When I put multiple GPUs in, I see that the /usr/libexec/Xorg process is threading over multiple GPUs. It’s still the same process, I get multiple lines in nvtop, one per GPU, but they all have the same process id.
I’m not saying anything is wrong. Just looking for an explanation why that is better, intuitively I would have thought that threading a process over separate hardware would incur more overhead. But I don’t know much about the Xorg process, perhaps it does actually perform better if it threads over multiple GPUs. But then: in what way is that better / faster / etc.? Thanks!
In the standard config, all gpus are auto-added to the same xserver so they can be used simultaneously for different screens.
So each of my four screens has its own GPU then?
By screen, I’m talking of ‘Xscreen’, a virtual construct of the Xserver, not a physical display. A Xscreen can contain multiple monitors (crtcs).
I use only one screen (one Xscreen), then why would that screen use different GPUs all the time? In nvtop I can see that about every 2 seconds another GPU is used, while all the others are idle. 2 seconds later, another one is used, with all the others idle. Every 2 seconds or so the “active” GPU switches for that Xserver. Wouldn’t it be less work to always stay on the same GPU?