Just skimmed the video…
If only you and I met up @wendell. I live in Vancouver. You could see my passthrough setup with working Looking Glass for Twitch streaming… *sigh*
Just skimmed the video…
If only you and I met up @wendell. I live in Vancouver. You could see my passthrough setup with working Looking Glass for Twitch streaming… *sigh*
@SgtAwesomesauce A casual perusal of eBay shows GRID GPUs starting in the $300 - $400 range. Not chump change, but achievable for a dedicated enthusiast.
For example: https://www.ebay.com/itm/Nvidia-GRID-K520-8GB-Cloud-GPU/292623079201?
EDIT: And if you prefer Team Red: https://www.ebay.com/p/AMD-FirePro-S9100-PCIe-Server-Graphics-Card-100-505925/20013763851
As far as I know, without SV-IOV, you can have multiple VMs using the same GPU, but only one VM can run at a time.
SR-IOV only helps if you want to run all VMs in the same time.
But then why would you run more then one VM with a dedicated GPU at a time ?
Can you play 2 video games in the same time ?
Unless you want to keep the other VMs on stand by, so you don’t have to wait until they boot, but then you’ll need a lot of RAM.
My use case would not be 2 VMs, but one host and one VM.
Additionally, I’d probably want one OSX VM and one Windows VM.
That said, enterprise use case is so they can buy a more powerful GPU, run multiple VMs and get a higher density of computing.
An even cheaper find, but I’m not sure if this card does SR-IOV. I think it’s designed for HPC, but there might be a way to persuade it to work with VMs.
Actually I’m curious about this one. Does anyone know if a Tesla GPU can be used for VM graphics acceleration?
But what if you also have integrated graphics, and you use that for the host.
What if that’s not powerful enough?
I also don’t have any more desktop systems with integrated graphics…
Would probably be better off using that in its own VM.
Passing through an iGPU is actually a fair bit more difficult.
It’s doable, but if your time is worth even minimum wage, it’s cheaper to buy a 1050ti to pass through instead.
Hmm. Didnt know that.
If you only do all graphics demanding tasks on the VM (like video games), then it doesn’t matter.
Has to do with initialization, IOMMU groups and the way onboard components communicate over the PCIe fabric.
That’s a bit of a narrow use case.