Re: Laptop Working Group Github Issue Request - Multi-GPU Support

The thing is: the AMD version of Optimus works, at least in the AMD +
AMD configuration. The presence of DRI_PRIME=1 or not will switch which
GPU is being used on the Framework 16.  I can't comment on the Nvidia
side because I don't have an example to test on.

I think basically what we need to answer is: "Why is X11's
autoconfiguration failing in the presence of multi-gpu?".  If that can
be figured out and fixed then I wouldn't be surprised if that fixes
Wayland as well and then multi-gpu works as expected. Unfortunately
that VGA Arbiter log message and it wanting you to manually configure
the devices is all I have. I can't find where seatd logs (which I
assume is responsible for this on Wayland's side), and sddm's log
doesn't seem to provide any clues either.

Thanks,
Luke

On Sat, 2025-06-21 at 15:25 -0400, Charlie Li wrote:
> Luke Wolf wrote:
> > this would seem to indicate
> > https://www.reddit.com/r/freebsd/comments/1ii0da4/wayland_on_freebsd_142r_tldr/
> > that Intel + Nvidia encounters the same issue.
> > 
> > However both of these would indicate that in the case of Nvidia
> > cards
> > that Wayland actually works, but X11 still needs configuration,
> > unlike
> > the Framework 16's AMD + AMD situation where X11 needs
> > configuration
> > and Wayland doesn't work.
> > 
> > Given the log complaint about:
> > VGA arbiter: cannot open kernel arbiter, no multi-card support.
> > 
> > My assumption is that this is what we're missing in order for it to
> > work:
> > https://docs.kernel.org/gpu/vgaarbiter.html
> > https://www.x.org/wiki/VgaArbiter/
> > 
> For nvidia on laptops specifically, if it's an Optimus setup, the 
> arbiter is irrelevant. A "proper" Optimus setup has the nvidia card 
> completely headless, where some other mechanism copies the rendered 
> output to the internal/integrated GPU framebuffer which has physical 
> outputs.
> 
> -- 
> Charlie Li
> ...nope, still don't have an exit line.