We had some issues on one of the GPU clouds with H100 instances related to PyTorch not supporting the required CUDA version. Then that was fixed and ran into a different issue.
I thought it might've been isolated but then saw this HN comment: https://news.ycombinator.com/item?id=36573601
Have you run into more issues with H100s than other GPUs? Should go away with time as software gets updated for them, but I'm curious how widespread issues are at the moment.