| *** mhen_ is now known as mhen | 01:36 | |
| *** ralonsoh_ is now known as ralonsoh | 06:20 | |
| jopdorp_ | Hi | 07:59 |
|---|---|---|
| jopdorp_ | I was wondering if anyone has experience with H100 HGX systems , or maybe DGX and GPU passthrough to KVM VMs in nova | 07:59 |
| jopdorp_ | We're seeing an issue where the nvidia-smi command hangs | 07:59 |
| jopdorp_ | we do install fabricmanager, and I've tried some different configs of it. | 07:59 |
| jopdorp_ | We're trying to pass one gpu per vm, on a host that has 8xH100 SXM gpus | 07:59 |
| jopdorp_ | nv_open_q takes 100% of a single cpu core when nvidia-smi is invoked and hangs | 07:59 |
| Mc- | gpu passthrough do work in nova | 10:09 |
| jopdorp_ | gpu passthrough with pcie GPUs works for our other models | 13:15 |
| jopdorp_ | this problem comes with the nvswitch HGX type machines that use SXM instead of normal pcie | 13:15 |
| Mc- | ah | 13:18 |
| Mc- | if it does not use pci, not sure how the passthrough will work, sorry :/ | 13:22 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!