RadeonSI Color Tiling Brings OpenGL Performance Boost

[phoronix]

Phoronix: RadeonSI Color Tiling Brings OpenGL Performance Boost

A few days back I delivered a fresh round of RadeonSI Gallium3D benchmarks showing the open-source AMD Radeon driver improving but still a long shot from the Catalyst driver. Shortly thereafter, it was announced the open-source driver would enable 2D color tiling by default. As earlier Phoronix benchmarks have shown, color tiling can be a big performance win so over the weekend I carried out some new performance tests looking at the impact of enabling color tiling for the RadeonSI driver that supports the HD 7000/8000 series and Rx 200 series graphics processors.

http://www.phoronix.com/vr.php?view=19303

[darkbasic]

HD7950 radeon+tiling vs catalyst: http://openbenchmarking.org/result/1...SO-1311018SO79

[darkbasic]

Michael why don't you upload logs?

[tomtomme]

xorg.conf

whats the needed entries for the xorg.conf to enable tiling? when will it be enabled by default?

[synaptix]

whats the needed entries for the xorg.conf to enable tiling? when will it be enabled by default?

I do believe it would be:

Option "ColorTiling" "on"

[Spittie]

Hi guys, long time lurker but first time poster.

Color Tiling is enabled for the HD 7790/R7 260X as well?
In general, what's the situation with bonaire's gpu? Sadly Michael doesn't seems to have one to deliver benchmarks.

I'll be in the market for a new gpu (kinda) soon, and the 260X seems like a great choice, if it has the same support from RadeonSI as the other cards not based on bonaire.

[marek]

whats the needed entries for the xorg.conf to enable tiling? when will it be enabled by default?

$ man radeon

[dimko]

really sad performance

Until AMD can offer better binary driver or sub par open source, i don't see how i shall become their customer. Don't get me wrong, i am not nvidia fanoy, my cpu is from AMD, but dam'... <50% performance? This is just wicked.
AFAIK quality of binary still sucks, would it still be the case?

[duby229]

Until AMD can offer better binary driver or sub par open source, i don't see how i shall become their customer. Don't get me wrong, i am not nvidia fanoy, my cpu is from AMD, but dam'... <50% performance? This is just wicked.
AFAIK quality of binary still sucks, would it still be the case?

The linux catalyst driver is about 85%-95% the performance of the windows driver. But as far as stability goes it's about 1%.

At least the OSS driver is stable. It's performance features are still being developed, but at least it won't treat you like a dummy.

[dffx]

Until AMD can offer better binary driver or sub par open source, i don't see how i shall become their customer. Don't get me wrong, i am not nvidia fanoy, my cpu is from AMD, but dam'... <50% performance? This is just wicked.
AFAIK quality of binary still sucks, would it still be the case?

This particular article is more about the open source drivers, and honestly, this is what you should be paying the most attention to. Right now, sure, Nvidia is top of the pile for performance, but because Nvidia relies on the closed driver Nvidia users are reliant on Nvidia to deliver -- this has panned out as an issue (though relatively small) in that the Nvidia driver currently doesn't support kernels higher than 3.11.

My prediction is that, by the end of next year, AMD will be the brand to go with for video cards in Linux. The open source driver is making such huge strides -- RadeonSI has taken more time than many would prefer to get to the point it's at right now, but a lot of these changes carry over to the new line of cards that were just announced, meaning that they will be seeing the performance increases much, much sooner than Southern Islands ever did.

Things like the color tiling, and when HyperZ and a couple other features finally gets implemented, are very quickly closing the gaps in performance, and it's very exciting to see.

I do believe it would be:

Option "ColorTiling" "on"

I believe it would be "True", not "on", as it's a boolean value. Though I'm not sure if Xorg parses on/off == true/false? Never thought about it until just now.