Does sVGA really work?

Does sVGA really work?

Recently I was finally able to get a vSGA environment working! It was not easy since nobody actually knows how it works and mainly where to get needed stuff!

My test environment was built from:

- an IBM System x3400 M3 server with 48GB of RAM and 2 6 core processors

- Quadro 4000 GPU with 2GB VRAM

- ESXi 5.1 .0 build 914609

-  View 5.2 build 987719

- nVidia vSGA driver version 304.59

- Virtual desktop VM with HW version 9 and Windows 7 64 bit

- Client device: TERA2321 PCoIP Zero-Client

First impressions:

After powering on the virtual desktop I installed some OpenGL and DirectX benchmarks as well as Google Earth setup in OpenGL mode.

It’s clear that everything got smooth and faster, and OpenGL applications that did not run before was now able to run without problems. I tried to run 3DMark benchmark, but I received some errors and I was not able to debug it yet.

For now, all I can assure you, is that performance is not as good as I expected! I was hopping to see my virtual desktop runs the same as a graphical workstation, but it’s not what I got. Of course we are in the very beginning of the implementation, so I’ll do my homework and post here every improvement we get.

Keep watching if you, like me, are sure this kind of environment is the future of edge computing!!

***  Returning to test bench, we have the following updates:

- The disappointing result I mentioned before was not caused by vSGA itself… the problem was that the VMware View Client limits bandwidth even in the default, (was not supposed to limit it!). We got in logs something around 110Kbps in a link of 20Mbps. After some adjustments (and a lot of help from Teradici) we were able to get 1,5Mbps from the client, what resulted in consistently +45fps in benchmarks.  Image quality got superb either! I am pretty sure we can run AutoCad and solidworks with good results – as soon as I test it I’ll post here.

- Even though vSGA should not affect 2D renderization, we actually saw some improvements in 1080p video playback (something that I don’t know why to run in a virtual desktop, but all customers requires…). We saw a lot of frame drops before vSGA and very little frame drops after hardware accelerating the vDesktop. Contradicting all expectations, 2D was improved too…

Next week we are going to install and test some CAD software and run complex modeling. I’ll post results here…


Hi Wilder.

What kind of 3DMark benchmark did you run?

Please be aware of that with vSGA you are limited to DirectX9.0c and OpenGL 2.1, maybe that is why the benchmark failed.

(These limitations are not present in vDGA that will be released "soon" but that is a 1:1 mapping from VM to GPU)

vSGA is performing about 50-100x faster then software rendered 3D graphics but it will not accelerate thing as video and 2D applications.

Cool thing about vSGA is that it is fully compatible with HA and vMotion, you can even move a VM from a host without a GPU to a host with a GPU without the user beeing affected and no restarts.

Please keep the feedback coming since we are very interested in hearing more.

// Linjo

Hi Linjo,

Since vSGA today only supports OpenGL 2.1 and DirectX9 I am using 3D Mark06. Talking to Teradici we agreed that nVidia driver is still not full functional, that’s why 3D Mark is not running. I heard some reports from HP running 3D Mark, but I found out they were using vDGA (passthrough) that, since VM uses nVidia reference drivers, supports all flavors of OpenGL and DirectX.

About performance, I am not sure we will get 50x better performance than VMware native 3D renderer, but running Glaze benchmark I got the following:

- VMware 3D renderer: about 12 fps in the most demanding tests

- vSGA: 45 fps (limited by software)

Results are very impressive: using VMware renderer, image quality was pretty poor with very low quality shading and textures; with vSGA quality was very good, with nice textures and shading! I am sure that I would get more then 100 fps if there were no software limits.

Something I discovered: View Software client limits bandwidth even when, by default, no limit is set. I had to manually override all limit parameters to get the result I wanted!

Are you using vSGA? What GPU? What are your impressions about it??


Not sure what Teradici is talking about, they have very little todo with vSGA.

The driver works fine for what it is intended to do, sure there are some bugs and improvements but nothing that is catastrophic.

I'll try some 3D Mark tests when I get time over and get a blogpost togheter with the results.

I usually run Gpucapsviewer to validate the functionality, test vmotion etc, have not tried the Glaze benchmark.

The clients are defaulting to settings that would be appropriate to the majority of the users, in these usecases you need to modify them for sure.

I am using both software rendering, vSGA and vDGA.

For me vSGA is the current sweetspot with accelerated graphics with acceptable consolidation ratios.

// Linjo


Kindly share the steps to achieve vDGA.

Which all GPU supports vDGA.


Thank you for sharing your configuration.  I am trying to find out more about vSGA before we head down this road.   Are you able to share the card to multiple Virtual Desktops?  I am getting mixed messages from NVidia:

"I know you are looking for a higher density solution but at this time  we have not yet enabled multiple users to share  a single GPU"

I am looking to setup a POC to ensure this will help us with our linked clone pools and some 3D apps that do not render properly with SW GPU's.  If i can't share the card I will wait until I can, but it appears that I can?

Thanks in advance


Version history
Revision #:
1 of 1
Last update:
‎03-25-2013 03:58 PM
Updated by: