At a press event held today in Seattle, Valve's Gabe Newell outlined the benchmark features integrated in Half-Life 2 and provided performance numbers comparing how the game will perform on various graphics cards. Newell endorsed ATI's graphics cards, which perform significantly better in the game, according to the provided performance scores. He also announced that the standalone Half-Life 2 benchmark will be available to the press on September 30.
Newell emphasized that Half-Life 2 has been developed as a DirectX 9 game, and said that it has taken Valve a large amount of additional effort to let Nvidia's range of DX9 cards play the game with DX9 effects enabled. Valve has created a special "mixed mode" option for Nvidia cards that substantially increases framerates for the high-end GeForce 5900 Ultra (from a score of roughly 30fps in the Half-Life 2 benchmark to a score of just under 50fps). However, Nvidia's less expensive cards, the GeForce 5200 Ultra and 5600 Ultra both post scores of just 10 to 15 frames per second in the same test. The provided test numbers were from a system equipped with a Pentium 4 2.8GHz processor.
In comparison with the high-end Nvidia card, the ATI Radeon 9600 and 9800 Pro perform much better in the Half-Life 2 test's "full precision" mode. According to Newell's presentation, the GeForce 5900 Ultra scores 30fps in this test, whereas the Radeon 9600 Pro scores 48fps and the Radeon 9800 Pro scores just over 60fps.
The mainstream Nvidia cards post higher numbers when running in DirectX 8 mode, though it's yet unclear as to how this affects the game's visual quality. Scores for both the 5200 Ultra and 5600 Ultra are about 10fps higher in DX8 than DX9. This indicates that the more affordable cards in Nvidia's lineup have to be run in DirectX 8 mode to achieve playable performance of over 20 frames per second.
We'll post more details based on running our own tests with the Half-Life 2 benchmark soon, which should shed more light on how the cards perform at different resolutions and settings.