Hi,
I´m using GRID K2 in ESX 6.
Atm there is sth like loadbalancing between GPU0 and GPU1. If I power on 2 K240q on a host the first VM is allocated to GPU0, the second GPU1. I cant power on any k220q vm on this host.
I want to achieve that both k240q power on at GPO0.
How do I configure this?
Any help would be appreciated
Marco
On ESX host, add "vGPU.consolidation = true" to file /etc/vmware/config
That then "fills" each GPU first before moving to the next GPU, instead of the default which spreads load across the available GPU’s.
With vSphere you cannot currently specify which profiles run on which physical GPU’s, though it is an area of investigation for VMware.
As JAson says we provide the virtualisation vendors the APIs to implement allocation but it is VMware/Citrix who implement and control it so feature requests/bugs against that behaviour should got to Citrix/VMware.
I am aware of one caveat (known issue) with vSPhere - that I’m not sure has been resolved by them it might be worth reading the driver release notes for the version you are using and checking if you see it: Known Issues and Resolved issues: Release Notes GRID VGPU FOR VMWARE VSPHERE - Version 361.40 / 362.13 - April 4th, 2016 | NVIDIA (ctrl-f for "allocation") - if you do see it log a ticket with VMware so they issue with a fix.
It might also be worth commenting on your needs on the VMware forums (there’s an NVIDIA specific one) so your need is highlighted with them.
Best wishes,
Rachel
Thank you - works like a charm!
Is there a place where I can look for informations like this?
Marco
Hi Marco,
Glad to hear Jason sorted you out :-D
I’d have thought it should be in the VMware deployment guide: http://www.vmware.com/content/dam/digitalmarketing/vmware/en/pdf/products/horizon/grid-vgpu-deployment-guide.pdf
I’m going to raise a ticket to see if we can get this better documented ourselves (although it is a hypervisor feature and we try to just provide links to the organisation that supports/maintains features)… VMware do document it in a useful section here: VMware Docs Home
Best wishes,
Rachel
It’s already documented in our driver release materials and in the VMware deployment guides.
Section 2.2.9 on Page 23 explains it.