RuntimeError: The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):
RuntimeError: nvrtc: error: invalid value for --gpu-architecture (-arch)
Set the environment variable HYDRA_FULL_ERROR=1 for a complete stack trace.
with cpu it works but when using gpu it does not work
I guess 4090 is not working for cuda10.2…
yes, its compatible only with cuda 12 and above but then these envs do not work and i would like to use nvidia 525 with cuda 12
Right.
I also tested cuda11.7 and cuda11.8 on RTX4090 with nvidia-driver 525, and both of them worked. But I believe cuda12+ could get the best performance.
Hello， I am working on a remote server through ssh and the GPU is RTX4090 as well. I tested cuda11.7 and cuda12.2 with nvidia-driver535, and neither of them worked. I got the same error as you do. I just ran python train.py task=Ant headless=True and I wonder what I could do now. Thank you!
I had the same problem with RTX4080. But I found that commenting out all “@torch.jit.script” from all the code can avoid nvrtc related errors.
Some code still doesn’t work though. (franka_cube_ik_…py)
Did you solve this problem? I also encounter this problem.
Hello All, Even Im facing the same issue
Ubuntu 20, RTX 4070Ti
conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia
This solved it for me
Specs:
RTX 4060 Laptop GPU 8GB RAM