Serious Sam Fusion 2017 Is Working Out Well For RADV Vulkan

[oooverclocker]

OGL perf = Vulkan perf. CPU bottleneck in Mesa?

It is even the best RADV result I have ever seen. But it really looks CPU limited in pretty much all cases. Usually RADV was much slower than OpenGL in most cases. I'm excited how it evolves but I guess the heavy performance tweaking comes later when most features are implemented and it passes all conformance tests. Just like RadeonSI which seems to be slowly getting closer to it's performance boundaries.

[Michael]

BTW, the PTS system summary in the article lists the 7700k as a 8-core CPU. Whatever it uses to fetch the info from is as buggy as Windows.

Basically shown as a logical core count, I don't bother changing that string to "Thread" for legacy reasons.

[tomtomme]

News are out that low perf of radv is due to bug in powersave governor. Use performance and you are fine.

[indepe]

We are yet to see a single solid proof that Vulkan is faster than OpenGL.

There seems to be a lack of valid benchmarks. Dota 2 and RADV are not.

The danger is that this lack of "proof" might slow down Vulkan adoption and support in the Linux space.

[LeJimster]

I did try it on my AMD Bonaire Chip (R7 360) and if i switch to Vulkan the Game hangs up and a core dump is sent to Steam ... I try it on my other AMD GPU i have and write in ....

Same Software Installation only Hardware Difference (AMD Athlon X4 860K vs. AMD FX-8320E) and (Bonaire XTX R7 360 vs. Curacao Pro R9 270) one X4&R7 360 did not Work and FX&R9 270 did not...

Ok maybe GCN1 Works vs GCN2 not? Why ever ...

So you tried with the R7 360 and R9 270 and both failed?

I have an R9 270 and I just managed to get radv working for the first time recently. For the longest time I didn't think it was supported, but it works fine.

I'm on arch using lordheavy's mesa-git repo. Using the linux-amd-staging kernel (4.9.6) and vulkan-radeon-git.

[Mystro256]

There has also been a lot of effort put into performance for Mesa GL but much less for radv. It's probably fair to think about radv as being where radeonsi was a year or two ago w.r.t. performance actual vs potential.

Considering how young radv is, it's pretty impressive how far it's gotten.

[VikingGe]

We are yet to see a single solid proof that Vulkan is faster than OpenGL.

Doom? I mean, it's a Windows thing, but it uses Vulkan, and it's clearly faster than OpenGL when CPU-bound.

Last time Phoronix did a benchmark comparison between RADV and the PRO driver, PRO was miles ahead, so there's definetly room for improvement.

Considering how young radv is, it's pretty impressive how far it's gotten.

Given what little manpower RADV has, I think it also shows that the Vulkan itself is a rather well-designed API for the most part. I mean, it can run pretty much all Linux Vulkan games in existence with merely ~31k lines of code.

[indepe]

Doom? I mean, it's a Windows thing, but it uses Vulkan, and it's clearly faster than OpenGL when CPU-bound.

About a month ago, mitch074 posted this link: https://www.reddit.com/r/linux_gamin...s_and_windows/

It shows that with DOOM on Nvidia, in actual gameplay, WINE Vulkan is almost as fast as Windows Vulkan.

However WINE OpenGL is more than 10% slower than Windows OpenGL, which in turns is close to 10% slower than Windows Vulkan.

[pal666]

We are yet to see a single solid proof that Vulkan is faster than OpenGL.

we had long ago with mobile gpu demos. here radv is slow and maybe croteam's renderer could see vulkan improvements

[pal666]

OGL perf = Vulkan perf. CPU bottleneck in Mesa?

no, with cpu bottleneck you would have 480 perf = fury perf