Going over GRID vGPU limit

Trying to find information on this. What happens if and when the limit of active vGPU nodes is exceeded? Will it still boot? does the driver just not load? will it affect pre-existing sessions using the vGPU on the same host?

For example, we use XenDesktop, with a single K1 per host and K140Q profiles. This gives us up to 16 active nodes per host. What will happen if I boot up number 17? Will it affect the other 16 sessions?


Basically, the VM will error (in the hypervisor logs) and not boot due to unavailable resources.

vGPU profiles are a consumable resource, you need to make sure you have enough available for your VMs. If you have 17 active VMs, you need 17 vGPU profiles (or GPU slices) available. They are not like Memory or CPU where you can just over-commit.

To get around that, use XenApp and assign a K1 in Passthrough. That way you are not bound by the limit of available vGPU profiles. However, if you’re going to do that, I’d recommend a K2 in Passthrough, rather than a K1, as you’ll soon run out of power with a K1.

I’m sure you’d rather stick with XenDesktop, in which case, to run 17+ VMs, you’ll need an additional K1.



Thanks for the response Ben,

So im still a bit curious on the affect that it will have on the active VMs with users already on them once the limit is passed. So you mention that the subsequent XenDesktop VMs on that particular host will just not boot. Can I assume it will not adversely affect the already active VMs?

You can’t pass the limit, it’s a hard limit. VMs that fail to power on won’t affect any currently powered on VMs or active users. All that happens is you’ll have an error in your Hypervisor.

Try it. Just spin up a few VMs into a temporary Catalog and power them all on. Once your GPU resource is fully allocated, that’s your lot! The only people that will notice anything are the admins and your 17th user ;-)

If you’re running XenServer, keeping track of your GPU allocation is easy, as it has class leading GPU management with great visuals and is really easy to understand. If you’re using vSphere, then it’s still really lacking any sort of GPU management compared to Xen and keeping track of what’s allocated is not so easy …



Good to know, thanks Ben!

No worries :-)