Closed cnr0724 closed 3 years ago
I figured it out that my computer was the problem. Sorry.
So this kind of things happen. I thought it was my computer, but I don't have much clue that it sure is. Why does this happen?
We will not investigate issues we can't reproduce. If it's specific to an environment, the environment needs to be provided (e.g. as a docker container).
@cnr0724 Do you not get an incompatibility error running the GTX 3090 (arch 8.6)? All the arch flags in your environment indicate it should only support up to 8.0.
We're trying to get a GTX 3090 working with detectron ourselves, and curious if those versions (detectron 0.4 + pytorch 1.7.1+cu110) actually supports the 3090.
@mdebeer Yes, it did not. Did you change the TORCH_CUDA_ARCH_LIST flag? export TORCH_CUDA_ARCH_LIST=8.0
might help.
I found out it was the problem on the CPU, so it doesn't have a relation with the code.
Instructions To Reproduce the 🐛 Bug: