OpenVZ Forum

Home » General » Support » Do you support GPU pass-through for VMs? (especially useful on headless multi-GPU hosts for CUDA or OpenCL GPU compute)
icon5.gif  Do you support GPU pass-through for VMs? [message #52683] Wed, 07 December 2016 11:59
abufrejoval is currently offline  abufrejoval
Messages: 21
Registered: November 2016
Location: Frankfurt
Junior Member
From: *
I hate asking before trying but since you guys are so reactive, a single question may well save me hours of investigation, trial and error...

I know given the proper hardware with VT-d support,BIOS, kernel and kernel parameters, twiddling, a recent enough Xen or KVM such things are possible and you can pass a GPU exclusively to a virtual machine (while some VDI enabled professional devices are even supposed to allow some degree of partitioning of the GPU: Not hoping for that--just yet).

But getting that to work is generally not trivial and I guess considered too 'niche' to receive a lot of attention.

With this AI/MI interest picking up and GPU compute that could change ever so slightly and in fact I'd be happy to have this available as general plan-B or if people come with pre-built VM images they want to run.

So is GPU passthrough something you support already or could be interested to support or something that you know to fail (perhaps because the RHEL7 base software ecosystem just is a little too old for that)?
Previous Topic: php-fpm core dumps inside container / how to enable?
Next Topic: OpenVZ docker version mismatch with CentOS and daemon fails to start
Goto Forum:

Current Time: Fri Oct 18 06:38:03 GMT 2019