NVIDIA vGPU and Linux - support and licensing

Hello,

I’m developer/architect for a software product for working with radiological images (CT, MR, X-ray, ultrasound, etc). The product uses VNC-based thin-client software to access VDI-like desktops that run the application software. The application software run on Linux-based servers with NVIDIA GPUs for both 2D and 3D rendering of the image data. Customers are increasingly requesting that our Linux-based servers should be virtualized. We have experimented with vDGA and it works, but it is not very flexible.

I have been reading about vGPU and it seems like a better option, but the documentation is very sketchy and focuses heavily on Windows-based VDI.

vGPU seems to target solutions where each VM have a single user. For our purposes we would have few (Linux-based) VMs with many simultaneous users. is vGPU still a fit?

Is it possible to use vGPU with Linux (CentOS 7)? I have read some disturbing things about vGPU requiring software licenses? Is this so?

Final question: Does NVIDIA have programs that provide developers with Tesla and GRID GPUs?

Yes, vGPU can be used for session based solutions such as this.

Yes it’s possible with CentOS 7, the Linux driver for GRID vGPU supports 64Bit Linux
Yes there is licensing based on feature set. Linux falls under the virtual Workstation Edition. There’s a document detailing how the softeare is packaged and licensed here:

There are frequently discount programs in place for registered ISV partners. I don’t have details on that to share it would be handled through developer relations.