Before you read this, I must disclose that I have never done this type of stuff before only seen this Wendell pull it off before. And now to the actual post …
Is it possible to slice up the compute units in RDNA2? by that i mean single amdgpu can be shared accros multiple VMs right? (using SR-IOV). So is it possible to only allocate some amount of compute units instead of sharing the whole die?
The reason why I want you to try is becuase, if you can pull this off we can essentially creat a virutal steam deck that has zen2 cores from threadripper and sliced compute units of rdna2 gpus form 6000 series that matches the steam deck in terms of frequency and number of cores and compute units they have, after that just install arch based distro to get the closest software experience, then you can benchmark the shitout of it. This maybe a completely dumb idea but if it works, we can already start to do funky things with it.
The issue here would be the gpu vram you might have to mess with it to get working just right.
Well, Craft Computing on YouTube, has done an entire series about making a cloud gaming server for his family and friends. He’s currently running 12 separate windows vm’s on a single server all running at 1080p/60. Though he uses nVidia cards to do it, but I’m sure the AMD process shouldn’t be too far off.
SR-IOV isn’t supported on AMD consumer GPUs, last time I checked. If we indeed want to get close to estimating the Steam Deck’s performance, I’d imagine a VM with 8 downclocked Zen2 threads and a severely downclocked RX6600 would probably give a moderately decent idea about the Steam Deck’s chops.
To virtualize something the software has to support it, which is doubtful out of the box on the Steam deck when it arrives. If u lucky and read the docs, of the hardware and software used. Perhaps with some tinkering or hacking/tricking the deck into doing, displaying something would be much more likely.