R600g+SI Dota 2 Benchmarks With Mesa 11.2, Linux 4.5 Show Open Driver Progress

[humbug]

It's been well known that DOTA2 likes Open Source drivers.

The new game engine seems to have a really good openGL implementation. Most linux players in the Dota2 community reported an improved experience with the engine switch. Apart from the new Fiji results performance seems to scale very well across the different GPUs according to the capabilities of the hardware.

[Michael]

Btw I don't think AMDGPU supports SwapBuffersWait at all? vblank_mode works however.

PTS unconditionally always sets vblank_mode=0. I hadn't checked to see if SwapBuffersWait is used by AMDGPU but have always kept it set whenever testing since usually when I do Radeon tests, generally doing AMDGPU tests too.

[ernstp]

Btw I don't think AMDGPU supports SwapBuffersWait at all? vblank_mode works however.

[ernstp]

Why do you need a PCI ID for modesetting? That's a generic driver! You don't need to install the amdgpu ddx.

https://cgit.freedesktop.org/xorg/xserver/commit/?id=ac2f27f1a9fa8cd88c5dbe7ec0f96238eecf2c3e

[Kano]

Why do you need a PCI ID for modesetting? That's a generic driver! You don't need to install the amdgpu ddx.

[ernstp]

Michael: I made a build of xorg-modesetting for 15.10 that has Fury in it's list of PCI IDs, if you're interested: https://launchpad.net/~ernstp/+archi...es_filter=wily

[Michael]

I had this error when running the test:

LoadSystemDependencies():
AppSystemDict: Error in Connect() of interface 'GameUI011'!

Missing dependency?

Yeah it's a weird one... I hit it initially as well. Unrelated to PTS, others have ran into it too: https://github.com/ValveSoftware/Dota-2/issues/105

The fix from that thread did work: sudo ln -s /lib/x86_64-linux-gnu/libudev.so.1 /lib/x86_64-linux-gnu/libudev.so.0

[abu_shawarib]

I had this error when running the test:

LoadSystemDependencies():
AppSystemDict: Error in Connect() of interface 'GameUI011'!

Missing dependency?

[Michael]

hmm, Can't wait to hear more about this discovery!

At first it looked like it was a regression in DRI3 as during that second run, still had accidentally had my xorg.conf.d of DRI3 (enabled) +SwapBuffersWait (disabled) set to the radeon DDX rather than the amdgpu ddx as I was testing some other non-AMDGPU hardware in between there... So I thought that was it, but now in going back through and trying out different DDX options for AMDGPU, not yet able to figure out if it was just a fluke or what.

[smitty3268]

So I guess the 4.5 rc3 kernel fixed your Fury powerplay issues?