Graphics Cards

[david_janicek]

We are a relatively small engineering office, with quite a unique IT-Infrastructure. AFAIK we are the only ones in our country doing proper Virtual GPU stuff in this kind of "line". The downside of this is, we are bit budget constrained.

We are in the process of updating our existing Citrix VDI. However, right now, you either have the option to "go all out" with Nvidia and be in astronomical price ranges, or you go a bit cheaper with direct passthrough and AMD RadeonPro GPUs. Tried Microsofts new GPU-P, which i seem to can not get to work properly (card shows up, but that's about it), so that's a nogo.

However, since HA is also one of the things we would like to retain from our Grid K2 systems, i thought about going the "used" way and get some FirePro S7150x2, since they seem to be supported by XenServer 8.2 LTSR (which in turn is going to be supported till 2025...).

I would like to know, how long AMD is planning on supporting the S7150x2 with drivers? And also, if there are alternatives? The V340 seems to have gone, Instinct Mi25s are unavailable (let alone drivers seem to be nowhere to be found) and nothing newer than the S7150x2 seems to be supporting MxGPU.

The alternative would be to put 4 W5500 GPUs in our shiny, new EPYC Servers...however, that's more like a "last resort".

[elstaci]

fsadough‌  would be the best AMD employee to answer your question.

[fsadough]

Indeed S7150 X2 is EOL'ed. Currently we are not offering any MxGPU product as retail product, that might change though in near future. In general, we want to know what our customers are planning to do in VDI environment:

• What applications are you running?
• Which remote protocol will you be running i.e. PCoIP, Blast, RDP?
• What are the system specs?
• Which guest OS and driver are you using?
• How many GPUs per server?
• How many VM-users per server?

[david_janicek]

Thank you for the answer. The quick question i have now is, how long are there going to be driver updates, containing the 7150's? Since we have to replace our GRID K2 "now" (NV doesn't offer ANY support anymore, even guest drivers for passthrough are "gone"), we don't have much of a choice anyway (As i mentioned, we are going to go AMD, no matter what). And since 7150s are used only anyway around here, it wouldn't be much of an issue, switching to a newer MxGPU capable GPU down the line...

• Applications we run:
  • Intergraph Smart3D (with CloudWorx Addon)
  • Leica Jetstream Viewer
  • Universal Plant Viewer (based on UE4) with point cloud
  • Occasional SolidEdge usage (only 2 users company wide)
  • Other Software that i cannot mention because of NDAs. (They are relatively "light" on the GPU load though, comparable to Smart3D, which really doesn't even use a WX2100 above 50%)
• Which remote protocol will you be running:
  • Citrix, mostly. We are going to evaluate other alternatives in the Future, though.
• What are the System Specs
  • Gigabyte G242-Z10, 256 GB DDR4 3200 MHz RAM, Epyc 7742, local SSD (NVME) Storage
• Which guest OS and driver are you using
  • Depends on the evaluation. Either Windows 10 Enterprise or Server 2016/2019 DC
  • Drivers depends on the GPU.
• GPUs per Server
  • Depends how efficient the system is going to be and what GPU. If W5500/W5700 via Passthrough, 4 GPUs. If the 7150 does the job, 1. I would have loved to get my Hands on a V340...
• How many VM-users per server:
  • Between a minimum of 10 and a maximum of 35

[fsadough]

Which driver updates are you referring to? Guest driver or Hypervisor driver?

[david_janicek]

Guest.

AFAIK, XenServer 8.2 LTSR has Support for SR-IOV capable GPUs "backed in", therefore Hypervisor drivers should not be needed? At least that's what it says on the Citrix Support pages...

[fsadough]

Hypervisor is required in any case on VMware or even on Citrix for VM functinalities. The 20.Q2 driver is the latest driver we offer fore S71xx-series.

[Arturas_R]

Hi David,

I know it is an old post but still, did you manage to get up and running with Firepro cards? we are taking the same path but it does not seem to perform as good as good old Nvidia Grid K1 and K2 cards. Maybe there something we can tweek