Well, now that we've seen what the new 3DMark consists of, courtesy of our 'Introduction of 3DMark03' article, that details the technology behind the various tests, lets take a look at the performance of the tests under a variety of currently available boards.

Test Setup

The graphics boards used in this test are as follows:

ATI Radeon 9700 PRO ATI Radeon 9700 PRO 128MB Cat 3.1 (7.83) Review
ATI Radeon 9700 PowerColor Evil Commando 2 Gold 128MB Review
ATI Radeon 9500 PRO ATI Radeon 9500 PRO 128MB Review
ATI Radeon 9500 Sapphire ATLANTIS 9000 64MB Review
ATI Radeon 9000 ATI Radeon 9000 64MB Review
Matrox Parhelia Matrox (Retail) Parhelia 128MB 1.03.00.043 Review
NVIDIA GeForce4 Ti 4600 Visiontek VisionTek XTASY 128MB 41.09 Review
NVIDIA GeForce4 Ti 4200 Gainward Ultra 650TV Golden Sample 64MB Review

Although there appears to be an overabundance of ATI parts here, the various 9500 and 9700 parts will be interesting to look at because they are all ostensibly the same chip but with different memory bandwidth and pipeline numbers. By analysing these we can begin to get a better understanding of where some of the bottlenecks may be in each of the tests.

Lets take a quick recap of the specifications of the parts used in these tests.

ATI Radeon 9700 PRO R300 325MHz 2600 pixels/s 2600 bilinear textures/s 19.2GB/s
ATI Radeon 9700 R300 275MHz 2200 pixels/s 2200 bilinear textures/s 17.2GB/s
ATI Radeon 9500 PRO R300 275MHz 2200 pixels/s 2200 bilinear textures/s 8.6GB/s
ATI Radeon 9500 R300 275MHz 1100 pixels/s 1100 bilinear textures/s 8.6GB/s
ATI Radeon 9000 RV250 275MHz 1100 pixels/s 1100 bilinear textures/s 8.8GB/s
Matrox Parhelia Parhelia 220MHz 880 pixels/s 3520 bilinear textures/s 17.6GB/s
NVIDIA GeForce4 Ti 4600 NV25 300MHz 1200 pixels/s 2400 bilinear textures/s 10.4GB/s
NVIDIA GeForce4 Ti 4200 NV25 270MHz 1080 pixels/s 2160 bilinear textures/s 9.1GB/s

Note, we are using the default clock speeds that Gainward ship at for their Ti 4200 Golden Sample board, which are a little above those of the standard Ti 4200 configuration.

Here is the configuration of the test PC and 3DMark03 software:

CPU Intel Pentium4 3.06GHz
Motherboard Intel D845GEBV2 845GE
RAM 512MB (1x512MB) PC2700 333DDR RAM
Network Integrated
OS WindowsXP Professional SP1
DirectX Runtime 9.0
 

3DMark03

Benchmark in default configuration except:
Trilinear Filtering Used
Anisotropic Filtering Set to 1x (None)

The new default settings for 3DMark03 have a setting of 'Optimal' for the texture filtering which may differ from one board to the next, so this was set to Trilinear filtering to ensure a consistent score throughout. Also the default Anisotropic filtering level is set to 4x which not all cards used in the test are capable of rendering at, so to ensure a fair comparison between all boards, Anisotropic filtering was disabled.