Looking to implement a POC using VMWare as the hypervisor and XenAPP and a single K2 card. The users will be architects using autoCAD. Since VMware only currently support vDGA and vSGA. We are looking to implement vDGA first a shared desktop. If that fails looking at dedicated desktops. I have the following questions:
1 - Will autoCAD work with this configuration.
2 - What are specific recommendation or where can I find them running it in this configuration
3 - What is the minimum hardware requirements? Do you know any company that will loan the cards for a POC.
4 - What are the gotchas for such a deployment?
5 - Should I rather be looking at Xenserver for vGPU pass through? The reason I want to use VMware as our current hosts are on VMware. As this is a POC and I need it signed off and spend as little as possible. If successful will purchase new hardware.
Yes
Check the Citrix eDocs.
Choose a CPU with a high clock speed as each XenApp session will have it’s own CPU based HDX encoder.
Too slow a CPU, not enough CPU cores allocated, not correctly configuring XenApp for GPU use (see Citrix documentation)
Nope. vDGA on vSphere will be absolutely fine. Note that vGPU and Passthrough are two separate things.
vDGA = Passthrough
Jason,
We are looking at providing a very similar setup for our customers- XA7.5 riding on ESXi5.5- but just applying GPU Acceleration to multiple Xenapp servers for Knowledge workers for general application acceleration. Everything these days is Graphically accelerated, Even Microsoft Office.
Does NVIDIA have any case studies or scaling information for this kind of setup?
We don’t have any documentation yet for XenApp workloads. However, one of our forum members Tobias Kriedl has been doing a lot of experimentation and performance testing with XenApp, and example of his work is here:
https://gridforums.nvidia.com/default/topic/450/configuring-multiple-grid-engines-on-a-single-xenapp-virtual-machine-running-on-xenserver/
Scaling is very subjective depending on the applications used. For VDI workloads we have work underway with a number of ISV’s to provide quidance on scalability / density for particular applications and user types.
There’s a couple of case studies such as this one from Drake University,
You’ll find more here:
Jason
Thanks Jason, Those Case Studies are precisely what I needed to show the other folks in our Department. We are quite interested in moving forward with this technology - how would I go about potentially getting some cards for a POC?
For PoC / pilot hardware I’d recommend going via your preferred OEM.
Different chassis’s have different card requirements plus cabling, risers and in some cases new PSU’s and heatsinks to become GPU ready. For this reason it’s best to initially go via the OEM to ensure that the card has all of the necessary components to fit and power the cards.