vGPU Remapping Issue with Multiple GPUs

B

bcchenbo

New Member
May 15, 2025
4
0
1
Hello everyone,


My server has two GPUs: an A5000 and an A6000. Both have been successfully split into vGPU instances and appear correctly under Datacenter → Resource Mapping. I created two separate mappings for the A5000 and A6000 to make management easier.

However, when I try to add a mapping device for the A6000 to a VM, I encounter an error in VM Start Task viewer:
Code: 
no nvidia sysfs api for '0000:0001:6f:00.4'
no nvidia sysfs api for '0000:0001:6f:00.5'
no nvidia sysfs api for '0000:0001:6f:00.6'
no nvidia sysfs api for '0000:0001:6f:00.7'
no nvidia sysfs api for '0000:0001:6f:01.0'
no nvidia sysfs api for '0000:0001:6f:01.1'
no nvidia sysfs api for '0000:0001:6f:01.2'
no nvidia sysfs api for '0000:0001:6f:01.3'
no nvidia sysfs api for '0000:0001:6f:01.4'
no nvidia sysfs api for '0000:0001:6f:01.5'
no nvidia sysfs api for '0000:0001:6f:01.6'
no nvidia sysfs api for '0000:0001:6f:01.7'
no nvidia sysfs api for '0000:0001:6f:02.0'
no nvidia sysfs api for '0000:0001:6f:02.1'
no nvidia sysfs api for '0000:0001:6f:02.2'
no nvidia sysfs api for '0000:0001:6f:02.3'
no nvidia sysfs api for '0000:0001:6f:02.4'
no nvidia sysfs api for '0000:0001:6f:02.5'
no nvidia sysfs api for '0000:0001:6f:02.6'
no nvidia sysfs api for '0000:0001:6f:02.7'
no nvidia sysfs api for '0000:0001:6f:03.0'
no nvidia sysfs api for '0000:0001:6f:03.1'
no nvidia sysfs api for '0000:0001:6f:03.2'
no nvidia sysfs api for '0000:0001:6f:03.3'
no nvidia sysfs api for '0000:0001:6f:03.4'
no nvidia sysfs api for '0000:0001:6f:03.5'
no nvidia sysfs api for '0000:0001:6f:03.6'
no nvidia sysfs api for '0000:0001:6f:03.7'
no nvidia sysfs api for '0000:0001:6f:04.0'
no nvidia sysfs api for '0000:0001:6f:04.1'
no nvidia sysfs api for '0000:0001:6f:04.2'
no nvidia sysfs api for '0000:0001:6f:04.3'
TASK ERROR: could not find a free device for 'hostpci0'

0001:6f:00 is A6000 PCI slot
I'd like to ask if this is due to a misconfiguration on my part, or if it's a Proxmox-related issue?


Environment:
  • Proxmox version: 8.4.1
  • Kernel version: 6.8.12-10
  • NVIDIA driver: 570.133.10

Thanks in advance for your help!
 
thanks for the report, the issue is that the device is on the second pci domain '0001' which is a bit uncommon
i can see our code does not handle that correctly, and tries to prefix it with the first pci domain '0000' which leads obviously to a wrong pci id

would you mind reporting this on our bugtracker? https://bugzilla.proxmox.com
if you do, please reference this thread and post
* the vm config
* the output of `lspci`

thanks !
 
dcsapak said:
thanks for the report, the issue is that the device is on the second pci domain '0001' which is a bit uncommon
i can see our code does not handle that correctly, and tries to prefix it with the first pci domain '0000' which leads obviously to a wrong pci id

would you mind reporting this on our bugtracker? https://bugzilla.proxmox.com
if you do, please reference this thread and post
* the vm config
* the output of `lspci`

thanks !
Click to expand...
Of course! I've already posted this under Bug 6400 and hope it will be fixed soon!
 
well the most correct way would be to checkout the two repositories with git, then use `git am <path-to-patch-file>` to apply them and use `make deb` to build the packages.

if you want to go that route, i'd recommend reading the developer docs https://pve.proxmox.com/wiki/Developer_Documentation

you also could try to manually apply the changes from the patches to the relevant files, but if you do only do so on a test system and not on your production system, I'm not responsible if something is not working right. if you want to revert the changes you can always do

Code: 
apt install --reinstall <package-name>

alternatively you wait until the fix is packaged
 
Sorry, but I still don’t quite understand how to apply the patch to my running system.
I’m just a Proxmox user and not familiar with how to apply patches to a live system.
I think I need a safe way to update the environment, or perhaps you could let me know where I can find or update to the correct released version!
 
it's not applied yet, so first a colleage of mine has to review the patches, then they have to be applied to our git repository. then when the package is bumped the next time will it started to be rolled out through our repositories
i'll see that i can accelerate it a bit, but i can make no promises when the patch reaches the repositories
 
You must log in or register to reply here.
Top Bottom