Total War: WARHAMMER - AMD RadeonSI Linux Performance

[phoronix]

Phoronix: Total War: WARHAMMER - AMD RadeonSI Linux Performance

Shortly after Total War: WARHAMMER was released for Linux by Feral Interactive we had out NVIDIA Linux WARHAMMER benchmarks. Now having more time since that OpenGL Linux game port release on Tuesday, here are benchmarks when using the open-source RadeonSI Gallium3D driver stack with various AMD GCN graphics cards.

http://www.phoronix.com/vr.php?view=23759

[oooverclocker]

A quite decent performance of the AMD cards. They could still perform about 20-25% better when the game and the driver were optimal(assuming that the NV drivers are already optimal which might be wrong) but for OpenGL the performance in the last tests where more shaders may be used is really nice compared to what we have sometimes seen so far.
I'm looking forward to the first Vulkan games that can keep the GPU utilization higher. Before that happens I would not believe that the performance improves more than something like 5-10% just on the driver side.

[DanglingPointer]

Michael, would be great if you included the R9 290!

[Michael]

Michael, would be great if you included the R9 290!

It's not included because with the open-source stack my card is still regressed (performing extremely poor). Obviously would have included it otherwise.

[theriddick]

If the 290 has issues then you can bet your arse the 390 also has regression issues. So that is TWO flagship cards that are essentially broken with the RadeonSI driver! allot of people still have 290/390 cards and are not going to waste money buying a rx480 at this time!!! so AMD you really should fix these regression issues ASAP because those two cards are VERY common.

[spstarr]

R9 290 and R( 390 are GCN 1.1 Hawaii cards and Amd has currently no interest to enable powerplay, see:

https://cgit.freedesktop.org/~agd5f/linux/tree/drivers/gpu/drm/amd/powerplay/hwmgr/hwmgr.c?h=drm-next-4.10-wip

Except you are wrong, GCN 1.1 has PowerPlay but it's not called PowerPlay since its older power management, that's my understanding.

[ua=42]

If the 290 has issues then you can bet your arse the 390 also has regression issues. So that is TWO flagship cards that are essentially broken with the RadeonSI driver! allot of people still have 290/390 cards and are not going to waste money buying a rx480 at this time!!! so AMD you really should fix these regression issues ASAP because those two cards are VERY common.

You did notice he was talking about and using git? As in the unstable, things break all the time, codebase. Stable doesn't have that problem, which is what everyone should be using unless you are a graphics driver tester.

[smitty3268]

If the 290 has issues then you can bet your arse the 390 also has regression issues. So that is TWO flagship cards that are essentially broken with the RadeonSI driver! allot of people still have 290/390 cards and are not going to waste money buying a rx480 at this time!!! so AMD you really should fix these regression issues ASAP because those two cards are VERY common.

Nobody else seems to be having any problems. Michael is the only one reporting any issues, and he hasn't provided any details that would help the devs fix it.

[bridgman]

AFAIK the amd/powerplay folder does not include CI support because the chunk of DPM code called "powerplay" (which AFAIK is organized more like the code we use in other OSes) was developed after we had already CI support including DPM.

If you reword your sentence as "AMD has currently no interest in refactoring CI DPM support just so they can call it powerplay" then that's probably true.

[bridgman]

Nobody else seems to be having any problems. Michael is the only one reporting any issues, and he hasn't provided any details that would help the devs fix it.

I'll check again tomorrow but last time I asked we had not been able to reproduce the regression in house yet.