-
Notifications
You must be signed in to change notification settings - Fork 1.2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
OSError: Unknown compute capability. Ensure PyTorch with CUDA support is installed. #1177
Comments
I ran into the same exact error today with conda installation of PyTorch with CUDA 11.7 (11.3 worked fine for me though). Seems like a problem with tinycuda rather than nerfstudio. |
I'm also using CUDA 11.3, did the tinycuda installation go entirely smoothly? I had a bunch of issues with it before I got it to work. |
@THE-COB I have both had tinycudann installation go smoothly and not work at all. I think @brawlawnzo is right, it's an issue with tinycuda. I just tried to install as per nerfstudio installation instructions on my work computer. Visual Studio 2019, CUDA 11.3, Python 3.8. Completely clean install. Got this error install tinycudann:
followed by thousands of lines of errors, and then:
|
@grshennan One of the things in that thread that I think got me passed that (and then to the next error lol) was conda installing ninja. |
@THE-COB I have VS 2019 installed but same idea, yes. I think it's the same as this issue: NVlabs/tiny-cuda-nn#147 Tried installing ninja and got a different error. Now it's
as per the tinycudann issue. Edit: I tried |
What is your current error? @grshennan |
|
@grshennan that is tough, I think I might have run into something similar but I don't remember. If you want (this may or may not be worth it) try building tinycudann in the way suggested in this comment? |
Oh wow, it works. Thanks, @THE-COB. I changed CUDA to 11.7 and Visual Studio to 2022. So still using
(cmake commands run in VS developer command prompt) Then just |
I'm fairly new to this topic. Would you explain, where you have placed the tiny-cuda-nn files and in which directory you have started the cmake process? I am receiving this error:
|
It sounds like tiny-cuda-nn wasn't installed correctly. I would try running |
Thank you very much! The reinstallation actually worked, although the pip installation method of Tiny CUDA Neural Networks has never worked before for me. To document what I have done:
I am not sure, why the installation via pip worked without errors during the last step, but nerfstudio is now fully functional. |
Installed nerfstudio with CUDA 11.3 as per setup page. Everything works fine. Followed the tutorial all the way to
ns-train nerfacto --data data/nerfstudio/poster
where I get this error:Removed CUDA 11.3, installed 11.7 and followed the install instructions on the Pytorch page, i.e.
conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
. Everything works as per thepip
install instructions.tinycudann
even seemed to build a lot faster. Ended up with exactly the same error:OSError: Unknown compute capability. Ensure PyTorch with CUDA support is installed.
CUDA variables are set as (not sure if this is relevant):
I understand devs are busy and I don't want to bother anyone but I'm trying really hard to get this working and just keep getting pipped at the post. Let me know if there's a better forum.
The text was updated successfully, but these errors were encountered: