Pytorch nightly install unsatisfiable Error

Hello,

When trying to install pytorch nightly (specifically pytorch-cuda) I get the following error:

conda install -y -c pytorch-nightly pytorch-cuda
Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Solving environment: failed with repodata from current_repodata.json, will retry with next repodata source.
Collecting package metadata (repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Solving environment: \ 
Found conflicts! Looking for incompatible packages.
This can take several minutes.  Press CTRL-C to abort.
failed                                                                                                                     

UnsatisfiableError: The following specifications were found to be incompatible with your system:

  - feature:/linux-64::__glibc==2.28=0
  - python=3.9 -> libgcc-ng[version='>=11.2.0'] -> __glibc[version='>=2.17']

Your installed version is: 2.28

Some additional Information:

which ldd
/usr/bin/ldd
ldd --version
ldd (GNU libc) 2.28
Copyright (C) 2018 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
Written by Roland McGrath and Ulrich Drepper.

I don’t understand the error message nor do I know if this is the right place to ask. (Perhaps a GitHub issue is the place to ask?)

slightly offtopic: I am trying to install pytorch with cuda 11.8 for the h100 GPU with CUDA 12. What would happen if I would use a 11.7 pytorch/cuda install? Would it only result in a performance loss or is not usable at all?

The current binaries do not support Hopper GPUs yet, as I’m still working on adding the sm_90 support into the nightly CUDA 11.8 binaries.

Your install command seems to be missing the -c nvidia channel as indicated in the install instructions.

EDIT: to run on your H100 you could either build PyTorch from source or use any NGC container with CUDA 11.8+ from here.