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.
Need last input from your end.
Have you tried APEX card will it help in improving performance?
Do u have GRID K2 do we get better performance in K2 when compared to Quadro and K1 GPUs.
Thanks in advance
Saibot, I modded my gt 640 to a k1 today but have been unable to get xorg to start on my host. just wondering if you had any success or any tips. The driver successfully loads but x fails to start. also nvidia-smi fails w/ an io error to /dev/nvidia0 though it does exist.
i've attached the nvidia bug report log if it is of use.
anyone tried a k4000? driver doesn't seem to be the same at all
unable to load nvidia module or start xorg
K4000 is not supported and currently do not work with vSGA.
NVidia have to support it in the driver for it work and they have not yet.
First off, thanks a ton for your contribution to this thread. It has been very useful, and has tought me a couple things along the way.
At this point, I am not getting an output from the gpuvm command. I've documented my environment and issues in the attachment.
I think the motherboard isn't compatible with the Quadro 6000 card, so I'm hoping somebody can confirm for me. Am I out of luck at this point, or is there something else I can try? How do I definitivly confirm that I have a compatibility issue?
I am working with a few that have similar problems but have not been able to figure out the issue yet.
Could you attach our xorg.log pleas? Its located in /var/log/Xorg.log
Are you able to install Windows 7 on the host to make sure that the card is working properly in your host?
Hey Linjo-- Thanks for the super fast response! I've added the log to my originial post, so that everything is in one place.
Yes, it seems that its the same issue as gigogig have mentioned before.
He has opened a support-issue on this.
Please file a Support Issue if you can and refer to SR number 13313482504
I have a Dell R720 with a K1 will try a K2 later, right now im doing a couple of tests
Performace an stability seems to be lacking reminds me of my times of PC Gamer.
I've applied the registry change to remove the devTap cap but i think PCoIP is the issue now...
PCoIP crashed and my session got disconnected when using 3D Mark06
Using the vsphere console it runs and finishes with a score of 5799
Windows Movie Maker 2012 does not run apparently its an issue with the DDI Version of 9Ex, fortunately 2011 does work...
Traditional benchmarks have is of limited use in a remoted desktop since it does not involve or measure one critical components and that is the remoting part. (PCoIP)
Another thing that theses benchmark does is to change/set the resolution multiple times to get a fair benchmark score and this is another thing that is not a great idea with a remoteing protocol since PCoIP and the client need to keep up and change resolution also, this is sometimes casusing PCoIP to fail.
So conventional graphics-benchmarks might not be the best way to test vSGA/vDGA instead I would suggest to perform some real test with 3D applications.
Anybody notice any performance differences with normal 2d stuff? Not 3d, but the 2d things such as highly graphical web pages, google satellite view, playing video's with VLC? In testing these things I show it is using the K1 GPU by the spike in the GPU utilization percentage, but the performance is slightly worse then if I run the same things on a host without the K1 card. Video's are less smooth/more jumpy, heavly imaged webpages tend to be more jumpy, etc. Same specs on both hosts down to the BIOS settings, only difference is one has a K1 card and the other does not.
Tried today AutoCAD, 3Ds Max, Inventor and Google earth and it ran ok which is mainly what we are going to be using in production .
smhvmware, I've noticed the same thing. I've had a ticket open with Teradici about this. Apparently the VMs are rendering all this at around 30fps but for whatever reason the frame rate that PCoIP is able to encode is only around 15fps. This weekend we'll be installing our apex 2800 cards in our K1 hosts so I'm hoping that will help. Otherwise I'm supposed to have a webex with Teradici but it hasn't been scheduled yet.
You may not gain anything better than what you are currently getting when installing the Apex 2800 card when playing video.
We are currently talking with Teradici support when using their apex 2800 card. Right now the scenario we have is two Windows 7 VMs being remoted into by quad display tera2 zero clients. One of the VM's displays are being offloaded by the apex card and the other one is not. Both VMs are clones so all settings are the same. When the offloaded VM plays a video file the PCoIP session's FPS (measured by the zero client's webpage statistics) is limited to ~21 FPS, however when the non-offloaded VM plays the same video file the PCoIP session's FPS is not limited and runs at about 30 FPS (what the actual video player application is able to decode at). The video application is stricly 2d video and is not utilizing the GPU (vSGA) to decode the video; it is all CPU. The response so far that we have gotten back from Teradici is that it looks to be a software bug with respect to the apex card and they are working to resolve it and will more than likely be pushed out with firmware 2.3 in the next few months. They mentioned some test builds may be avaliable in the next week or so, but I havent gotten a confirmation on that yet.
This tool is also useful for gathering PCoIP Data PCoIP Configuration Utility Release &#8211; Version 1.0 | MindFlux Inc.
I have been running APEX 2800 v2 cards and it doesn't seem to matter. Can I get your Teradici/Vmware ticket numbers so I can reference them? I am having a hard time proving it to VMware as it is subtle and they are now telling me that the K1 cards are only supported with ESXi 5.1 U1 which ironically causes the APEX 2800 cards to cease to work.
I played around with that PCoIP tool in the above post and got interesting results. I was playing a 1080p Star Trek movie trailer in VLC. Both hosts are identicle aside from the K1 card and both have APEX 2800 v2 cards.
-With build-to-lossless enabled total max B/W was 35Mb/s, with K1 card was jumpy and without K1 card was less jumpy (though still slighly jumpy on panning frames)
-With build-to-lossless disabled, with K1 card was now very smooth with a total max B/W of 10Mb/s and without the K1 card became jumpy with a total max B/W of 4Mb/s
With build-to-lossless disabled the GPU was also running about 5% more utilization
Do you know what the difference is between v1 and v2 of the apex 2800 cards? Is it as simple as looking at the part number or some identifying information on the card itself? I'm not sure what version we have (eval cards), but would like to know.
v2 is the low profile version. v1 is full height.
That tool is helpful and I normally use it, however when you have more than one monitor and need a per display FPS reading the zero client's web statistics page is more detailed.
smhvmware; thanks for the info.