I’ve been dual booting Linux and windows for about two years now, but in those two years, I have never booted into windows, except by mistake.
This made me think about removing windows and just saving that wasted space for Linux. I only ever dual booted for the off chance the peer pressure to play anti cheat games was too great, but so far it hasn’t.
For the off chance where I want to play a game that doesn’t run well on Linux, is it a good idea to do that via VM instead of dual boot, or is it too much hassle? Will there be performance hit or any issues with those games?
depending on how you virtualize your windows install, yes, there are ways to kind of “spoof” your machine details (like manufacturer, desktop instead of vm) but again, it won’t be fool-proof and it might take a lot of time to get working well. i definitely have not tested that many triple-a windows games with anti-cheat because that’s just not my gamer wheelhouse, but so far after spoofing some of my machine details (spoofed as a baremetal oem install) i haven’t had any issues.
What hypervisor are you using?
I use proxmox and run a couple windows VMs for Remote Desktop. I’ve passed through nvidia gpus and even at point had a nvidia grid setup running splitting up a P40 across multiple VMs.
The nvidia gpu’s require several config options to ‘spoof’ a real desktop and prevent the code 43 error but windows still identifies them as virtual machines. I’ve never found a way for trick windows itself into thinking it’s stand alone.
Correct me if I’m wrong, but I believe Nvidia stopped producing the stupid code 43 error on their (consumer) GPUs a while ago (version 465 of their drivers). The P40 is possibly different because it’s not a consumer chip, but for gaming GPUs it should work out of the box now.
I say should because Nvidias driver are hot piles of garbage so even if they intend to make it work in VMs now there’s a good chance it just doesn’t.
Its been a while since I’ve had any gpu’s attached to a windows VM but I think my time pre dates the change. I realized they were not actually doing anything for my given workloads so I sold most of them. The P40 is setup in a VM for tensorflow now and one of these days Ill get the time to go back to that.
Yeah, that makes a lot of sense. Nvidia intentionally broke consumer GPUs for so long that I’m pretty sure most people who know GPU passthrough exists still think you need to do the weird Nvidia workarounds!
I have recently been playing with VMs in Unraid and in the video tutorials I’ve seen they talk about about grouping together (by editing xml file[1]) the video with audio that comes the GPU to avoid that error. Also about passing a modified BIOS. Are those the workarounds been talked about here?
[1] multifunction=‘on’
When I passed through my 1080 to macOS I didn’t need to specify the multifunction flag. I did need to pass a BIOS, though. I also had to mess with the VM XML but I don’t remember exactly what I needed to do to get the 1080 to be recognised right.
All I know is that my tiny Arch hypervisor ran in 256MB of memory while macOS got most of the hardware. That was months before Apple and Nvidia stopped working together to provide drivers, though.
These are the workarounds I’ve heard of, but I don’t think multi-function is an Nvidia-specific fix (there are probably more flags like those, depending on the layout of your PCIe device tree, I had to pass through my USB 3 controller or the passthrough would fail).
for hypervisor i’m using libvirt+qemu. by doing this, a few lines in the .xml for the vm is all that is needed for me to enable some hyper-v feature flags to spoof to windows that it’s not a vm. check to see if proxmox has some hyper-v features you can enable for this purpose.