Regards to the latest video from Wendell on Intel Flex GPU. When the KVM is setup and running Windows Guest OS, is it the normal RDP client that are used and is that a fast enought client for VDI or are other Remote Desktop Clients used in a VDI setup with the Intel Flex Cards. Does it just work out of the box?
Alternative there is this install guide, but it never talks of options to the Client used and so on.
I have the idea that the normal windows RDP client is to slow or not made for the Intel Flex GPU or am i totally wrong, since it can work with high enough framerates.
So maybe I could run a Promox Server as native HyperVisor, but since Proxmox dont support Flex GPU Native as I see it, I would have to make Ubuntu Server running KVM as Nested virtualization where the proxmox does full GPU passthrough to the Ubuntu KVM Hypervisor. And on the KVM make the Guest OS. Maybe it could work.
I have 4 workers working with CAD programs, that need to go over to VDI instead of local computers. Thats why I think the new GPU Flex from Intel was very interresting.
Does anyone know the allocation ratio for these flex gpu’s? I wanna buy/requisition a pizza box server with 2 flex gpu’s but wondered if it was possible to assign those 2 gpu’s to 4 vdi client’s in an esxi horizon pool?
Or
Is this a 1:1 allocation (1 real gpu to 1 vdi instance) unlike the vcpu allocation insanity we already have.
To be clear
We have an esxi 8.x environment running 400 vdi instances on Horizon across 6 hosts (12 nodes) with 2vcpu per client. No we don’t have any hardware acceleration currently, we’re data science heavy now though so… We need it now. Nutanix boxes may support flex but not 100% sure.
I would say that 7:1 would be a reasonable vdi experice for data science with flex170s. So 14 people, about 2gb vram each? Vmware is even more advanced than sriov in that you can have asymmetric allocation
