Half-Life 2 On Wine Is Faster On AMD R600g Over Catalyst

[marek]

So in a CPU limited test, the driver with less CPU intensive shader optimizations wins? Surprise.

Shader optimizations have nothing to do with this. Shaders are usually NOT compiled in the middle of rendering (if they were, there would be stuttering).

[marek]

I doubt that. r600g doesn't have a shader optimizer, how possibly it can beat Catalyst?

Not exactly true. r600g uses the common shader optimizations, which are part of the glsl compiler. Only the hardware-specific optimizations are missing.

Also, if the bottleneck is on the CPU, the speed of the GPU usually doesn't matter.

[rohcQaH]

Then again, it's somewhat understandable that people don't read the text, because it's usually a pageful of filler crap and self-promotion.

So in a CPU limited test, the driver with less CPU intensive shader optimizations wins? Surprise.

[AJenbo]

That would explain a lot of the comments on phoronix in general.

[smitty3268]

Which boggles my mind completely. Because I don't see howcould've been written any clearer.

Ha, yeah, I think what you're seeing here is a bunch of people who glanced through at the charts without reading the text beforehand.

The charts on their own were a little confusing if you didn't bother reading the explanation.

[AJenbo]

"As far as why the open-source driver is outperforming the highly optimized closed-source driver, Stefan wrote, "I don't have an explanation for this yet, especially considering that fglrx beats r600g by a factor of 5 in most of your Linux-native tests. A possible cause is that my test setup is heavily CPU limited (usually 800x600 resolution, no multisampling). This was a conscious decision when I set up the tests because Wine's main performance issues are on the CPU side, not the GPU side. However, at least r300g's main problems seem to be GPU-related (not sure about r600g).""

Unless he means, WINE has CPU issues when using AMD drivers, I disagree. WINE had no CPU bottlenecks in all games I played (NOLF2 and stuff).
Still, even with 800x600 this shows open driver has no resolution-associated CPU issues, which is good.
Running 1920 parallel would be good though, even if opensource looses.

I think he might be refereing to that Wine it self dosn't do a lot in GPU space so optimizations are focused on CPU and there for his test is also set to mainly stress CPU bound functions.

[oliver]

I'm surprised that blizzard games aren't in there. I do believe all 3 current blizzard games are the most played ones, if not the world, at least under Linux?

[crazycheese]

800x600 as res used for all tests? Most be a joke or?

"As far as why the open-source driver is outperforming the highly optimized closed-source driver, Stefan wrote, "I don't have an explanation for this yet, especially considering that fglrx beats r600g by a factor of 5 in most of your Linux-native tests. A possible cause is that my test setup is heavily CPU limited (usually 800x600 resolution, no multisampling). This was a conscious decision when I set up the tests because Wine's main performance issues are on the CPU side, not the GPU side. However, at least r300g's main problems seem to be GPU-related (not sure about r600g).""

Unless he means, WINE has CPU issues when using AMD drivers, I disagree. WINE had no CPU bottlenecks in all games I played (NOLF2 and stuff).
Still, even with 800x600 this shows open driver has no resolution-associated CPU issues, which is good.
Running 1920 parallel would be good though, even if opensource looses.

[Gusar]

So far there have been more comments on the terrible graphs than when they're supposed to be displaying

Which boggles my mind completely. Because I don't see how

What we are looking at is the result on the far right side, which is the fglrx/Catalyst performance while the rest of the results are from Wine on the Radeon Gallium3D open-source driver.

could've been written any clearer.

[_ONH_]

I'm sorry, but "clearly" is wrong wording. The labeling is very confusing and also: the last two result seemed to be Catalyst, according to the first page, not only the last one.

No, the two points rows in the first page are for the two lines with arb and glsl so just one point of the x axis is the catalyst driver, eg the last one.