Depending on the card, one single benchmark run took a lot of time- up to 20 minutes per run. It was impossible to test all interesting configurations like FSAA and Anisotropic filtering combinations on all cards in the one and a half hours ATI provided for the test. We cooperated with Dave Baumann (Beyond3D), Brandon Bell (Firingsquad) and Bob Calayco (Gamespot) and used our shared time to run more tests. Since we also brought a GeForce FX 5600 Ultra (Flipchip) card with us, we are also able to show performance numbers with that card. ATI was only able to provide a FX 5600 (non-Ultra).
Due to the fact that we only had very limited time and we had to cooperate without any previous planning, we missed some scores and certain configurations were not run on one card. We're also unsure about the correctness of some numbers with Anisotropic Filtering enabled. Considering these factors and the hurry we were under, the following numbers should be read very carefully. We were not able to verify single scores in a second run to make sure that all settings were set correctly.
Halflife 2 - 1024x768-32
The table shows the result of the cards while using different codepaths of Halflife 2's "Source" 3D Engine. DX9 FP means "Full precision using Pixel Shader 2.0". DX9 HP stands for "Half Precision using Pixel Shader 2.0" and can only be run on NVIDIA cards. The third configuration we ran was the DirectX 8.1 codepath.
The ATI cards lead the pack. Most impressive is that the Radeon 9600 Pro beats the FX 5900 Ultra. The FX5600 needs to be run in DX 8.1 mode to get scores over 25 FPS. The level of the scores of each time demo is very different.