VGA Charts VII: AGP Update Summer 2005

Test Setup

We ran all tests at the most common TFT resolutions - 1024x768 and 1280x1024 pixels - at standard quality settings and with 4x full-scene anti-aliasing (FSAA) and 8x anisotropic filtering (AF). Additionally, we benchmarked the fastest models at 1600x1200 pixels. To keep the time and effort within reasonable limits, we decided not to test the slowest cards with 4x FSAA and 8x AF, since their performance would simply not have been sufficient for these quality enhancing features. Some of the benchmark runs would have taken too long to complete when crawling along at 25fps, and at that frame rate the games would have been unplayable on these cards anyway. In some of our tests, we also encountered reproducible crashes. In these cases, we entered a score of 0.0 fps and noted that there was a problem.

All cards used the default driver settings provided by the manufacturer ; the only exception is that V-Sync was disabled. If the games didn’t offer corresponding settings, 4xFSAA and 8xAF were forced through the driver menus.

In some games, NVIDIA’s GeForce FX 5xxx family of graphics cards render pixel shaders at reduced precision to make up for lower performance, which results in more or less noticeably reduced image quality. This is also the reason why a GeForce 5950 Ultra is surprisingly able to beat a GeForce 6800 Ultra in Half -Life 2 in lower resolutions. However, when the cards are forced to render the shaders with full precision, for example in the game "Chronicles of Riddick : Escape from Butcher Bay", the cards of the GeForce FX series show a pronounced performance penalty. This phenomenon can also be observed in 3DMark 2005. Covering the image quality of the individual cards tested here would go beyond the scope of our VGA Charts, so we have compiled a list of articles that deal with the topic of image quality of graphics cards at the end of this article.

The Testing Platform In Detail