mirror of
https://github.com/RVC-Project/Retrieval-based-Voice-Conversion.git
synced 2026-11-05 00:27:21 +00:00
[GH-ISSUE #51] Problem with Nvidia 50 series #38
Labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
RVC-Project/Retrieval-based-Voice-Conversion#38
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @RodriguezNL on GitHub (Jul 13, 2025).
Original GitHub issue: https://github.com/RVC-Project/Retrieval-based-Voice-Conversion/issues/51
Hello everyone!
I have a problem with my graphics card. I changed my graphics card, and now the program is giving me an error.
The program does not work with Nvidia 50 series graphics cards.
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_37 sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90 compute_37.
If you want to use the NVIDIA GeForce RTX 5080 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/
Please tell me how to solve this problem.
Windows 10
@alcoftTAO commented on GitHub (Jul 13, 2025):
Hi, I also have a NVIDIA 50 series card and I fixed this problem by upgrading PyTorch to >=2.7.
You can use this command to upgrade:
pip install --upgrade "torch>=2.7.0" torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128. If it doesn't work, trypython -m pip install --upgrade "torch>=2.7.0" torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128.It fixes that error, but currently RVC does not support this version of PyTorch due to some changes to the library itself. I'm currently working on a fix to this error and I'll create a pull request fixing it, but it seems like @Tps-F is not updating this project anymore, so I'm not sure if the changes will be merged.
@Tps-F commented on GitHub (Jul 13, 2025):
@alcoftTAO Sorry it's been open for so long.
I will test it and if everything is ok, I will merge it.
and I will resume development on this project.
@alcoftTAO commented on GitHub (Jul 13, 2025):
Thanks!
@Tps-F commented on GitHub (Jul 13, 2025):
I merged
3b7e87831a!Hope that problem was solved
@alcoftTAO commented on GitHub (Jul 13, 2025):
Oh, that PR was only a fix for Intel GPUs, not NVIDIA GPUs.
I'm still working on a fix for this issue. I think it's a pytorch function that, in older versions (<2.6) was set to
Falseby default, and now it's set toTrueby default. But I'm not 100% sure yet.@alcoftTAO commented on GitHub (Jul 13, 2025):
I've found the error. In every
torch.load()function, you must setweights_only=False(if not set already).Unfortunately, this error is also in the
fairseqlibrary.This error happens because in torch<2.6 this was set to
Falseby default if it was not specified, but now, in torch>=2.6, this is set toTrueby default.In this image, the error is in
rvc/modules/vc/modules.py.I've updated my local copy of RVC to add the fix, and the error now is in the fairseq library (I'm using Tps-F's fork).
To fix this, the fairseq library must be updated.
When giving the argument
weights_only=Falsein the torch.load() function inrvc/modules/vc/modules.py (line 50)andfairseq/checkpoint_utils.py (line 340), the error fixes and everything seems to work fine.@RodriguezNL commented on GitHub (Jul 13, 2025):
I used the RVC1006Nvidia archive. It's no longer available in the file list. Where can I download the RVC version for win10?