A Batch Of Graphics Cards On Gallium3D

[phoronix]

Phoronix: A Batch Of Graphics Cards On Gallium3D

Yesterday the results for the Intel Core i5 2500K graphics on Linux were finally published after receiving a new motherboard and CPU from Intel that did not encounter the earlier Sandy Bridge problems. That article included results for several ATI Radeon graphics cards using both the proprietary Catalyst driver as well as AMD's open-source Gallium3D driver, there was also the testing done from a NVIDIA GPU under the reverse-engineered Nouveau driver that's also written against the Gallium3D architecture. In this article is an even larger round up of graphics cards being tested under open-source Gallium3D drivers. There are also results from the Gallium3D-based LLVMpipe driver.

http://www.phoronix.com/vr.php?view=15675

[whitecat]

more games

It would be great to have benchmark of other games (Doom3/Quake 4/ETQW/TASpring).

[whitecat]

In fact, I would know if r300g (with Radeon X1000) is able to run smootly ETQW on normal render (so with libtxc_dxtn *)

* on Fedora, configure RPM Fusion, then "yum install libtxc_dxtn" :-)

Also, benchmark the 32 vs 64 bits free games and drivers will be great :-)

[smitty3268]

It would be great to have benchmark of other games (Doom3/Quake 4/ETQW/TASpring).

Yes. [10 char limit is stupid]

[Wyatt]

Hey Michael, don't suppose you could include an HD2400 in your testing? I know it's pretty much the lowest-end of the R600 parts, but I think that's sort of useful for checking the lower bound.

It's also the lowest-power of the r600g cards and can even be passively cooled. I'm not a fan of annoying fan buzz, and GPUs (also G5 desktops) tend to be the worst about this in my experience.

[crazycheese]

I wonder how a card from 2005 (x1800) with 8:16:16:16 config core and only 8 GP/s in pixel and texture fillrate is able to outbeat with nearly 2x speed a card from 2009 with 640:32:16 config core and 9.2, 18.4 fillrates accordingly.
The difference in tests also varies to great degree.
Maybe there is a lot of unimplemented hardware features and 1800 is card which arch have got most of attention?..

[TeoLinuX]

I wonder how a card from 2005 (x1800) with 8:16:16:16 config core and only 8 GP/s in pixel and texture fillrate is able to outbeat with nearly 2x speed a card from 2009 with 640:32:16 config core and 9.2, 18.4 fillrates accordingly.
The difference in tests also varies to great degree.
Maybe there is a lot of unimplemented hardware features and 1800 is card which arch have got most of attention?..

Experts will explain better than I do, but long story short I guess x1800 runs on R300g driver which has been developed for longer and more optimized rather than R600g which powers the newer AMD cards. not to talk about nouveau, which is being developed by heroes (not supported by nVidia) and it's quite recent on 3D features

[darkbasic]

If someone is interested I recently benchmarked my HD3870.

97 fps in world of padman @1920x1200 (using phoronix-test-suite)
54 fps in nexuiz demo1 @1920x1200,normal quality
120 fps in openarena anholt @2560x1600, very high quality

Athlon64 3800+X2@2,6GHz, 2GB ram, HD3870 (RV670) with R600g, 2.6.38-rc4, xorg-server-1.9.4, mesa git, libdrm git, xf86-video-ati git, pageflipping on, colortiling on, swapbufferwaits off.

The real star here is llvmpipe

[Lynxeye]

I dont know how Michael manages to get so low framerates in OpenArena with nouveau.

Even my nv47 card is able to do a constant 90fps and I saw no performance regressions there in the last half year. And this 90fps top is likely caused by Xorg host taskswitches, not the card maxing out.

[pvtcupcakes]

I noticed that this time the r600 gpus actually had some speed differences even if they were minor. I remember an older benchmark where all the r600 cards got the same performance (slow) because they were limited by the CPU. It's nice to see that that is starting to go away.