Menu
Catalyst 3.0 under the microscope: Are all good things really 3.0?

Catalyst 3.0 under the microscope: Are all good things really 3.0?

Synthetic benchmarks

3DMark2001 SE

First of all, the well-known 3DMark2001SE, which is impatiently waiting for its replacement, refreshed by the patch to version 3.30.

At this popular goal of benchmark optimization, we actually expect the greatest differences between the individual driver versions.

3DMark2001 SE v3.30
Unit: points
  • 1024x768x32:
    • Catalyst 2.5
      9.490
    • Catalyst 3.0
      9.436
    • Catalyst 2.4 (DX8)
      9.387
  • 1024x768x32 2xFSAA/2xAF:
    • Catalyst 3.0
      6.928
    • Catalyst 2.4 (DX8)
      6,866
    • Catalyst 2.5
      0
      Crash
  • 1024x768x32 4xFSAA/8xAF:
    • Catalyst 3.0
      4.852
    • Catalyst 2.4 (DX8)
      4.702
    • Catalyst 2.5
      0
      Crash
  • 1280x1024x32:
    • Catalyst 3.0
      7.262
    • Catalyst 2.5
      7.252
    • Catalyst 2.4 (DX8)
      7.158
  • 1280x1024x32 2xFSAA/2xAF:
    • Catalyst 3.0
      4.087
    • Catalyst 2.4 (DX8)
      4.036
    • Catalyst 2.5
      0
      Crash
  • 1280x1024x32 4xFSAA/8xAF:
    • Catalyst 3.0
      2.216
    • Catalyst 2.4 (DX8)
      1.990
    • Catalyst 2.5
      0
      Crash

For inexplicable reasons our Test system with the official Catalyst 2.5 not capable of the 3DMark2001 mi t to complete FSAA and AF. Each time at the same point in the Car Chase High Quality Test, the system was completely rebooted. Limiting the AGP setting, switching off fast writes or the like could not bring any improvement here, so that we unfortunately had to do without these values.

An increase in the number of points by just 150 stupid counters, and that only with the help of full screen anti-aliasing and anisotropic filtering, could be put within the scope of the measurement accuracy in view of the absolute number of points achieved.

Either our Radeon9500 was not taken into account in the optimizations that had taken place in the meantime, or there was simply nothing left to optimize in the driver that would have achieved a higher 3DMark score.

TempleMark

Even if the TempleMark is one of the rarely used benchmarks, it is available hereInteresting to watch. On the one hand, there is a small bug that runs through pretty much all Radeon cards and that prevents the lighting in the pool chamber from working, as well as an anomaly that causes the flames in the braziers on almost all cards known to us (including other manufacturers , even by PowerVR itself) cut off or appear strangely patterned as soon as anisotropic filtering is enforced.

Furthermore, due to the various effects that are used and the constancy of the results, statements about possible optimizations can be made very well in the area of ​​multitexturings or bump mappings.

TempleMark D3D v1.06
  • 1024x768x32:
    • Catalyst 3.0
      103.8
    • Catalyst 2.5
      103.5
    • Catalyst 2.4 (DX8)
      102.8
  • 10x7x3 2xFSAA/2xAF:
    • Catalyst 3.0
      78.5
    • Catalyst 2.5
      78.0
    • Catalyst 2.4 (DX8)
      77,2
  • 10x7x3 4xFSAA/8xAF:
    • Catalyst 3.0
      54,6
    • Catalyst 2.5
      53,8
    • Catalyst 2.4 (DX8)
      53.4

Interestingly, the advancing driver versions seem to continuously result in small but constantly observable improvements in contrast to 3DMark.

A rogue who now assumes that all driver work would first flow into the 3DMark. The exploitation of multitexturing and complex, but already standardbelonging texture effects, has obviously improved.

On the next page: Aquamark