Using GPUs as accelerators for application workloads.
- David Power (powerd)
- Blair Bethwaite (b1airo)
- Initial setup of GPU environment with passthrough configured per PCI device. (Blog: http://www.vscaler.com/gpu-passthrough)
- M3 cluster production GPU passthrough deployment at Monash University
- OVH gave a nice talk at the Sydney summit - https://www.youtube.com/watch?v=1tdvz3ejokM
- For some GPUs, the hypervisor signature needs to be hidden
- By image, img_hide_hypervisor_id in https://docs.openstack.org/python-glanceclient/latest/cli/property-keys.html
- By flavor, patch - https://review.openstack.org/#/c/555861/
- Performance evaluation of the GPUs within a virtualised environment in comparison to a bare metal environment.
- Upstream guide docs on how to build a GPU-enabled OpenStack cloud
- initial draft notes underway at https://etherpad.openstack.org/p/scientific-wg-gpu-config-guide
- collating details of known working and not-working GPU-hardware-hypervisor combinations at https://etherpad.openstack.org/p/GPU-passthrough-model-success-failure