unigine heaven 3.0: Radeon gets slower and slower (mesa9.0 vs 9.1-devel)

[necro-lover]

Yes yes i try it again and upgraded from kubuntu 12.10 mesa9.0 to oibaf mesa9.1-devel and the result is: its slower.

Yes maybe the mesa9.0 is only faster because 9.1 calculates more openGL effects but in the end its complete useless on my hardware.

is there any improvement planet for this kind of hardware generation? hd4000?

or is it just another way to say: fuck you customer buy new hardware you dump ass ?

[Kano]

It should be clear that you just need to downgrade the xserver with a ppa to use fgrlx 12-6-legacy driver. If you can not live without latest xserver then get rid of the card, you should know that since 12-4 which was the last official driver (12-6 legacy was just a bonus).

[marek]

Yes, some things are slower, but other things are a lot faster. This is a known issue with TTM. There has been a Phoronix article about it.

[necro-lover]

Yes, some things are slower, but other things are a lot faster. This is a known issue with TTM. There has been a Phoronix article about it.

Hi Marek are you sure that the problem is the TTM (in-kernel memory management) instead of the old compiler compared to the new LLVM Compiler from hd5000-hd6970 series ?
What’s the status of back-porting the LLVM Compiler from hd5000-hd6970 series to hd2000-hd4000 series cards?
In my knowledge its the compiler for VLIW what makes the difference.
I have some euro/dollar/bitcoins for you if you fix the LLVM Compiler for the hd2000-4000 cards

I think other people also would honour this.

[chithanh]

in the end its complete useless

That I would consider a spot-on characterization of synthetic benchmarks.

[necro-lover]

That I would consider a spot-on characterization of synthetic benchmarks.

Good point but on the other side I also bought oil-rush with the same engine.
This means I can use oil rush with the same result then its not a synthetic benchmark.

Do you feel better if I do this ?

[agd5f]

Hi Marek are you sure that the problem is the TTM (in-kernel memory management) instead of the old compiler compared to the new LLVM Compiler from hd5000-hd6970 series ?

There are lots of things that could be done to improve performance. This particular change is due to limitations in ttm.

What’s the status of back-porting the LLVM Compiler from hd5000-hd6970 series to hd2000-hd4000 series cards?

The r600 llvm work covers all asics handled by r600g (hd2xxx-hd6xxx).

[necro-lover]

I found the cause of the bad result surprise surprise...

I installed windows7 on a harddriver to check this the i installed the catalyst 12.6 and run the unigine heaven 3.0 bechmark with directX and openGL..

And the result is really a surprise... because the opensource radeon driver is ~3times faster than the windows openGL catalyst driver.

I think the OpenGL renderer of unigine heaven 3.0 is broken by design.

test results:





so its complete bullshit bingo the radeon linux driver is already 3 times faster than the catalyst 12.6 in windows7

and the DirectX renderer of the unigine engine is ~10 times faster than there openGL renderer.

[marek]

If you revert the Mesa commit fa58644855e44830e0b91dc627703c236fa, you might be surprised...