AMD Radeon HD 6000 Series Open-Source Driver Becomes More Competitive

[phoronix]

Phoronix: AMD Radeon HD 6000 Series Open-Source Driver Becomes More Competitive

The "R600" Gallium3D driver that provides open-source 3D/OpenGL Linux graphics support for AMD GPUs up through the Radeon HD 6000 graphics cards is becoming increasingly competitive. The open-source AMD stack isn't yet ready to overtake the proprietary AMD Catalyst driver in terms of raw performance or OpenGL compliance, but a lot of ground has been made up in recent months.

http://www.phoronix.com/vr.php?view=19048

[uid313]

What is the point of including benchmarks when the setting is set to low quality?
This benchmarks the CPU, not the GPU.

By the way, great that new Phoronix Test Suite have better colors, now the charts are actually readable.
Before I was often unable to distinct between the colors.

[Lemonzest]

Did this have the new shader backend enabled? (R600_DEBUG=sb)

[d2kx]

Did this have the new shader backend enabled? (R600_DEBUG=sb)

No, as stated in the article.

[GreatEmerald]

I assume that this is also with SwapBuffersWait. Would be nice if it was disabled in such benchmarks, but ah well.

Pretty nice results overall, 60-100% of the performance across the board (except for 6570 ? it would seem that DPM didn't work for it).

And yes, finally the good colour code is in

[xeekei]

I have a HD6950, and it saddens me to see that it's slower than the HD6870 in so many cases.

[duby229]

I noticed the 6950 as well. It's a VLIW-4 architecture where the 6870 is a VLIW-5 architecture. I believe the performance difference comes down to that.

[peapa]

VLIW5 and HD6000

Yes, to me this benchmark is more about the VLIW5 architecture than HD6000 cards. The title is a bit strange in this regard as all the cards benchmarked here use the same architecture as HD5000, VLIW5.
In the end, this article shows that the driver is mature for VLIW5, and that doesn't include the whole HD6000 family as HD6900 use VLIW4.

[krasnoglaz]

I don't understand why test target for drivers are decade old shaderless games or opensource relatively light games like Xonotic. Why not Team Fortress 2 and Dota 2?

[Michael]

I assume that this is also with SwapBuffersWait. Would be nice if it was disabled in such benchmarks, but ah well.

No, tests are always done with it disabled, as can be seen from the system logs.