VMware Horizon Community
fredr3k
Contributor
Contributor
Jump to solution

Nvidia Quadro <3 vSphere 5.1?

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.

202 Replies
TomMar
Contributor
Contributor
Jump to solution

I just updated our R720's BIOS from 1.4.8 to 1.6.0 and applied the the settings you suggested in the BIOS.  It made a significant improvement to playback performance.  It appears that the last bit may be related to encoding the PCoIP data fast enough, I have a Teradici 2800 card I can try once I can get it out of our current View hosts, but that will take some downtime so it probably won't happen today.

Reply
0 Kudos
Bhargava_Shriva
Contributor
Contributor
Jump to solution

Hi Tom,

Kindly share the output of following command

~ # nvidia-smi -pl 0

Kindly observe the changes and let us know will this help in GPU performance Smiley Happy

If u have dell support take the help of them to optimize the BIOS settings. Here my local Dell Support is good for noting. Smiley Sad

And also share the result of APEX card will it help in improving View environment / not.

Reply
0 Kudos
smhvmware
Contributor
Contributor
Jump to solution

I am having the same performance problems with the GRID K1 in a R720.  For most tasks the nVidia card actually seems slower.   For example google maps, from the default satallite view of the USA zoom in several levels.  In a host without the GRID K1 it will take 5-10 seconds to render.  With the GRID K1 card it will take upwards of 45 seconds, sometimes timing out and crashing the browser.  Issue is repeatable.

Also of note, the nvidia-smi -pl 0 does not seem to be a valid command for the GRID K1.  Says to pick a power rating between 10w and 31w.  By default it is 31w (max)

Edit:

Just saw this in the nvidia-smi -q output.

GPU Link Info
            PCIe Generation
                Max             : 3
                Current         : 1
            Link Width
                Max             : 16x
                Current         : 8x

Reply
0 Kudos
Bhargava_Shriva
Contributor
Contributor
Jump to solution

What best can be done as of now is

Check the BIOS (BIOS.docx) settings as you see in my previous post.

Check the ESXTOP result CPU ready value should not exceed 5.

If all the above things are done from our end.

Try taking help of Dell support for optimizing BIOS even more.

Contact Nvidia and VMware support stating vSGA not working as expected.

Good luck Smiley Happy kindly share the result.

Reply
0 Kudos
Bhargava_Shriva
Contributor
Contributor
Jump to solution

Hi Tom,

Kindly share your inputs Smiley Happy

Reply
0 Kudos
TomMar
Contributor
Contributor
Jump to solution

~ # nvidia-smi -pl 0
Provided power limit 0.00 W is not a valid power limit which should be between 10.00 W and 31.00 W for GPU 0000:09:00.0
Terminating early due to previous errors.

This is what I get when I try that command.

I do see the performance level fluctuate but its always related to gpu utilization  I go between P8 at idle and P0 when there's anything that taxes the GPU.

Reply
0 Kudos
TomMar
Contributor
Contributor
Jump to solution

        GPU Link Info
            PCIe Generation
                Max             : 3
                Current         : 1
            Link Width
                Max             : 16x
                Current         : 8x

Are my results for the PCIe bandwidth, I don't think this is your issue though.  I tried google earth yesterday and everything was very smooth.  The thing that made the biggest change was updating my R720's bios to 1.6.0 and making the BIOS changes listed previously.  I'll take some video right now of google earth running and post it.

Reply
0 Kudos
smhvmware
Contributor
Contributor
Jump to solution

How is your performance in the normal Google Maps satellite view and playing a 1080p video on VLC, host with GRID K1 vs  host without?

Reply
0 Kudos
TomMar
Contributor
Contributor
Jump to solution

That's with the Grid K1.  I just started up my other test VM that is still set to software rendering, I'll take some videos of a few applications between the two and post the videos in a little bit.

Reply
0 Kudos
waynej
Contributor
Contributor
Jump to solution

Has anyone tried some of the graphic demos from AMD/NVIDIA?

http://www.geforce.com/games-applications/fermi-hair-demo/downloads

http://developer.amd.com/resources/documentation-articles/samples-demos/gpu-demos/amd-radeon-hd-7900-series-graphics-real-time-demos/

I'm not sure if the AMD Leo demo will run on an NVIDIA card but the hair demo should run. 

