The docker image does not work with modern CUDA builds (which prevents new cards from working altogether)

[SacredSkull]

I initially tried to get this running with CUDA on my rack server which has a Quadro P2200 - but it would always fail with a device error (this was with the latest CUDA driver).

I'm now trying this on Windows/docker and It seems that the RTX 5000 series might be too new for the version of torch this project is set up to use my 5080.

[UltraSinger] ********************************
[UltraSinger] UltraSinger Version: 0.0.13-dev8
[UltraSinger] ********************************
[UltraSinger] Checking GPU support.
/app/UltraSinger/.venv/lib/python3.10/site-packages/torch/cuda/__init__.py:209: UserWarning:
NVIDIA GeForce RTX 5080 with CUDA capability sm_120 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_70 sm_75 sm_80 sm_86 sm_90.
If you want to use the NVIDIA GeForce RTX 5080 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

Are there any ways forward with this? Or do I have to live with CPU evaluation :(

[SacredSkull]

Has anyone had any success with updating the dependencies to use pytorch 2.71 or later?

I read that the project locks itself to 3.10 to ensure that windows compatibility is maintained - but I would counter that docker should provide all of the CUDA support necessary.

[rakuri255]

There is a pytorch problem with these rtx5000 cards. Someone mentioned that you will need to use newer pytorch version.

#266 (comment)