gpu-unlock

Ada Lovelace · RTX 40 · AD103

GPU Unlock for RTX 4070 (AD103)

Experimental10de:2709

The RTX 4070 (AD103) is supported experimentally. Results come from one or a few cards, so check your own log carefully.

Measured BF16 (dense matmul, FP32 accumulate)

No per-card BF16 number is published yet. The tensor rate itself doubles on supported cards; real gains depend on power headroom.

  • Experimental Ada IDs need their own opt-in flag and an exact PCI address; targets=all never selects them in gpu-unlock-ada.
  • This model has Windows UEFI success reports. Check the exact PCI ID, board and boot-route requirements in the Windows guide.

How to unlock the RTX 4070 (AD103)

01Clone and install the project on Linux.

git clone https://github.com/JustAResearcher/gpu-unlock-ada
cd gpu-unlock-ada
make install

02Add this to your kernel command line. Replace the address with yours from lspci -D.

gpu_unlock.targets=0000:01:00.0 gpu_unlock.ad103_experimental=1

03Power off, start again, and check the log.

journalctl -k -b | grep gpu_unlock
Full install guide →

FAQ

Can the RTX 4070 (AD103) be unlocked on Windows?

Windows UEFI success has been reported for the RTX 4070 (AD103). Check the exact PCI ID, board, display route and the selected port’s requirements before trying. See the Windows page.

Which PCI device IDs does the RTX 4070 (AD103) use?

10de:2709. Check yours with lspci -nn | grep -i nvidia.

Other Ada Lovelace · RTX 40 cards

All supported GPUs →
RTX 4090ExperimentalRTX 4090 DExperimentalRTX 4080 SUPERExperimentalRTX 4080ExperimentalRTX 4070 Ti SUPERExperimentalRTX 4070 Ti SUPER (AD102)ExperimentalRTX 4070 TiExperimentalRTX 4070 SUPERExperimentalRTX 4070ExperimentalRTX 4060 TiExperimentalRTX 4060ExperimentalRTX 4090 Laptop GPUExperimentalRTX 4060 Laptop GPUExperimentalRTX 4050 Laptop GPUExperimental