Anyone familiar in configuring NVIDIA's vGPU on RHEL with KVM?

[hoon0304]

Hello, I am new to this community. 

I have been configuring NVIDIA vGPU (with NVIDIA Telsa p40 -- was too expensive imho) on RHEL with KVM. 

I gave each VM: 12gb of Memory, 4vCPU, and allocated with vGPU type P40-4Q.

The VMs are properly licensed with the NVIDIA's vGPU licenses (Quadro vDWS).

I am connecting to my VMs via RDP. 

However, I am convinced that the performance is not quite strong.

Is anyone familiar in this project?

Thanks for your contributions!

-Kihoon

[DDo]

A long time ago, long before there was cloud or docker or container, we use GPU for scientific computing. Don't know where we are now, but when you stated that "performance is not quite strong", I hope you can elaborate on that. Something for me to learn with the hope that I might able to contribute.

Thanks

Dzung

[hoon0304]

Hello, Dzung

Thanks for your interest in this post.

Keep in mind that what I am working with is Virtual GPU (vGPU) - where one physical GPU is shared among multiple VMs.

I downloaded a designer application called AutoCAD Inventor inside my Win10 VM.

I ran a sample 3D car engine file with that application. Whenever I rotate the engine, there is a slight delay (and it's buggy) when rendering the 3D model.

If I run the same 3D file in my laptop (12gb RAM, 4CPU, win10) it is a lot smoother when I rotate the file. There is no sense of bugginess(?) either.

Hope this kind of gives you an example of what I mean by "performance [being] not quite strong"

-Kihoon