AMD A10-7800 Kaveri: Radeon Gallium3D Improves, But Still Tough Catalyst Fight

[dungeon]

Sure, APUs do better with fast RAM, but he tested the difference between drivers here. The hardware is equal between test cases.

Yup, but looking at those numbers my Kabini with one 1600MHz 4GB module is faster with fglrx in Tesseract, faster in OpenArena with radeon and very nearly equal with fglrx on Kaveri, etc. and so on... So i can't explain why it is slower with Micheal's dual channel 2x2133Mhz that is 34.1 vs 12.8 bandwidth, so this Kaveri has 60% better bandwidth right from the start!!! then one on 1600MHz i use...

BTW radeon is faster for me in Xonotic low then flgrx by ~13%!!! But i knew that is because sound for some reason is not disabled properly in Xonotic benchmark with that 3.16 kernel where i run fglrx So yeah it is faster, but that is a joke

And this radeon "win" for example:

The open-source Tesseract game on the A10-7800 APU seems to do a bit better on the open-source AMD Gallium3D driver than with Catalyst fglrx 14.20.

Actually those are spilling issues, so radeon disable global illumination and render less

https://bugs.freedesktop.org/show_bug.cgi?id=79155

Again i have 25.76 fps with fglrx, again Kabini is "faster" then Kaveri - he, he, that can't be true if setup is right

[haplo602]

hmm looks interesting ... seems I can finally change my aging Radeon 5830 for a Kaveri chip (7800 looks the best) and save on money and power and don't drop in performance. Or wait for Carrizo ...

[Adarion]

But Carrizo will likely need more driver adaptions. It looks interesting but then you might need to have some patience.

On the Kabini vs. Kaveri: I don't have direct comparison, but my Kabini setup (Athl. 5350 using R3 HD 8400 which should be Sea Islands, DDR3-1600, SSD, Kernel 3.17.4, mesa 10.3.x and xf86-video-ati 7.5.0) still does have a few rendering issues in KDE (seems mainly related to a few context menus and transparency) and either Aquaria (which is not absolutely high end in terms of requirements) as well as Shadowrun Dragonfall (Dir.Cut) run, ehm, "rather slow". SR-DF even on low res. And it has somehow horrible loading times (unlike on my big box).
I'm sure the driver stack in my setup doesn't use it to its fullest extent but I would be sure that Kaveri must me faster in any case.

[dungeon]

hmm looks interesting ... seems I can finally change my aging Radeon 5830 for a Kaveri chip (7800 looks the best) and save on money and power and don't drop in performance. Or wait for Carrizo ...

I think you expect too much, those can't be compared gpu in Kaveri 7800 should perform let say something like or slighlty faster then HD6570/6670 or in fastest Kabini similar or slighlty faster then HD6450... and of course that compared with DDR3 equiped dedicated cards, not GDDR5.

Your 5830 is 175W device, that is double config, double buswidth, quadruple mem bandwidth... i guess you get an idea

[halfmanhalfamazing]

Woot! 80%+ of Catalyst performance in almost all tests!
Don't stop being awesome!
Now there is no reason except OGL4 (which hopefully comes in the next half or full year) to use the proprietary driver anymore, especially since that driver (on some distros, like Arch) is a PAIN to set up and keep working.

Crossfire.



And sadly, the Mesa driver desperately needs a modern GUI configuration tool. :-(

[bridgman]

And sadly, the Mesa driver desperately needs some form of GUI configuration tool. :-(

Do you mean something other than DriConf ? I thought the .drirc file was still being used by Mesa & drivers, although I haven't played with it for a long time...

http://dri.freedesktop.org/wiki/DriConf/

[halfmanhalfamazing]

Do you mean something other than DriConf?

DriConf is good in theory, but I meant something that's a little bit newer than 2006.

[haplo602]

I think you expect too much, those can't be compared gpu in Kaveri 7800 should perform let say something like or slighlty faster then HD6570/6670 or in fastest Kabini similar or slighlty faster then HD6450... and of course that compared with DDR3 equiped dedicated cards, not GDDR5.

Your 5830 is 175W device, that is double config, double buswidth, quadruple mem bandwidth... i guess you get an idea

and 2 generations older.... looking at various test and reviews, I am getting about the same framerate from my system as I am seeing in Kaveri benchmarks. also it's those 175W that bug me the most. Since I only play CS Source and GO in any amount of hours, I think the system would get me a sub 100W tops (all together) with the same GPU power and some more modern features (I have a dual opeteron 2389 system). all in all it looks like an improvement in almost any way I can think of (power consumption being the highest on the list).

[Martin76]

AMD A10-7850k , memory 8gb 1866mhz

Xonotic low 1920x1080

ubuntu 14.10 32 bit kernel 3.18 + oibaf ppa : 130 fps
ubuntu 14.04 64 bit kernel 3.18 + oibaf ppa : 140 fps
ubuntu 14.10 64 bit kernel 3.18 + oibaf ppa : 145 fps
ubuntu 14.04 64 bit kernel 3.13 fglrx 14.30.4 : 184 fps

Bioshock Wine (gallium nine needs oibaf nine and dri3 sarnex. dri3 could be disabled with LIBGL_DRI3_DISABLE)

ubuntu 14.10 32 bit kernel 3.18 + oibaf ppa : 32 fps
ubuntu 14.10 32 bit kernel 3.18 +
oibaf ppa + oibaf nine ppa + sarnex dri3 ppa : 53 fps

Xonotic with oibaf nine + dri3 : 80 fps(130 without nine and dri3)
with LIBGL_DRI3_DISABLE: 113 fps


x264

ubuntu 14.10 32 bit kernel 3.16 : 66 fps

ubuntu 14.04 64 bit kernel 3.13 : 91 fps
ubuntu 14.04 64 bit kernel 3.18 : 80 fps
regression?

[MasterCATZ]

Thanks for posting that

I have been wondering why I am 1/4 the FPS since upgrading to ubuntu 14.10 from kernels 3.13 - 3.16~13.7

will give 3.18 a go now and stop finding something wrong with ppaibaf/graphics-drivers