[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-API] GPU passthrough: multiple vGPUs on the same VM

Hi Rob,
Âthanks for the clarification. I googled a little bit and found the "other-config:pci" option but I guess I probably need GPU passthrough, however we'll try that.

Actually we have single NVIDIA K2 card which carries 2 GPUs so they are correctly seen as a gpu-group of 2. We would like to attach the whole card (i.e. 2 PCI devices) to a single VM. The VM OS is Linux (CentOS 6 64bit).

The reason for having both GPUs on the same VM is a little complicated to explain.

We make a software which receives graphic rendering requests from the network and passes them to the available GPUs. We have a simple GPU load-balancer to take advantage of multiple GPUs on the same machine.
When switching from physical machines to virtual machines we can just attach a single GPU to a VM and so we need multiple VMs to be able to use multiple GPUs. This increases complexity and we cannot leverage our load-balancing logic across multiple machines at the moment.

Do you know if there are some serious architectural/technical reasons making this support very difficult to achieve or this is a limitation which could be removed with a limited effort?


On Thu, Jul 11, 2013 at 6:08 PM, Rob Hoes <Rob.Hoes@xxxxxxxxxx> wrote:

Hi Andrea,


You are right: the feature is currently restricted to a single GPU per VM. You could try manually passing through the PCI devices of multiple GPUs using âother-config:pciâ.


I donât think there are any plans to change this at the moment. Could you explain why passing through multiple GPUs to a single VM would be useful for you?




From: xen-api-bounces@xxxxxxxxxxxxx [mailto:xen-api-bounces@xxxxxxxxxxxxx] On Behalf Of Andrea Curtoni
Sent: 11 July 2013 4:36 PM
To: xen-api@xxxxxxxxxxxxx
Subject: [Xen-API] GPU passthrough: multiple vGPUs on the same VM



Âit looks like it is currently not possible to attach multiple vGPUs to the same VM. I tried with XenServer 6.2 and suspect the same happens with XCP 1.6.


I'm successfully using an NVIDIA card in GPU passthrough from a VM but when I try to attach a second vGPU I get the following error:


[root@xenserver ~]# xe vgpu-create vm-uuid=... gpu-group-uuid=...

A device with the name given already exists on the selected VM

device: 0


Trying to specify a different device did not help:


[root@xenserver ~]# xe vgpu-create vm-uuid=... gpu-group-uuid=... device=1

The device name is invalid

device: 1


I think the support for multiple vGPUs on a single VM is not implemented at the moment, I'd like to understand the following:

1) is it possible to workaround this limitation?

2) are there any plans to introduce this support in the future?


Thanks in advance,


Andrea Curtoni
NICE srl
TEL: +39 0141 90.15.16
FAX: +39 0141 01.98.29
Xen-api mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.