On Sat, 22 Feb 2020 at 13:19, Patrick O'Callaghan <pocallaghan@gmail.com> wrote:
On Sat, 2020-02-22 at 06:40 -0600, Richard Shaw wrote:
> On Sat, Feb 22, 2020 at 6:02 AM Patrick O'Callaghan <pocallaghan@gmail.com>
> wrote:
>
> > On Fri, 2020-02-21 at 16:18 -0800, Samuel Sieb wrote:
> > > On 2/21/20 2:46 PM, Patrick O'Callaghan wrote:
> > > > That's a good point which I hadn't thought of. I actually only have a
> > > > single monitor connected via an HMDI switch to both video outputs. I've
> > > > been so used to this I forgot to mention it, but clearly I have to
> > > > figure out how to run my desktop off the Nvidia card (I don't mind
> > > > losing the IGP so multimonitor isn't important). Do you know if it's
> > > > possible to blacklist the IGP? That might be the simplest solution.
> > >
> > > You could try blacklisting the Intel one.  You would need to check which
> > > module it is, but probably "i915".  Or check in your BIOS, you might be
> > > able to set the other card as primary.
> >
> > Blacklisting the i915 didn't work, so I changed the BIOS settings and
> > that did it.
> >
> > Unfortunately, although some games work well (e.g. Witcher 3), others
> > don't (Assassin's Creed Odyssey - lots of stuttering), so I'll probably
> > back off from this for the moment, though a more high-end GPU might
> > have better results (mine's a lowly GTX-1050). It's interesting that
> > the GPU passthrough with QEMU/KVM works so well in comparison, given
> > that it's a whole virtual machine.
> >
>
> I don't thing the 1050 should be THAT bad... I did just watch a youtube
> video on gaming on linux and it specifically talked about some games need
> to compile shaders on first run. You could try leaving the game running for
> 10-30 minutes and see if it improves.

What I did was run the AC Odyssey in-game benchmark, which gave less
than half the framerate I get in the VM version. I'd guess that
probably reflects a worst-case scenario but a brief try at the game
itself convinced me it wasn't playable. Shaders would need to be
recompiled for each scene, which I don't think is practical.

Do you have a link to that video?

poc

Ah! So you just did a direct Windows only title in a KVM+GPUPassThru+Windows(?) vs  Linux+Steam+Proton

Now, it just so happens that there are comments on the protonDB suggesting that in this specific title there are slowdowns (performance loss) Windows vs Proton.
Could you try a few more games for a better baseline.  I would really like to know the results.

Thanks!