[Openstack] State of virtualised GPU use in instances?

Simon Pasquier pasquier.simon at gmail.com
Wed Oct 8 08:07:51 UTC 2014


Hi,

If you're using XenAPI then the good news is that GPU passthrough is
implemented [1] and it works with NVIDIA. I've copied Guillaume that
developed and tested the feature, he may be able to provide more details.
I know at least that Guillaume had difficulties to do GPU passthrough with
NVIDIA cards and libvirt/KVM.

BR,
Simon

[1] https://blueprints.launchpad.net/nova/+spec/pci-passthrough-xenapi

2014-10-02 12:49 GMT+02:00 Jake Carroll <zebra.x0r at gmail.com>:

> Hi.
>
> I’m interested in accelerated high throughput microscopy workloads
> involving deconvolution using stream processors via CUDA interfaces to
> GPU’s.
>
> Last I checked, things were not going all that well in terms of GPU inside
> instances:
>
> https://review.openstack.org/#/c/32803/
>
> I got the impression that the direction being pushed was that a more
> cohesive or heterogenous use of SRIOV for all PCI based devices was being
> considered instead, as opposed to just GPU.
>
> Could someone give me a brief explanation as to where we sit with
> virtualised GPU inside instances? For sake of argument, it could be a Xen
> or a KVM guest. Whatever. Either way.
>
> Thank you.
>
> -z
>
> _______________________________________________
> Mailing list:
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
> Post to     : openstack at lists.openstack.org
> Unsubscribe :
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack/attachments/20141008/174f4360/attachment.html>


More information about the Openstack mailing list