rFactor was tested with a custom demo recorded on the Sardian Heights race track. The DX9 shader profile was selected for all tests. Worth noting is that the configuration tool lists rFactor’s application support for anti-aliasing as a super-sampling scheme.
rFactor returns the test scores back to the previous pattern of smaller performance deltas between the two boards, with the 1900 holding a 5-10% lead at various resolutions.
High quality testing continues the pattern with the X1900 again maintaining a fairly consistent 10% performance margin over the X1800 XT.
Higher Image Quality Testing
The X1900 continues ATI’s pattern of supporting 2x, 4x, and 6x multi-sampling anti-aliasing. It is unfortunate, however, that the company decided not to support higher samples of its standard anti-aliasing with this new generation; the 6x option has been available since the 9700 Pro’s launch in the fall of 2002. Not increasing the number of samples was most likely based on performance concerns, though, since the current ROP units are capable of handling 2x multi-sampling AA in a single clock cycle, whereas 4x and 6x require two and three cycles, respectively. Running 8x AA with the current ROP engine would therefore consume four clock cycles, cutting deeply into the raster performance of even the X1900 XTX; yet for older games, particularly ones that by design have always been CPU-dependent, having the option available for higher amounts of AA samples would be a great feature, though naturally at some point the law of diminishing returns would rear its head. Yet with the Radeon 1000s ATI did introduce its Adaptive AA (AAA), a mode that combines multi- and super-sampling anti-aliasing intended for dealing with the aliasing produced by transparent textures.
The beginning of Call of Duty 2’s Prisoner of War map contains lots of transparent textures due to the grass, bushes, and trees scattered across this portion of the level. SimHQ’s custom demo thus represents a good test for ATI’s Adaptive AA support. AAA has two settings, a performance mode that applies super-sampling at half the rate of the requested AA, and a quality mode that matches the specified number of samples. Both test boards were tested at standard 4x AA and then again with the performance AAA mode to examine how the frame rate was impacted by the heavy amount of transparent textures scattered across this demo.
Click here for an enlarged image.
Both test boards see a 10-15% performance loss from enabling 4x multi-sampling; AAA, however, greatly increases the performance hit due to the fill rate and bandwidth costs of creating additional color data required by super-sampling. For the X1800 the frame rate impact ranges from 25-40% peformance loss, and due to its higher initial scores the X1900 actually sees a larger percentile frame rate hit. Again, the performance impact of Adaptive AA over regular multi-sampling will be largely dependent upon the amount of transparent textures in a scene.
The Prisoner of War map also has a fair amount of off-angle surfaces for testing the performance hit of ATI’s High Quality anisotropic filtering. 8x AF was tested in both standard and HQ modes on both boards to examine the performance impact the new filtering option would have in this test; bear in mind that the High Quality mode’s performance is dependent upon the number of off-angle surfaces being rendered in a given scene.
Click here for an enlarged image.
Enabling regular (translation: angle dependent) anisotropic filtering appears to cost slightly over 10% performance loss for both boards at all resolutions, though the 1900 saw a larger percentage drop due to its considerably higher starting frame rate. Going to High Quality mode in this particular demo really isn’t too costly, cost both test boards very little performance at the higher resolutions.