I've read that GPU passthrough on linux (ubuntu/mint) is possible with the correct types of hardware. I'm looking for a specific use case of passthrough and I'm wondering whether technology has advanced enough to allow for it to happen.
I have a linux mint host, and wanting a windows 8/10 guest. CPU/motherboard support vt-d (i7-5820k, asus x99-a). gpus are a pair of gtx970s. I want to: 1) Set up the guest so that it runs within a window on the host, thus allowing me to use something like a unity mode 2) Pass 1 of the GPUs through to the guest 3) When I shut down the guest VM, I want the passed through GPU to return to the host so I can use the pair of GPUs for compute/cuda heavy tasks
There are times where I'd like to game (hence the passthrough), but when I'm actually doing work I often need access to the cuda cores on both GPUs. A lot of the old threads I've read about this suggest that 1 card completely disappears from the host, is there a way to bring it back into action without a reboot?
Normally you'd need 2 monitors for this type of thing, plugging each into a separate GPU. But is it possible to use the second GPU to render a windowed VM within the host, instead of to a 2nd monitor?
Regarding windowed mode, I did see this on the virtualbox site, but I'm not sure if the VM is still windowed in this case: https://www.virtualbox.org/manual/ch09.html#pcipassthrough
I've searched for this and have come up short, but having said that, most of the search results are quite a few years old so it doesn't speak to any advancements in technology since then. The only thing I've found is a video on youtube that suggests it might be possible as it looks like a passed through GPU on a VM running in windowed mode: https://www.youtube.com/watch?v=XY1zDgCxARw