Please add cu128 support for NVIDIA 50xx-series GPUs (Blackwell)
The recent NVIDIA 50xx series GPUs (Blackwell) like my 5060-Ti do not work with the old CUDA versions for whatever reason.
Which broke virtually all AI tools (ComfyUI, SD.Next/Auto1111, your vcclient etc) for us 50-series users when they came out in early 2025.
When trying to start the app we instantly get a runtime error:
RuntimeError: CUDA error: no kernel image is available for execution on the device
NVIDIA GeForce RTX 5060 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation.
While pretty much all of them have fixed their requirements to PyTorch >= 2.7.0 built against CUDA >= 12.8 by now (which is the minumum for 50-series GPUs) - or they could easily be fixed by running something like...
[...]\python_embeded\python.exe -m pip install -U torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
... neither the old MMVCServer versions nor the newest 2.1.4-alpha seem to have an embedded python environment where we run that. So basically we're all stuck with only CPU mode for the past 6 months or so.
Any chance we could get this fixed from 2.1.4 on, please? ♥
Im not sure if you tried Using the 2.0.78 but it seems to work fine, and doesnt give me the Sm_120 Error when using my RTX 5090 even though every other version does,
But I also asked Chatgpt if there was anything i could use in the 2.0.78 files and replace them in the 2.1.3 and 2.1.4a versions and it said to try coping the torch folder from 2.0.78 vcclient_win_cuda_2.0.78-beta\dist\main_internal\torch and replacing that folder in the other two newer versions, So I made copies and deleted the torch folder in both the newest client versions and pasted in the one from the 2.0.78,
Crazy enough it actually worked 0.o, Havent tried doing any Ai training with it though but at least i can somewhat use the ones I have already trained pre RTX 5090 XD