Reply
0 Kudos
TomMar
Contributor
Contributor
Jump to solution

The problem with those is that the VM only sees a card with DirectX 9 and OpenGL 2.1 and even then I don't think its the full feature set.

Reply
0 Kudos
Bhargava_Shriva
Contributor
Contributor
Jump to solution

Good Tom,

Can you check 1080p sample video avail in Win7 Smiley Happy

Kindly share me the deatils about your  CPU and storage where your VM is running.

Reply
0 Kudos
nicobal
Contributor
Contributor
Jump to solution

Hum... I thought using nVIDIA's new Grid VGX K line of product would provide full compatibility with: DirectX 9, 10 & 11, OpenGL 4.3 as well as Cuda!

BTW, what about this software?
http://www.nvidia.com/object/grid-vgx-software.html


Does it have to be installed on the hypervisor to get full horse power from the graphics card and/or to unlock specific features?
Thanks

Reply
0 Kudos
TomMar
Contributor
Contributor
Jump to solution

http://www.nvidia.com/object/grid-software-partners.html

If you look on that page, View only supports DirectX 9 and OpenGL 2.1.  I find that not all the OpenGL 2.1 benchmarks will run though because of missing features.

Reply
0 Kudos
waynej
Contributor
Contributor
Jump to solution

Good point on the DirectX version.  The big issue for us is scaling for # of users per K1 or K2 since there is a max of 2 cards per R720.  It would be nice to see a comparison of what load some of the guided tours in google earth put on the different cards.  Seeing the spec sheet of "low-end Kepler" (GK208?) with 192 cuda cores per GPU vs a "high-end Kepler" (GK104?) w/ 1536 cuda cores per GPU leaves more questions than answers as we don't know how their GPU virtualization layer splits up work on the CUDA cores, etc. 

We have some student labs that use AutoCAD (and other CAD programs), some GIS applications (ArcGIS) and Google Earth.  I'd be willing to bet that for us Google Earth will be the heaviest GPU load.  It is fairly easy for even a novice to stress a GPU in that program and I doubt student assignments in the CAD programs will be all that 3D intensive.  But it is important for us to ensure that Google Earth doesn't become a 1FPS slide show if 30 students in the lab are using it at once. 

Reply
0 Kudos
smhvmware
Contributor
Contributor
Jump to solution

The R720 can fit two cards yes, but I was under the impression that View only supported one per host at this time.

Reply
0 Kudos
TomMar
Contributor
Contributor
Jump to solution

I took two videos one with the Grid and the other in software.

Both are running on the same host, the specs for that host are:

Dell R720

Bios 1.6.0

384Gb RAM at 1333Mhz (I think)

Dual Xeon E5-2660 (2.2GHz)

nVidia Grid K1

No Teradici 2800

Both VMs have 2 vCPUs and 4Gb of RAM.  Both have 512Mb of video ram.  Both VMs are stored on the single 500gb 7200rpm local drive on the ESX host.  I haven't setup a ILIO datastore on that host yet and that probably won't happen until next week.

With hardware acceleration:

With software acceleration:

The software video is still being encoded by Vimeo so its going to be 30 minutes or so before you can view it.

Reply
0 Kudos
waynej
Contributor
Contributor
Jump to solution

Does anyone have confirmation from VMware on the # of supported GRID cards in a single server?

Reply
0 Kudos
Linjo
Leadership
Leadership
Jump to solution

Up to 8 GPU:s are supported with vSGA, but I think you will run into problems with power/cooling way before that.

Since the GRID:s are passivly cooled it needs to be cooled by the chassis somehow and that seems to be tricky for the servers.

// Linjo

Best regards, Linjo Please follow me on twitter: @viewgeek If you find this information useful, please award points for "correct" or "helpful".
Reply
0 Kudos
waynej
Contributor
Contributor
Jump to solution

That you for that clarification.

So now I need to figure out scalability.  For example, you have 8 users doing 3D work simultaneously (Google Earth).  Lets assume the hardware is a single server with a single GRID card.  Would you be better off with a single K1 or a single K2? 

The K1 has four slow GPUs and 768 CUDA cores while the K2 has two fast GPUs and 3072 CUDA cores. 

Should I assume that as long as I have enough video memory to handle all of my client VMs, the K2 solution would deliver a better user experience?

Reply
0 Kudos