vSGA performance Nvidia grid K2 with Solidworks

[arjandv1]

Hello there,

I am currently testing our new R720 server with grid K2 card.

I want to configure our vm's with vsga hardware acceleration.

The problem is that I don’t see any performance improvement when configuring for vSGA.

As a matter effect, no 3d renderer is faster than when I configure a hardware renderer in the view pool.

we are running:

vsphere 5.5

vmware view 5.3

ESX server with a Nvidia grid K2 card

Windows 7 64 bit client

I did the following configuration untill now:

• I installed the Nvidia grid drivers on ESX / NVIDIA-VMware-x86_64-319.65-1OEM.550.0.0.1331820-bundle.zip

       Installation was successful.

• Installed vmware tools latest version
• Installed the view 5.3  agent
• Vm hardware: version 10 (also tried version 9)
• I have configured 3d renderer on hardware in the vmware view pool (also tried to configure this in vcenter)

When working in solidworks i'm opening fairly complex model and try to turn it.

The performance is better when I set the pool to 3d renderer to disabled !

I don't see any performance difference between software or hardware 3d renderer.

Did I forget something here?

I know that vdga gives better performance but we have about 10 concurrent light solidworks users and not much money for more grid cards

I hope you have some good tips to improve the performance.

Kind regards,

Arjan

[Linjo]

Maybe solidworks are using software rendering? Are you monitoring the GPU?

// Linjo

[arjandv1]

Linjo, thanks for your reaction.

the flag software openGL is disabled.

The performance is better when I enable softwareopengl :smileyshocked:

I monitored the gpu with gpuvm and then i get this so it looks that it is working.

~ # gpuvm

Xserver unix:0, PCI ID 0:6:0:0, GPU maximum memory 4173824KB

        pid 2934629, VM "WKS01", reserved 131072KB of GPU memory.

        GPU memory left 4042752KB.

Xserver unix:1, PCI ID 0:7:0:0, GPU maximum memory 4173824KB

        GPU memory left 4173824KB.

~ #

thanks,

Arjan

[Linjo]

How much graphics memory did you provide to the guest? (Max is 512Mb)

Are you sure that its the GPU that is the bottleneck and not CPU or Memory?

You should use the nvidia-smi command to monitor the GPU, gpuvm will only give basic information about the GPU and what guests that are using it.

Is Solidworks using OpenGL or DirectX?

// Linjo

[arjandv1]

I tried with 64 mb and with 512 mb. I dont see much difference in performance.

The gpu is not the bottleneck, i just don't know what it is.

I am sure it is not the memory or cpu. memory is using max 2 gb of the 4 cpu is not getting above 30%

I did the redway turbine demo and in the first two test i get about 30 fps, so i think it is not to bad?

Solidworks is using opengl.

I think the problem might be that i am testing over a wan connection.... can this be it?

Then still i think it is strange that when i put solidworks on software opengl the experience is better then hardware.

see the nvidia smi command below, this is when running the redturbine demo.

+-----------------------------------------------------------------------------+

~ # nvidia-smi

Wed Mar 26 20:19:55 2014

+------------------------------------------------------+

| NVIDIA-SMI 5.319.65   Driver Version: 319.65         |

|-------------------------------+----------------------+----------------------+

| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |

| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |

|===============================+======================+======================|

|   0  GRID K2             Off  | 0000:06:00.0     Off |                  Off |

| N/A   32C    P0    44W / 117W |      128MB /  4095MB |     27%      Default |

+-------------------------------+----------------------+----------------------+

|   1  GRID K2             Off  | 0000:07:00.0     Off |                  Off |

| N/A   23C    P8    17W / 117W |       13MB /  4095MB |      0%      Default |

+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+

| Compute processes:                                               GPU Memory |

|  GPU       PID  Process name                                     Usage      |

|=============================================================================|

|  No running compute processes found                                         |

+-----------------------------------------------------------------------------+

Thanks,

Arjan

[Linjo]

Try to add this registry setting:

HKLM\SOFTWARE\VMware, Inc.\VMware SVGA DevTap\MaxAppFrameRate=dword:00000000

// Linjo

[arjandv1]

Linjo you are the greatest thanks for your help!

Look at the results in Passmark:

Before changing the registry entry:

After adding the registry entry:

I missed  the performance tuning tips in the documentation: "Graphics Acceleration in VMware Horizon View Virtual Desktops.pdf"

Guess I was to exited to start

The performance in Solidworks is still not what I would expect but can this be due to I am testing over a WAN connection?

I will ask someone spin a model onsite.

Do you have experience with solidworks to? the performance benchmark from solidworks 2013 itself only gives 0.0 scores...

Regards,

Arjan

[Linjo]

Did you do any PCoIP performance tuning?

[arjandv1]

Linjo,

No I didn't do any tuning yet, are you refering to this document: VMware-View-5-PCoIP-Network-Optimization-Guide?

Can you suggest specific settings to tune?

Thanks,

Arjan

[bjm534]

Good thread guys. For PCoIP tuning I would suggest that you immediately drop down the bandwidth for audio, because by default it's set to 500. Error message tones can consume a lot of bandwidth with that much freedom. Use some of the tools located here PCoIP | MindFlux Inc. this is a blog written by someone who works in the office of the CTO for EUC at VMware. He has written a few VMware Flings that are very useful in your situation.

-Brad