vGPU use with Opennebula 7.0.1

[jj197005]

I would like to know if the way I’m working with the new version 7.0.1 is correct because I thought that this version would manage the vGPU resources and could attach vGPU device to any VM without create the UUID in the vGPU profile as this link specifies:

vGPU experience and profiles

I managed to figure out vGPU setup and profile creation. Here’s the information needed for a successful vGPU launch and profile selection. First, you need to refer to the NVIDIA vGPU support guide by ON: NVIDIA vGPU support. For Enabling Virtual Functions, I recommend creating a systemd service. Create the service: nano /etc/systemd/system/nvidia-sriov.service Add the following: [Unit] Description=Enable NVIDIA SR-IOV After=network.target [Service] Type=oneshot ExecStart=/usr/lib/nvidia/s…

With the command that this documentation gives as example:

echo 811ba00f-b95c-5d0d-a870-99b48162add9 > /sys/class/mdev_bus/0000:01:04:3/mdev_supported_types/nvidia-919/create

Please, tell me if this is the working way, or ONE 7.0.1 already manage this.

---

Versions of the related components and OS (frontend, hypervisors, VMs): I have front-end and nodes 7.0.1 version installed and working.

Steps to reproduce: I have configured all step except MIG-backed vGPU support showed in PCI Passthrough | and PCI Passthrough |

Current results: Using vGPU experience and profiles - #2 by Mayhem I can create and attach vGPU devices with a choose profile to the VMs previous create the adequate template.

Expected results: I thought that version 7.0.1 wiould be able to do this without this previous configuration in the nodes.

[FrancJP (Francisco Picolini)]

Hello @jj197005,

Short answer is yes, vGPU support works with OpenNebula 7.0.1, but it requires the right combination of hypervisor drivers and host configuration.

In OpenNebula this is handled via the VM Templates (GPU attributes) and the KVM/QEMU stack on the host.

Make sure:

• The host has a supported NVIDIA driver and the vGPU profiles enabled.
• You configure the GPU attributes in your VM template according to the documentation examples.
• The hypervisor sees the vGPU devices before trying to schedule VMs.

Please check the updated documentation for a full overview: Nvidia GPU Passtrough.

Let us know if you need more information.

Regards,

[jj197005]

Thank you very much FrancJP,

I undestand. As your explanation says I have already the correct driver and the vGPU profiles enabled. The last point confirms my configuration, it is necessary that the hypervisor sees the vGPUs and for this you have to choose the profile you want and create the mdev devices.

Regards.