News for hackers. Stuff that matters.

A comment on Talk write-up: How to build a PaaS for 1500 engineers. Parent comment

Re:Talk write-up: How to build a PaaS for 1500 engineers by mushufasa on Thursday January 09, 2020 @06:33PM (#22007030)
I wonder if AMD will try to change that in the next couple years. AFAIK the reason for this is some clever licensing clauses by NVIDIA that effect market segmentation. If there were more competition on the high end, there would be downward pricing pressure.
[ Parent | on HN ]
Re:Talk write-up: How to build a PaaS for 1500 engineers by anon9001 on Friday January 10, 2020 @01:54AM (#22008892)
It's also annoying because SR-IOV would be a wonderful thing for consumer GPUs, but it would make it too easy to use consumer GPUs for cloud providers.

Right now, you can run a VM with qemu and pass through the GPU to the guest OS, getting pretty close to native performance. With SR-IOV, every VM could have the same GPU attached, and you could manage performance with the hypervisor. This would let you toggle between VMs instantly, getting full performance on each one (assuming the others are idle).

AMD and nVidia do make SR-IOV cards, but they're extremely expensive, intended for data centers, and don't have display output. If it ever hits consumer cards, Linux will be the hypervisor of choice for pretty much everyone, because there will be minimal performance penalty for using VMs.

[ Parent | on HN ]
Re:Talk write-up: How to build a PaaS for 1500 engineers by latchkey on Friday January 10, 2020 @11:08AM (#22011918)
https://www.blockchaintechnology-news.com/2019/01/07/consens...
[ Parent | on HN ]
Re:Talk write-up: How to build a PaaS for 1500 engineers by streetcat1 on Thursday January 09, 2020 @06:46PM (#22007111)
I agree.

Another option would be custom chips for inference or training. IF we can get something like a TPU in house.

[ Parent | on HN ]