Gpu is not showing in the host

[TheKunalSen]

I have created a basic zone in cloudstack and added few host machines and all of them are having a rtx 3060 graphic card in them but the gpu section is always showing empty but the rest of the things are working like launching vms and all

Can you tell me how can I enable the gpu for the hosts !

[vishesh92]

@TheKunalSen This is probably because nvidia rtx 3060 is a consumer grade GPU and the type of device is probably something other than 3D controller or Processing accelerators.

To make it work,

1. Execute the below command to check the pci class of the nvidia GPU card

   sudo lspci -nnm | grep -i nvidia
   01:00.0 "3D controller [0302]" "NVIDIA Corporation [10de]" "GA107M [GeForce RTX 3050 Ti Mobile] [25a0]" -ra1 -p00 "Dell [1028]" "Device [0b19]"

   In this case, "3D controller" is the PCI class.
2. Edit the gpudiscovery.sh script on the hosts and add the PCI class to the regex if it's missing.
   

   cloudstack/scripts/vm/hypervisor/kvm/gpudiscovery.sh

   Line 752 in 34b8870

   if [[ ! "$PCI_CLASS" =~ (3D\ controller|Processing\ accelerators) ]]; then

3. Restart the cloudstack-agent service.

Also, ensure that you have pciutils & xmlstarlet installed on the hosts as well.

If all this doesn't work, please share the output of the gpudiscovery.sh script on the host.

[tatay188]

thank you @prashanthr2 However, @arpanbht The understanding is we all have knowledge of unix or linux:

You can use the find command from linux/Unix, for example:

sudo find / -name "gpudiscovery.sh" -print

GPU One step at the time:
Then is not clear to me what is your intention, If you want to run a GPU passthrough to make the GPU Cards available to the VMs, then read both links I am sharing with you, also get a good book on linux.

STEP ONE FOR GPU PASSTHROUGH
Link one: OpenStack GPU Passthrough for HPC
follow the instructions here for OpenStack; are the same for cloudstack, you stop before the NOVA settings, is quite straight forward, and depending on your GPU card it may differ which existing one you will have to blacklist.

STEP TWO You can go to step 3 if you understand already what is step2 about.
Link two: https://lab.piszki.pl/cloudstack-kvm-and-running-vm-with-vgpu/
The second one is more specific to Cloudstack, in case the there are problems with the gpudiscovery.sh, you can use the manual setup shown on this link, and will give you a better grasp on understanding of the whole functionality.

STEP THREE
After you have the GPU with the vfio-pci driver then you can start working on the gpudiscovery.sh
|

[TheKunalSen]

the driver i have changed to vfio but still in the gpu section it is empty

root@host-62-gpu:~# lspci -nnk -d 10de:
01:00.0 VGA compatible controller [0300]: NVIDIA Corporation GA107 [GeForce RTX 3050 8GB] [10de:2582] (rev a1)
Subsystem: ZOTAC International (MCO) Ltd. GA107 [GeForce RTX 3050 8GB] [19da:2641]
Kernel driver in use: vfio-pci
Kernel modules: nvidiafb, nouveau
01:00.1 Audio device [0403]: NVIDIA Corporation Device [10de:2291] (rev a1)
Subsystem: ZOTAC International (MCO) Ltd. Device [19da:2641]
Kernel driver in use: vfio-pci
Kernel modules: snd_hda_intel

[tatay188]

@TheKunalSen You may need :
Ensure that everything is running on release 4.22.0.
Restart the Agent on you KVM host.
Remove and add the KVM host to your POD/Zones
Modify the gpudiscovery.sh to include the GeForce RTX 3050, I know will work with RTX 3070, you will have to dig on it.

[vishesh92]

@TheKunalSen did you modify the gpudiscovery.sh and restart the cloudstack agent?

[TheKunalSen]

Yes I did modified the gpudiscovery.sh file and now the host is gpu enabled!
The problem was with the different pci class!
Thanks for the help!