A Big Comparison Of The AMD Catalyst, Mesa & Gallium3D Drive

[89c51]

How come the open source drivers perform so poorly in comparison to Catalyst? I mean, obviously they are in development still and so lower performance is expected. But is there a concrete reason/missing feature which causes them to be so far behind?

developers developers developers

[evolution]

Well, talking about the results, I'm satisfied that even classic Mesa / Gallium3D OS ATI drivers can run most standard OpenGL games on Linux , (although their frame rates are almost an order of magnitude slower than Catalyst). With some time, I hope (maybe Gallium3D) Mesa drivers can run at best, about 50% the performance of 3D proprietary Catalyst. But, until then, I'll stick with Catalyst (I need some fast 3D perfomance ATM).

And... there's also another point that wasn't tested here (and I'd liked to see there): Currently, if you use Wine, you're almost out of luck if you use any of the OSS ATI driver stacks. Some games need SM3.0 and that's only avaiable with OpenGL3 or higher... (currently, only Catalyst supports it. )

My 2c, cheers

[evolution]

The 11.1 drivers (coming out this month) are supposed to fix the tearing issue.

I hope they also fix the UVD1/UVD1+ H264/VC-1 problems in r600 generation cards (HD2xxx/HD3xxx) on both Windows / Linux OS's.

Cheers

[zouboulou]

the open source 3D drivers just don't cut it. the gap is huuuuuuuuuge.
one may wonder what is the difference between the drivers. That's not just "poorly" optimized drivers, some things must be plain missing (and probably kept secret by ATI)

[RealNC]

This article is just sad The drivers for any recent card suck and it's 2011 by now.

Maybe by 2018 we get an r600 that's somewhat decent?

[ChrisXY]

again no color tiling or pageflipping enabled...
using git but not using the most important performance improvements

And only with compiz enabled...

[BlackStar]

How come the open source drivers perform so poorly in comparison to Catalyst? I mean, obviously they are in development still and so lower performance is expected. But is there a concrete reason/missing feature which causes them to be so far behind?

Open-source drivers: <10 paid developers for all drivers (R300/R600/Nouveau/Intel/Via/...) Limited access to hardware specs (AMD, Intel) or reverse engineering (Nouveau, Via, ...) Development starts after the hardware is released. Stack designed for ease of maintainance and stability.

Close-source drivers: large, separate, *paid* teams for each company (AMD, Intel, Nvidia, ...). AMD, for instance, employees as many software developers as it does hardware designers (Brigdman could give a better estimate, but something like 50 developers sounds right? Remember, this is for a single company only.) Full access to hardware specifications. Development starts 6-12 months before the hardware is released. Stack designed for performance.

The amazing thing is that R300 has already surpassed the old Catalyst drivers in image quality, stability and 2d performance, with 3d performance improving daily.

[BlackStar]

And only with compiz enabled...

It wouldn't make sense otherwise. It's 2011.

[bridgman]

How come the open source drivers perform so poorly in comparison to Catalyst? I mean, obviously they are in development still and so lower performance is expected. But is there a concrete reason/missing feature which causes them to be so far behind?

If you look at the RadeonFeature matrix you'll see that "functionality" work was done first, while "performance" work (things like texture tiling, HyperZ along with general optimization) are just starting to happen now.

A couple of posts here suggest that performance options already implemented (eg colour tiling, page flipping) may not have been used in the tests, not sure though.

[crispy]

Generally the performance looks very decent considering this is open source and coded with relatively few resources. This gives me a lot of hope for future versions. I would consider 50% performance of the proprietary driver a huge success, especially since it will just work out of the box and solve for example the tearing issue. And I guess getting to ~20% as now is probably much much closer to 50% than it looks