Server End Graphics Card

[TSher]

Hi,

Would you benefit from putting in a high end graphics card in the server (vSphere & View 4 environment) when using PCoIP or RDP protocol to a Thin Client or Zero Client? Assuming the VM is either XP SP3 or Win v7. If not does anyone know when the Teradici cards will be fully supported by VMWare for PCoIP?

Thanks.

[SWCS833]

Currently there is no way to put a graphics card in a server and offload the graphics processing from VM/ESX with View. The only way to do that is using the 1:1 model from Teradici.

I have had discussion with VMWare and Teradici at a very high level, but I think it is something they are looking at. It sounds like it is just a design challenge of somehow offloading the graphics processing from all the VMs in ESX to a GPU. They would either need to create a special GPU or somehow get the main GPU vendors like Nvidia and ATI to create a driver or what ever it would be to allow this.

To sum it up, in order to have a dedicate GPU to do the processing, you need to do the 1:1 model with Teradici. You can broke this with the View agent, but its not VMWare View with VMs in ESX. Lets just hope VMWare and Teradici are truly think about this concept of GPU offloading.

I didn't know that MS and Citrix had announced RemoteFX when I first posted, which I guess allows Hyper V to use a shared GPU. I guess Microsoft is able to do this now. Here is a good link from Brian Madden

The question now is, where is VMWare on this type of technology?

[gunnarb]

I have heard right out of Teradici's mouth that this is something they are working on. It's a two fold problem, one being that the graphic card vendors see this is a major concern becuase if one graphics card can serve 20 users that is not a good thing for their sales. Personally I think they should think like Wyse on this and instead of being 100% hardware focused you add some software licensing into the mix (think Wyse TCX). The second problem is that VMware/ESX doesn't support virtualization of a GPU, I imagine the problem being how do you support all the GPUs that exist, plus how to use one card in multiple sessions. Teradici was telling me about how Microsoft is going about this and explained something I can't explain, I know they monitor these forums like a hawk so they may chime in.

So the short responce is, they know, and are working on it. A solution will come becuase frankly everyone using View demands it, so VMware will have to open up the ESX side to support it, and the graphic vendors will figure out a model that keeps them profitable.

-Gunnar

[TSher]

Hi,

Thanks to the both of you for taking the time out to explain the situation. So it looks positive, which is the main thing. Once again, thanks.

[admin]

We for sure have some things cooking in this space and have for sometime, we talked about what the future will look like in this area in 2009 at VMworld during one of the break out sessions. Anyone who knows VMware well knows the progression of things typically starts with personal Desktop Virtualization. So, Workstation and Fusion and then makes it' way to hosted ESX/ESXi. We were the first to enable and deliver 3D and Aero with Virtualization. It obviously is a complex problem to solve but we know it well from all the work we have done in this space to date.

As other have said, we see a progression here and things are staged out, we try to map things to the world today and the world tomorrow to cover the most broad set of possibilities.

Today, you will be hard pressed to find a datacenter class server with an x16 slot or more than one x8 slot. On the flip side, it will be tough to find a GPU that is not x16. So, there is a lot of convergence and still some innovation by all parties Server guys, GPU guys, Software guys that will need to happen before this becomes main stream. So, it's more a question of when not if. When will it converge for main stream use? I am sure early adopters might take rack workstations and do it, but, I doubt they will look to fill up their datacenters with them anytime soon

We think we are on track to meet today's demands and use cases. If you have specific requirements in this, example of the app's you want to enable and the class of GPU you deliver it with today in the physical world, feel free to shoot me a PM. I would love to hear them and make sure we are on the right path to meet both your short and long term goals.

WP

[gunnarb]

Warren,

I'd be happy if my Chess worked again, my lunch shift schedule has been all messed up since my chess was taken away from me. Even in 2D mode it is extermely slow and clunky. What do you think time wize it will be before we see some simple 3D rendering available within a View VM? (Chess is just an easy example)

-Gunnar

[admin]

You know I cannot talk about time publicly

WP

[PCoIPinsider]

TSher, Regarding your first question, multiple VMs sharing a

server-based GPU is not currently supported. Regarding the second

question, next-generation PCoIP silicon that addresses VDI requirements

is indeed in development. Teradici will be releasing further details

later this year.

-Ian