Tim Gu

Hi @trixirt, thank you for your effort trying to improve the out-of-box user experience of ROCm on Fedora/CentOS, we really do appreciate it! However, after some discussions, we felt that...

Hi @oldxie, this issue will be closed for now due to inactivity. Please feel free to reopen for more follow-ups. Thanks!

@nix-wolf Hi there! I took a look at your initial steps. Would you mind providing a bit more details regarding how you set up your system? For example, how did...

> Hey, I do believe if you read through the subsequent comments I made relative to trying to figure it out this is one thing I did figure out thay...

@nix-wolf Thanks for the additional information! We were able to successfully reproduce your issue on RHEL 9.3 + ROCM 6.2 with the official docker pytorch image for CentOS+Rocm6.1. We are...

Hi @nix-wolf, a brief update here: It seemed like the error was fairly transient. While I was able to get it in the first run, I did not get it...

@illwieckz Thanks for reporting the issue! A ticket has been opened internally and we will be working actively to resolve this. Thanks!

Hi @illwieckz. Thank you again for pointing out the issue. We were able to reproduce it on our side as well. For now, we recommend following the official [ROCm upgrade...

Hi @void-Man, can you try running your workload with lowvram mode? Automatically using shared memory when on device memory is insufficient should by no means be a responsibility of ROCm....

@void-Man A bit of additional context, the reason why on windows shared memory is automatically used seems to be a feature of the windows graphics driver as per this post...