Radeon Gallium3D Moved Closer To Performance Parity With AMD's Catalyst In 2014

[Dukenukemx]

Why is GPUTest such a huge difference? Strikes me as odd.

[duby229]

The bottom line I think is that Phoronix is a site that linux gamers gravitate to. Whether that was the intention or not doesnt matter, that's what happened. More and more than ever, gaming on linux is getting better.

As more time passes there will be more linux gamers that gravitate to this site. I think it's inevitable.

[Michael]

Why is GPUTest such a huge difference? Strikes me as odd.

GPUTest is very synthetic and just stressing a subset of operations.

[duby229]

Why is GPUTest such a huge difference? Strikes me as odd.

It's because it stresses the hardware, where as the games running at hundreds of frames per second can't. Those games aren't bottlenecked at the gpu. In those cases the bottleneck is either at the framebuffer or the cpu. Which is what you are seeing regardless of catalyst or OSS.

[smitty3268]

It's because it stresses the hardware, where as the games running at hundreds of frames per second can't. Those games aren't bottlenecked at the gpu. In those cases the bottleneck is either at the framebuffer or the cpu. Which is what you are seeing regardless of catalyst or OSS.

That's not true. It's testing various limitations in the driver (software), not hardware.

For example, the triangle test will only get fixed when DRI3/Present is supported, and allows the driver to bypass synchronous X calls which are currently bottlenecking it.

The Furmark test is more what you describe, with a set of simple shaders designed to max out the most hardware as possible. It was originally designed as something which could cause thermal problems in the hardware on windows, since drivers weren't really designed to throttle the hardware and it could use more than intended.

I'm not sure exactly what the other tests are bottlenecked on, but it's certainly something slow in the drivers causing a CPU limitation, not a GPU bottleneck.

[Rakot]

Phoronix: Radeon Gallium3D Moved Closer To Performance Parity With AMD's Catalyst In 2014

Last week I wrote about the incredible improvements to AMD's open-source Linux driver over the course of 2014 that showed many significant OpenGL performance improvements for the open-source driver on various Radeon GPUs... But how does the latest open-source code compare to the closed-source Catalyst driver? In this article are benchmarks from an even larger assortment of Radeon GPUs while testing the latest Radeon Gallium3D and Catalyst drivers at the end of 2014.

http://www.phoronix.com/vr.php?view=21284

Michael, thank you for the articles you wrote this year.

I have two questions. One is for you specifically and the other is for mesa developers.

First one is for mesa developers. Is there any piglit test dedicated to performance other than OpenGL compliance? In this case it could be much easier to optimiser drivers.

Second question is to Michael. Probably it is not a question but rather suggestion. You almost always concentrate on performance of the drivers and rarely on stability and spec compliance. What if you publish a couple of articles related to OpenGL compliance. I think running full piglit for a single GPU from a generation on every driver would be great. The bottom line is that even if fglrx shows 300+ fps and "officially" support OpenGL 4.4 it may fail in more number of piglit tests compared to Mesa.

[Michael]

The bottom line I think is that Phoronix is a site that linux gamers gravitate to. Whether that was the intention or not doesnt matter, that's what happened. More and more than ever, gaming on linux is getting better.

As more time passes there will be more linux gamers that gravitate to this site. I think it's inevitable.

And inevitably more games will suit the necessary requirements for testing.... Heck even if the recent tests like Civ Beyond Earth were tried on the end of 2013 drivers for these recent comparison articles, it probably wouldn't run. In the end, 2015 looks brighter for Linux users.

[Michael]

Second question is to Michael. Probably it is not a question but rather suggestion. You almost always concentrate on performance of the drivers and rarely on stability and spec compliance. What if you publish a couple of articles related to OpenGL compliance. I think running full piglit for a single GPU from a generation on every driver would be great. The bottom line is that even if fglrx shows 300+ fps and "officially" support OpenGL 4.4 it may fail in more number of piglit tests compared to Mesa.

The problem with Piglit is that (at least traditionally) when things fail they fail hard.... Lockups, etc. Harder to automate for recovery, etc. And generally probably something less interesting to normal users.

[duby229]

And inevitably more games will suit the necessary requirements for testing.... Heck even if the recent tests like Civ Beyond Earth were tried on the end of 2013 drivers for these recent comparison articles, it probably wouldn't run. In the end, 2015 looks brighter for Linux users.

It's gonna be a great day. I can't wait for it. The more I think about it, the more excited I get.

[duby229]

That's not true. It's testing various limitations in the driver (software), not hardware.

For example, the triangle test will only get fixed when DRI3/Present is supported, and allows the driver to bypass synchronous X calls which are currently bottlenecking it.

The Furmark test is more what you describe, with a set of simple shaders designed to max out the most hardware as possible. It was originally designed as something which could cause thermal problems in the hardware on windows, since drivers weren't really designed to throttle the hardware and it could use more than intended.

I'm not sure exactly what the other tests are bottlenecked on, but it's certainly something slow in the drivers causing a CPU limitation, not a GPU bottleneck.

Thank you. I guess I should insert foot into mouth now...