Is it possible to run GPU passes on Linux?
Is it possible to run GPU passes on Linux?
I mainly play Star Trek Online and Company of Heroes 2. The first one performed well on my system, and the second has a native Linux version from Feral. I’d look for other games if these three weren’t available. For me, games aren’t the main reason for choosing specific hardware or systems, except perhaps Pokémon. I’m thinking about getting a Nintendo Switch just for playing Pokémon every day...
Fallout 4 and League of Legends have both been stable for me after playing months ago without problems. On the battlefield, the 3 isn't just due to browser issues—it performed well with solid single-player and decent speed. For Witcher 3, I heard some problems with Mesa drivers on Intel and Radeon cards, so updating drivers via Ubuntu repos is a good idea. dxvk is still considered alpha but worked fine for me; overall, it's the surrounding improvements that kept me from quitting. A slight FPS drop (around 10-20) is typical during API transitions in dxvk. Still, other AAA titles are running smoothly thanks to Proton support.
I saw a similar approach mentioned during my search after posting. It seems X.Org can't directly swap GPUs, but a workaround exists: the main X.Org instance doesn't recognize the GPU at all—it sees a framebuffer on a secondary GPU. A fast Looking Glass copies the rendered output from the VM into the active X.Org instance. If the VM shuts down, a separate running instance can use its own native framebuffer. This involves multiple layers of X.Org? I'm not sure, but a YouTube video explained it with some scripts and notes about latency issues.
He relies mainly on the intel iGPU while executing games in a separate fake server environment after the VM stops. The vfio driver is detached from the GPU and replaced with the nvidia one, then frames are transferred to the iGPU. This method aligns with how Bumblebee functions, though it lacks Vulkan support unless paired with nvidia-prime, which demands a full shutdown of X.org. It’s not ideal since the GPU has real video output and using Prime with Intel is only practical if you ignore the video connection. The performance issues stem from Look-glass, and the setup resembles a laptop-like setup with an optimus GPU. Input lag remains tied to this approach.