Intel Clarkdale Linux Graphics Performance

[phoronix]

Phoronix: Intel Clarkdale Linux Graphics Performance

Last week we delivered our first Linux benchmarks of Intel's Core i3 Clarkdale processor with a variety of computational tests through the Phoronix Test Suite. While the Core i3 packs a nice performance punch, that is not all it has to offer. Also found on the Clarkdale (and mobile Arrandale) processors is an integrated 45nm graphics processor that is supposed to offer a decent level of performance in comparison to earlier Intel IGPs normally found on the motherboard's Chipset. In this article are these first Intel benchmarks for the Clarkdale graphics processor as we see how its open-source Intel driver stack compares to that of AMD with their open-source Radeon stack up through the Radeon R700 series.

http://www.phoronix.com/vr.php?view=14560

[Wyatt]

"...was magnitudes faster..."
There was only one order of magnitude there.

Also, I take some amount of issue with the assumption that after a year and a half, the performance of the X4500 hasn't changed relative to the others. I have an X4500 and have observed it getting better over time.

[bridgman]

Nice article. Curious about what happened with OpenArena and Warsow - I haven't run Warsow on my system but I know OA was running quite fast last time I tried it.

One minor point about performance expectations. The older cards used were a bit more "high end" than the newer ones, and for these types of tests (pixel-pushing vs shader intensive) the relative performance of the cards wasn't that surprising.

The closest "simple model" for performance on these tests would be the width of the render backend (aka ROPs, aka DB/CB, aka...) times the engine clock. For the four cards under test :

X800XL - 16 pixel/clock, 400 MHz -> 4800

X1950PRO - 12 pixel/clock, 581 MHz -> 6972

HD2600 - 4 pixel/clock, 600 MHz -> 2400

HD4670 - 8 pixel/clock, 750 MHz -> 6000

This is a pretty simplistic model, of course - the newer render backend blocks are faster for things like Z-only passes - and ignores shader power completely, but I think this is appropriate for the tests in question.

If you model shader power instead, the numbers would be more like :

X800XL - 35,200
X1950PRO - 97,608
HD2600 - 72,000
HD4670 - 240,000

The performance increase in newer chips mostly comes from higher clock rates and the ability to perform more shader processing on each pixel.

[Michael]

Curious about what happened with OpenArena and Warsow - I haven't run Warsow on my system but I know OA was running quite fast last time I tried it.

Hmm, odd. I never have been able to get OpenArena running fast with Radeon Mesa, at least with the map/settings used by PTS. This is going back to when the R500 3D support first came about I've had this problem and continue to have it... I know back then I talked with Airlie about it, but don't remember what was said.

[L33F3R]

Very interesting. I just popped in my 6200 and at 800/600 its about par with the x800xl. as soon as i scale to 1080 fps drops to about 50 as it should under this card. Mind you my card is crusty, old and passively cooled however i am very very impressed with the open source results .

I do find those OA results weird. I would expect that all ioquake3 games would run more-less the same.

[Apopas]

The only thing that keeps me from buying a X1950 Pro card is that it doesn't work with Kwin atm

Btw the JXRenderMark test doing transformed blit linear, what affects in real life?

[Zhick]

That's weird, I also never had a problem with neither OpenArena nor War$ow on my X1900XT when using mesa (except for OpenArena being terribly slow in the menu when "Bloom" was enabled). Even with r300g OpenArena is already running just fine.

[ausmusj]

Any plans on benchmarking the higher-GPU-clock i5-661?

[mendieta]

These tests are great. But I think the real interesting comparison is the one actually mentioned in the article: to compare this with previous Intel IGP's in Linux (GMA 950, etc). Would that be possible, at all?

Thanks!

[Jonno]

But I think the real interesting comparison is ... to compare this with previous Intel IGP's in Linux (GMA 950, etc). Would that be possible, at all?

Not in a fair comparison where all other componens are the same, as there is no motherboard with an integrated Intel GPU that is compatible with a Clarkdale CPU/GPU combo.

You would have to use different motherbosrds and CPUs for the different GPUs tested. However, if you pick "equivlent" motherboards and CPUs (same number of cores and clock frequency mainly) the comparason would still be interesting imho.