Multi gpu peer to peer not available in vm
-
Hi everyone,
I am trying to get P2P communication working between two Nvidia L40s PCI GPU cards.
I did this in qemu/kvm by adding the "x-nv-gpudirect-clique=1" flag to each -device entry in the VM spec. Performance wasn't great (~13GB/s in the cuda p2pBandwidthLatencyTest), but the p2p function seems to work.
I can't achieve similar behavior in the latest xcp-ng 8.2 or 8.3 versions.Is it possible in xcp-ng? Has anyone struggled with such a problem?
Thank you all in advance for your time and input!
-
Hi,
I'm not entirely sure to get it. You want P2P working with 2x GPUs via PCI passthrough AND inside a VM?
-
exactly
-
Hmm I'm not sure it's even possible due to the nature of isolation provided by Xen
Let me ask @Team-Hypervisor-Kernel
Hello! It looks like you're interested in this conversation, but you don't have an account yet.
Getting fed up of having to scroll through the same posts each visit? When you register for an account, you'll always come back to exactly where you were before, and choose to be notified of new replies (either via email, or push notification). You'll also be able to save bookmarks and upvote posts to show your appreciation to other community members.
With your input, this post could be even better 💗
Register Login