A Look At The RadeonSI/RADV Performance From Mesa 17.2 To Mesa 18.2

[phoronix]

Phoronix: A Look At The RadeonSI/RADV Performance From Mesa 17.2 To Mesa 18.2

As part of the Phoronix 14th birthday benchmarking this week, I have carried out a fresh comparison to see how the RadeonSI and RADV open-source AMD Linux graphics driver performance has evolved over the past year in Mesa. Tests were done using a Radeon RX 580 and testing every release branch from Mesa 17.2 to the current 18.2-devel.

http://www.phoronix.com/vr.php?view=26389

[schmidtbag]

Haha I don't recall seeing these vertical bars in years. I'm not against it, just merely pointing it out.

Despite there not really being any significant performance gains in the later versions, I don't see this as a bad thing. It suggests the drivers are already highly optimized, and updates have more to do with adding missing features. Optimizations could go on indefinitely, but there are a finite amount of features that can be added. So, as long as the "todo list" is shrinking after every update, we're in very good shape.

[TemplarGR]

Nice tests, although i have to ask, why not Xserver 1.20? You are using other unstable versions in your test, like LLVM 7.0 or the just-released kernel 4.17. In my experience with 1.20 on Arch it feels snappier in general. Perhaps it is my imagination, but it would be worth a try.

[Michael]

Haha we haven't seen these vertical bars in years. I'm not against it, just merely pointing it out.

Despite there not really being any significant performance gains in the later versions, I don't see this as a bad thing. It suggests the drivers are already highly optimized, and updates have more to do with adding missing features. Optimizations could go on indefinitely, but there are a finite amount of features that can be added. So, as long as the "todo list" is shrinking after every update, we're in very good shape.

PTS supports doing either/or and OpenBenchmarking.org you can check 'prefer vertical bars'. In this case it made the results easier to look at for seeing how the performance changed over the past year, and made more sense than line graph since only a few data points.

[Michael]

Nice tests, although i have to ask, why not Xserver 1.20? You are using other unstable versions in your test, like LLVM 7.0 or the just-released kernel 4.17. In my experience with 1.20 on Arch it feels snappier in general. Perhaps it is my imagination, but it would be worth a try.

Don't feel like rebuilding it every few days when I do a new install on the test systems... for the xserver plus all the dependent libs/DDX. Not aware offhand of any PPAs to make it faster, but didn't check much anyhow.

[VikingGe]

Not sure what should surprise me more, the fact that Mesa 17.2 is actually able to run Rise of the Tomb Raider which was released like 8 months later, or that performance has been very stable for the past couple of months overall (which is a good thing since it's mostly very competitive these days).

[tildearrow]

Typo:

a noticealbe

[davidbepo]

surprised and disappointed by the small gains

[tomtomme]

surprised and disappointed by the small gains

I guess on a Vega56 or 64 it would be another story... but also there the kernel version is important too

[smitty3268]

surprised and disappointed by the small gains

Considering it's faster than the corresponding NVidia card, I'm not exactly sure what you were expecting. Small gains here are actually good news, because it means it's been fast for quite a while instead of only recently getting there.