I assume that many of you with exictement have read through all news in vSphere 5.1
What cought my eye in View point of view was this:
Improved 3D Graphics Support – (View Only) – hardware acceleration with possiblity to leverage NVIDIA’s hardware cards installed in ESXi server, where those graphics cards are virtualized and used in View Desktops. It’s targeted for graphics intensive workloads, CAD designers, medical imaging etc…
NVIDIA Quadro 4000/5000/6000 and NVIDIA Tesla M2070Q are supported graphics cards. Note that the ESXi Image profile must be installed with the NVIDIA GPU VIB file."
It certainly raises some questions for me:
a) The Nvidia GPU VIB-files is that free of charge or is it licensed from Nvidia or Vmware?
b) Have anyone tested 5.1 with Quadro-cards (and VMware Techie out there?) Pros? Cons?
c) Is the GPU fully accessible from the View guest or is there some kind of abstraction layer in between? After some research it seems like the abstraction layer is utilizing Xorg in a clever way.
That is good to hear! When you do this do you notice much of a difference in application performance? How do you do this; is it with the "nvidia-smi" command?
I do not believe you can oversubscribe VMs utilizing GPU RAM like you can with CPU RAM... yet.
Right now I would just use the 256 MB of GPU RAM as your guide since that will support a VM with a total of 512 MB of GPU RAM (256 MB GPU RAM / 256 CPU RAM). Using that figure and a K1 GPU you could theoretically support 64 VMs with 512 MB of GPU RAM. If you want your VMs to only utilize 256 MB of GPU RAM (128 MB GPU/ 128 MB CPU) it would be double the amount of vSGA VMs (128 VMs).
Nathan is correct, there are currently no oversubsscription of Video RAM with vSGA.
His other points are also correct, you could however also put several cards in the host,(if you can power and cool them properly, I know that we fried a few cards in our testing...:-)
When interfacing with Nvidia at a tradeshow, they gave us a few utilization points. I believe it was on the K1 (lower-end) and it really depends on what programs you will be using. Revit for example will support 20 users per 1 card (AutoCAD I think is 10:1 or 14:1). I don't think the RAM is going to be the issue, rather the GPU chip itself.
Since only RAM can be assigned, from what I see now, I don't think there is a setting to limit GPU over-utilization.
How are performance settings of the card being modified (these P0-P12 states)?
Yes, you are right about it being more about the GPU power (core count) than the available VRAM when using 3D intensive apps within each VM. The figure I gave was merely just a calculated number of VMs per the available GPU VRAM. All those VMs would be able to use vSGA, however the performance wouldn't be very good if they were all using something like AutoCAD. If you are using intensive 3D apps like AutoCAD, etc... the limiting factor will definitely be the GPU chip. If you are planning on using intensive 3D apps I would say go ahead and use the K2 over the K1 since you would probably hit the GPU performance limit before you run out of available VRAM.
It’s simple as of now if u take currently there are 2 technologies
vSGA – will support 50% H/W and 50 S/W GPU as of now. Need to wait and see in feature outcomes of Nvidia and ESXi % of H/W GPU utilization may increase.
However advantages of K1 and K2 is you can set power state to max
Kindly run the following command and check and share the result will it helps are not currently these is not supported by Quadro but on GRID K1 and K2 it should.
nvidia-smi -pl 0
I am running it on quadro 4000 I am getting following error
~ # nvidia-smi -pl 0
Changing power management limit is not supported for GPU: 0000:05:00.0.
Treating as warning and moving on.
Changing power management limit is not supported for GPU: 0000:42:00.0.
Treating as warning and moving on.
There is one more technology vDGA currently not supported by VMware View
in this technology
K1 – 4Entry level users
K2 – 2 High end users
this is more like assigning 1 GPU for 1 VM
vSGA is memory based sharing. vDGA is GPU based sharing.
Where did you find that vDGA is currently not supported by VMware View?
I think the key word here is "supported". When View was in beta earlier this year I successfully used a passed through Quadro 4000 with View. It is currently "unsupported", but I believe there are plans to support this with View at some point in the future. I have not tried this again with the GA View 5.2 so Im not positive it still works with the GA version. With the beta version there wasnt any settings specific to view to get it to work. Just pass the GPU through, switch primary display to the Quadro, and connect with the view client.
What kind of performance is everybody seeing with this. We just upgraded to 5.2 over the weekend and installed a Grid K1 into one of our R720s. There's nothing else running on the server besides my desktop. I see memory being reserved on the GPU, but dragging windows around is still laggy and full screen video is not smooth. GPU utilization is about 40%.
For people who have it working, what kind of user experiance are you seeing? So far I'm very disappointed.
We're using ESX 5.1 Build 1021289 with the 5.2 agent in the VMs.
The GPU will not help with window dragging and full screen video, its purpouse is to accelerate 3D content like OpenGL and DirectX.
You would probably get better experience with windows dragging and video by disabling 3D on your pool.
Have you done any other tweaks to PCoIP etc?
For the window issue you may want to try this fix:
Since I have the McGyver'd GTX 680 into a K2 I can say performance is lacking. I was expecting it to be able to smoothly run Windows Media Center to a Tera2 terminal (it doesn't).
I'm correct in that there is no management software right?
I don't believe that Linjo. If I start playing a flash video from youtube, I see GPU usage go up on the host. Aero in windows 7 requires directx, I don't see how this would not accelerate that type of application. In another example, I run the GPUCapsViewer you posted earlier in this thread. The benchmark for the furry box reports 27-31fps, the resultant display on the zero client or on the software client isn't smooth at all.
Also, doing Windows Key + Tab with a hardware GPU VM is much much faster than if you do it without hardware acceleration. So I can't believe that the GPU is not assisting in windows aero performance.
Sure, Windows Aero will be accelerated but just basic window-dragging will not be...
My point was that the primary function of vSGA is to accelerate 3D not video or dragging windows, just want to set the right expectations on this technology.
So why does windows key+tab still run badly then? It's basically a glorified Rolodex of flat 3d surfaces accelerated by your gpu. I'm just wondering at what point do I stop believing all the vmware sales reps who said hardware gpu support would solve these basic user interface issues and now they aren't.
I don't know why, in my setup Aero works fine including windows key+tab even when not using any hardware acceleration.
Maybe the problem is elsewhere in your setup? Have you gone through the standard PCoIP optimization and the Windows tuning best practises?
Here is the solution for your H/W.
Unfortunately I am also using the same H/W
Dell R720 performance makes big difference with small settings at BIOS.
Unfortunately our Dell R720 is in production I am not in the position to make changes and check it frequently every weekend I do it.
I have tried my level best and now it’s better than before.
As u have Grid K1 kindly try following command and share me the result
~ # nvidia-smi -pl 0
PFA for BIOS setting for Dell R720
Linjo , when you say Aero works fine do you mean fine = acceptable or fine = really well same as physical pc .
Do you remember generally what the settings were related to in the BIOS?
The only thing on this R720 right now is my desktop, when I get to the office today I'll shut it down and check the BIOS settings for anything that seems off.
Sorry just saw the attachment, I'll check these out on our R720.