It may seem weird that somebody is complaining that something is working correctly, but I do want to know if there are others that see improvements in Intel’s GVT-G virtualized GPU technology with the newest NVIDIA drivers. It was noted in several community wiki’s on the interweb, that prime offload and gvt-g or hybid graphics and gvt-g setup is problematic and has issues.
I am happy that for some reason this now works correctly, but I am afraid that regressions may slip in in the future.
Basically, I have this setup: OMEN dc0019ur laptop (dedicated gtx1050 + intel uhd 630) and i use virt-manager (libvirt) to run a Windows 10 virtual machine. I run Manjaro Linux (testing branch), on X11 with prime offload (Plasma 5 desktop), for virtualization programs: qemu-git, libvirt-git.
My virt-manager entry for windows 10 has the following: mdev virtual intel iGPU, virtio disks, with UEFI, and other less important stuff attached.
Before, I could not suspend nor hibernate my laptop when: 1) intel gvt-g iGPU device was created in /sys/…/…; 2) while running virtual machine with gvt-g iGPU attached.
Also, the setup experienced frequent crashes/freezes - tyhe virtual machine would freeze up every hour or so, requiring “force stop” and starting it again.
Apparently none of the mentioned issues are affecting me now. Suspend, hibernate works. The VM has been running for 5 hours with various workloads, both on guest and on host OS.
This improvement appeared after I installed latest NVIDIA driver - 465.27.
I am not entirely sure that the newer NVIDIA driver changed that situation, but its highly suspected (at least for me). Is anyone using such a setup too? Could someone replicate the results with older/newer NVIDIA drivers, using gvt-g and with prime render offload?
I am delighted that everything suddenly works fine for me, but I do want this to last.
Sorry for the long rant and lack of technical details on my part. I will write other info in subsequent hours.