XenApp 7.15, vsphere 6.5, Grid K1 questions

Hello

I’ve done some XenDesktop installations with Grid K2, but never any XenApp installations and need some advice.

We’re about to do a new installation of XenApp at one of our customers.
They already have 3 vmware hosts with 1 Nvidia Grid K1 installed on each host that are not used for anything :P
So i thought that maybe i can use these K1s in their new XenApp environment.

They have around 600 XenApp users.
Is it possible to split these K1s in any way that would support that many users?
They don’t need the gpus for anything special, just for general performance improvments where they can be used.
If not, how many servers/users can we get out of these 3 cards. If we can’t use them for everyone, maybe we can at least give some users a better performance.

Say it is possible to split 1 card on 4 XenApp servers on one host so that each get 1 GB video ram.
What happens if 1 user consume all resources from the gpu, and another users tries to start something that also use the gpu?

From what i understand the Grid K1 and K2 don’t need any special licensing to use? But the newer Mxx need licensing.

Thanks

Hi there,

The GRID K1 card has 4 physical GPUs. The GPUs can be assigned to XenApp servers using either GPU passthrough or by using vGPU profiles. If you use GPU passthrough, you assign 1 GPU to 1 XenApp server. If you use vGPU profiles, which are preset GPU configurations (frame buffer, display heads, resolution), you can divide each physical GPU into multiple virtual GPUs (vGPU). The idea is that you assign one vGPU profile to a XenApp server (giving it one vGPU). You should find tables listing all vGPU profiles and configurations on the NVIDIA website. I’ve listed 3 of the vGPU profiles that might be appropriate for you, including GPU passthrough:

  • GPU passthrough - Would give 4 GPUs, 4 XenApp servers per host, 50 users pr. XenApp server
  • K180Q - Would give 4 vGPUs (4096MB frame buffer), 4 XenApp servers per host, 50 users pr. XenApp server (600 / 12)
  • K160Q - Would give 8 vGPUs (2048MB frame buffer), 8 XenApp servers per host, 25 users pr. XenApp server (600 / 24)
  • K140Q - Would give 16 vGPUs (1024MB frame buffer), 16 XenApp servers per host, 12,5 users pr. XenApp server (600 / 48)

Which configuration gives the best performance? Well, that depends on the usage, and whether the applications/resources you use are able to utilize a GPU. Most newer operating systems and applications are able to do this, ie. Windows Server 2016, Windows 10, Internet Explorer and Office, but there are issues with XenApp not utilizing GPU for these. Another thing is whether to use GPU passthrough or vGPU profile. I’m not sure which gives the better performance to be honest (ie. GPU passthrough vs. K180Q). I’d probably go with the K160Q or K140Q vGPU profiles in your environment.

K1/K2 does not require licensing.

Regards
Tore

Hi,

just to add you also need FB, therefore I would only try to use the Passthrough or K180Q config. And 50 users is also a bit high. Needs to be tested.

regards

Simon

Thanks, that helped alot.
One thing to add, i don’t think it will be 600 concurrent users, they are spread worldwide so maybe 300-400 at the same time.

Any more input for which profile to use?

Echoing what Simon says above, this is your best option.

Use the entire GPU with XenApp, do not split the GPU between multiple XenApp servers. You only have 2 (realistic) options, K180Q or Passthrough. Anything else won’t have enough Framebuffer.

As for 50 users per XenApp VM, this is a bit optimistic. Depending on all the usual resource factors, I’d plan for 20 - 30 per XenApp VM, and if you get more then great! But that’s still 80 - 120 users per K1

Regards