AMD RadeonSI Driver 2D Performance Is Getting Better

[phoronix]

Phoronix: AMD RadeonSI Driver 2D Performance Is Getting Better

The latest Phoronix article covering AMD's latest graphics processors on Linux was earlier this article in pointing out the Radeon R9 270 is far from perfect. One of the big problems with any Radeon HD 7000 series or newer GPU is the poor 2D acceleration performance with the open-source Linux driver, but performance improvements are coming...

http://www.phoronix.com/vr.php?view=MTU2NzI

[bakgwailo]

Hopefully it gets up to be at least on par with the Radeon/r600 driver. I finally switched to it and it is a dream compared to Catalyst.

[przemoli]

Hopefully it gets up to be at least on par with the Radeon/r600 driver. I finally switched to it and it is a dream compared to Catalyst.

It will be better.

1) For radeonSI gen hw, radeon crew have early access to development process. That means (or should at least), access to ingeneering team, debuggers, hw emulators, firmware crew (from the beginning), etc.
2) radeonSI code started with its hw. Before r600g, there was one more driver for that hw gen.
3) Radeon team is bigger.
4) More and more game developers point to r600g as better driver, that ought to at least bring more users, but also (maybe) some involvement from game devs themselfs.



Hopefully AMD quickly greenlight Mantle for Mesa.
(Anyone seen public discussion on the topic of implementing Mantle on top of Gallium3D? Would it need much change?)

[zanny]

(Anyone seen public discussion on the topic of implementing Mantle on top of Gallium3D? Would it need much change?)

Gallium is designed to work with any API as a state tracker (such as how it has d3d9 and d3d10 state trackers that aren't well maintained), and Mantle (should be) no different. However, by having that two layer indirection (meeting common state tracker APIs on top of conforming to the Mantle standard) it probably wouldn't have the performance one would expect from it, because it is designed for the hardware.

[przemoli]

Gallium is designed to work with any API as a state tracker (such as how it has d3d9 and d3d10 state trackers that aren't well maintained), and Mantle (should be) no different. However, by having that two layer indirection (meeting common state tracker APIs on top of conforming to the Mantle standard) it probably wouldn't have the performance one would expect from it, because it is designed for the hardware.

For me Mantle is way below OpenGL or DX.

For one, its client code that is responsible for assembling command buffers, and other such stuff that generally would be executed by driver BELOW gallium3D.
And Mantle do not "track" state. State is given by app (and is assembled beforehand). That also may be counterintuitive to what Gallium does.

[darkbasic]

Hopefully it gets up to be at least on par with the Radeon/r600 driver. I finally switched to it and it is a dream compared to Catalyst.

Catalyst 2D acceleration really sucks and there are no workarounds (ex native instead of raster). If you're lucky to not stumble on some nasty bugs then radeonsi 2D acceleration is much better than fglrx, unfortunately there are still lots of unoptimized paths.

[A Laggy Grunt]

Maybe RadeonSI will be in a good working state in time for AMD's next GPU release?

[zanny]

Maybe RadeonSI will be in a good working state in time for AMD's next GPU release?

The next series is Pirates Islands, the 20nm shrink of SI. So they will still be using the same driver.

After that, though, who knows. AMD hasn't said anything about their 2016 line.

Also, if radeonSI has at least opengl 4 and 90% of Catalysts performance next year, I'll be buying whatever their highest end card is just to support development.

Right now I'm waiting to see how radeonSI does in the 3.14 / hopefully Mesa 11 release on Arch before I buy a gpu this year.

[GreatEmerald]

Also, if radeonSI has at least opengl 4 and 90% of Catalysts performance next year, I'll be buying whatever their highest end card is just to support development.

Well if it does, then there won't be much development left to support

[zanny]

Well if it does, then there won't be much development left to support

Whatever their next gen cards are, their chipsets, their cpus in the kernel, gles 4, opencl, and gl 4.1 -> 4.4 is a lot of stuff.