NVIDIA
Talks with Andy Currid - Why the K100/K200 vGPU profiles, shouldn't they have more users?
Q: Looking at the vGPU profile K200, shouldn’t the max number of users per board be higher than 16 if you are allocating 256MB to each VM? Andy: The maximum number of users isn’t solely constrained by the amount of GPU framebuffer available. We currently have a limit of 8 users per physical GPU in Kepler. We sized the FB for the various vGPU types in line with the expected requirements of specific use cases, and 256MB is about right for entry knowledge worker, hence the K100/K200 vGPU types. With future GPU architectures that go beyond 8 users per physical GPU, you’d be able to accommodate more than 8.
Q: Looking at the vGPU profile K200, shouldn’t the max number of users per board be higher than 16 if you are allocating 256MB to each VM?

Andy: The maximum number of users isn’t solely constrained by the amount of GPU framebuffer available. We currently have a limit of 8 users per physical GPU in Kepler. We sized the FB for the various vGPU types in line with the expected requirements of specific use cases, and 256MB is about right for entry knowledge worker, hence the K100/K200 vGPU types. With future GPU architectures that go beyond 8 users per physical GPU, you’d be able to accommodate more than 8.

Regards,

Luke Wignall
Performance Engineering Manager
NVIDIA | Worldwide Sales ­ GRID Computing
http://www.linkedin.com/in/lukewignall/
https://twitter.com/lwignall

#1
Posted 10/10/2014 10:45 PM   
Scroll To Top

Add Reply