A Batch Of Graphics Cards On Gallium3D

[marek]

Can you explain ? You mean libtxc_dxtn cannot run on 64 bits ? Sorry if I misunderstood.

The problem is you need a 32-bit driver in order to accelerate those closed 32-bit apps, i.e. to get direct rendering. If you don't have it, indirect rendering is used, which doesn't have all the features Mesa has - it's stuck at OpenGL 1.4 with a lot less extensions. Things might work but the performance will suck to say the least.

[Michael]

Michael, can you clarify what Phoronix's position on that is? Are you afraid of getting sued if you publish an article using ST3C? I seem to remember some a long time ago, but perhaps not.

Because it's not "out of the box" configuration... While I'm sure there's a fair number of active Phoronix members that may install it, as far as overall Linux usage goes, how many people do you think will actually go forward and do it or even know about it? Not many at all.

[whitecat]

ta-spring works without s3tc... and without openGL3 effects.

Really ? how disable S3TC on ta-spring ? On my machine with r600g the textures are corrupted !

[whitecat]

Benchmarking those games without S3TC and direct rendering is pointless, which means you must have a 32-bit linux with libtxc_dxtn.

Can you explain ? You mean libtxc_dxtn cannot run on 64 bits ? Sorry if I misunderstood.

[smitty3268]

I meant ETQW, Doom3, and other closed commercial games.

Yeah, nobody except Q cares about ta-spring.

I forgot r600g still doesn't support ST3C, so you're right that it wouldn't be useful. Still, it would be nice to see that for an r300g test.

Michael, can you clarify what Phoronix's position on that is? Are you afraid of getting sued if you publish an article using ST3C? I seem to remember some a long time ago, but perhaps not.

[Michael]

Not to be picky, and I know that high-end cards can be extremely expensive and that you only get a few free samples here and there from manufacturers, but:

What's the possibility of also testing high-end cards in the latest 3 generations? For example, Radeon HD4850 or better, HD5850 or better, HD6950 or better. For the most part I have only seen low-end cards in your tests lately, some of them are even passively cooled because they've had so much of the bulk of the card stripped out.

Earlier today I actually tried doing a comparison of classic Mesa, Gallium3D, and Catalyst on a Radeon HD 4890, Radeon HD 5770, and Radeon HD 6870. (I finally got a Radeon HD 6870.) But the Linux 2.6.38 Radeon DRM as of this morning fails to light up the display for the HD 6870.

One other thing: Can you offer incentives to people who run phoronix-test-suite and successfully submit non-outlier data into openbenchmarking.org? For example, you could have a drawing to give away a free HD5850 to one lucky person, randomly selected from the list of people who submitted their results for a given test suite and had their results accepted into openbenchmarking.org. Having events like this occasionally would give you a nice burst of data flooding openbenchmarking.org, and having a large data set is key for people to have confidence in the reliability of the data.

We've already been working on a competition and hope to be giving away a complete Linux system from a Linux-friendly vendor as part of the OpenBenchmarking.org launch yet, but that sponsorship isn't finalized yet.

[marek]

ta-spring works without s3tc... and without openGL3 effects.

I meant ETQW, Doom3, and other closed commercial games.

[allquixotic]

Not to be picky, and I know that high-end cards can be extremely expensive and that you only get a few free samples here and there from manufacturers, but:

What's the possibility of also testing high-end cards in the latest 3 generations? For example, Radeon HD4850 or better, HD5850 or better, HD6950 or better. For the most part I have only seen low-end cards in your tests lately, some of them are even passively cooled because they've had so much of the bulk of the card stripped out.

I'll use OpenBenchmarking.org (once it's launched) to see how the high end compares, I guess, because that's where I mostly shop. Unless I'm in the market for a low-power mid-sized laptop with a (small) discrete GPU, which I'm currently not, I would never, ever, ever buy a low-end GPU. For my desktops I would at a minimum get the high-end single-GPU solution, but more likely I go for the dual GPU cards out of preference. I just buy them infrequently, and use them until I need to run a program that won't run without the next-gen features

One other thing: Can you offer incentives to people who run phoronix-test-suite and successfully submit non-outlier data into openbenchmarking.org? For example, you could have a drawing to give away a free HD5850 to one lucky person, randomly selected from the list of people who submitted their results for a given test suite and had their results accepted into openbenchmarking.org. Having events like this occasionally would give you a nice burst of data flooding openbenchmarking.org, and having a large data set is key for people to have confidence in the reliability of the data.

[agd5f]

r600g driver is not working truly right in performance section: hd3850 is usually performing faster than hd4830 and hd5750. hd4830 has just two times the processing power than a hd3850. hd5750 have more than two times the processing power of a hd3850 and some more memory bandwidth and still performing worse.

There are still a number of features that have yet to be implemented in r600g:
- fast clears
- 2D tiling
- hyper-Z

[marek]

Benchmarking those games without S3TC and direct rendering is pointless, which means you must have a 32-bit linux with libtxc_dxtn.