Support GPU computing in OpenNebula

Community Support
1

Hello,

Is anyone on this forum supporting/integrating GPU computing with ON? If it is so, could you please indicate me how did you make this integration and how are you supporting it?

Also, Is anyone else on this forum interested in getting some facilities in ON to support GPU computing?

We would like to get something similar to this, https://cwiki.apache.org/confluence/display/CLOUDSTACK/GPU+and+vGPU+support+for+CloudStack+Guest+VMs but also for KVM (which needs vfio-pcio and is available from QEMU2.0 onwards).

Thanks in advance,
Esteban

2

Have you found an answer ? I also have a similar question.

3

Hello @ITenthusiast,

Yes, OpenNebula 4.14 supports to add GPU PCI devices to the VM :slightly_smiling:

Your can find more information about it on the following links:

http://docs.opennebula.org/4.14/release_notes/release_notes/whats_new.html

http://docs.opennebula.org/4.14/administration/virtualization/kvmg.html#kvm-pci-passthrough

Best regards,
Esteban

4

But I am thinking of sharing the GPU by multiple VMs
(e.g. rCUDA) and not GPU passthourgh method(since only one VM can access the GPU), is it possible ?

5

As far as I know, It is only possible to do it through passthrough method.

If you have 4 GPU PCI devices on the host, you could assign each of these GPU devices to different VMs and have 4 VMs running with one GPU device each one but I don’t think it is possible to share the same GPU by multiple VMs.

Thanks,
Esteban

6

Thank you Esteban