X850XT PE appears slower than X1600XT
YOU'D EXPECT wouldn't you, that the Radeon X850 XT PE - equipped as it is with a full sixteen pipelines - would be much faster than the new eight-pipeline X1600 XT card.
We're not suggesting that the RV530-based X1600XT is a bad card. On a contrary it's a great card for the money, but it's certainly not designed to be a match for the R480-based Radeon X850 XT PE.
But 3Dmark06 seems confused. Since, in this test, the Radeon X1600 XT card will kick the Radeon X850 XT PE's performance butt.
A X1600XT card supports Shader model 3.0 while the latter cannot do HDR or Shader model 3.0 and this significantly affects the score. The card with Shader model 3.0 / HDR support will automatically gain a lot of points and end up a little bit faster than the usually much faster card.
While we can allow that Shader model 3.0 is becoming important, as at least a few titles are shipping with HDR enabled, it seems that the weighting in the tests is skewed. And besides, every single one of these titles has an option to use Shader model 2.0 instead. So, except some of fancy visual experience, an X850 card can run all these games just fine.
This leads us to conclusion that 3Dmark06 doesn’t indicate the real performance of the card. It gives us some bogus number that might be used only for Shader model 3.0 cards. It is a very relative thing rather than just based on pure logic.
We are still running 3Dmark06 on various cards and there is a good indication that the 6600 GT will look faster than an X850 card, which is just wrong. Even if it just comes close, it is wrong. Those cards appear faster as they can do the HDR test while an X850 won't run them.
As far as we know, it is possible to run all Shader Model 3.0 effects on Shader Model 2.0b hardware it just might be a little bit slower - sometimes it is even faster to use Shader model 2.0b instead of 3.0.
Still, Futuremark tends to ignore this, as SM 2.0b cards simply won't run the test at all. We will investigate this thoroughly and have many questions to ask. Some of our developer friends already indicated that there are different code paths for different types of hardware which is just wrong. We will sniff around.
In every single game, the X850 XT PE beats the heck out of X1600 or Geforce 6600 GT cards so 3Dmark06 won't give you a realistic performance number.
What can we say is that 3Dmark06 might make some sense for new latest and greatest cards. And since we have supported this Madonion, Futuremark gang for ages now, we intend to use the benchmark in the future, as it gives you one performance number that you should be able to rely on. At least, we'll do so for Shader Model 3.0 cards, but not for older ones.
