Hacker News Reader

Not really the point of the paper's pitch, but I work for a company that does cloud GPU stuff. Currently, we have no safe way to share GPUs and over-provision them in the cloud in a cluster deploy. So we have one GPU per kubernetes pod. There are all sorts of problems with that. The idea of vGPUs is very exciting to me. I would like to test this out.

View on HN

Ultimately that's a driver issue. GPUs have had the equivalent of an IOMMU for almost two decades; longer than virtualization has existed in commodity hardware. But the heavily closed nature of the hardware, and it not really being a priority for the GPU vendors has led to it not getting the love it needs in that regard. You don't need to add vGPUs to get there.

https://www.youtube.com/watch?v=_CQmomyOiRM

This seems to claims vGPU works with NVidia Grid.

The real issue you're facing is with over-provision them?

Yeah, it's industry wide. We work directly with nvidia. They use our product. I thought the paper was interesting, definitely not a magic bullet for the problem. You used to be able to over provision with kubernetes 1.8 and before, but they fixed that useful bug for good reasons. I expect we'll have an official solution in Q1 next year, but I'm always on the look out to explore related implementations.

Reply on news.ycombinator.com