3090 has the most reports, with BF16 GEMM gains of about 25% to 67% depending on the benchmark and power limit.
Does GPU Unlock work on the RTX 3080?
The RTX 3080 10 GB and the memory-upgraded 3080 20 GB are reported unlocked, with +70–73% and +56–59% BF16 GEMM gains. The OEM RTX 3080 Ti 20 GB (GA102-250) completes the unlock, but BF16 stays at half rate.
Does it work on the RTX 3060?
The RTX 3060 12 GB is reported unlocked with about +80% BF16 GEMM throughput.
Does it work on laptop GPUs or RTX A-series cards?
They are not in the reported list. The module only acts on the GeForce device IDs listed in its source, and the Ada fork says professional RTX A-series and data-centre parts are not accepted.
What about the RTX 4090?
Use the GPU Unlock Ada fork. See the RTX 4090 page.
Ready? Go to the install guide.