Once again I have changed the format of the OcPrices video card benchmarking methods. I was highly impressed with the new [H]ard|OCP time-based analysis benchmarking using fraps, but to learn how do to do all the graphs was impossible within the time frame of this review. Instead, I am running all the games at what I consider to be the current optimal "gamers settings". The days of 1024x768 have been well and truly superceded, and with the migration to TFTs, 1600x1200 is being replaced by 1280x1024 with FSAA and Anistotropic filtering enabled as the settings for high end users. I think the majority of experts would much prefer an antialiased 1280 experience to a 1600x1200 picture with all IQ turned off. I have chosen a mixture of Open GL, DirectX 7 8 and 9 benchmarks, with the majority being the latter, simply because the modern cards being tested are being touted as DX9 cards.
For drivers I used the latest offical WHQL certified drivers for both nVIDIA and ATi cards. The Detonator 52.16s were pitted against the Catalyst 3.9 drivers. It's interesting to see how nVIDIA and ATi have come full turn with regards to their attitudes to drivers. ATi is now commited to releasing bug fixes and new drivers as often as possible, whereas nVIDIA, who used to sometimes have three new revisions in a week, are committed to providing much less regular - but very stable driver updates a la the now defunct 3dfx.
First up, let's get the older titles out of the way. Quake III with IQ enabled at 1280x1024 can still tax the graphics card enough to show scalable differences between, for example, a 9800 pro and a GeForce FX 5950. For that reason I still have no qualms about using it as an OpenGL test, especially considering it still a very popular online game.
With image quality settings enabled, the Radeon card is 10fps faster than the GeForce FX5950, with the 9800 Pro being a little quicker than the 5900. No noticeable difference during game play with any of these cards however, but to be honest if all you do is play Quake - even at these settings a 9700 Pro is overkill.
3DMark 2001SE is a useful DirectX 7 and 8 benchmark, and is considerably more useful than its now humiliated replacement. 2003 has lost all credibility in the eyes of many reviewers and readers alike, so we won't be testing components of it in this review.
With image quality settings enabled, the GeForce FX5950 is just a touch slower than the 9800 XT, although both nVIDIA cards can overtake the 9800 Pro for easy second and third places. Interestingly, with IQ disabled the ATi cards run amock, destroying the FX cards by around two thousand points - even the aging 9700 Pro was within snapping distance of the 5950s heels.
Another Open GL game, this time Return to Castle Wolfenstein. RTCW is based on a highly tweaked version of the Quake III engine with more detailed textures, many added special effects and other eye candy.
This time we see the raw power of the FX5950 able to overtake the 9800 Pro. The 9800 XT card, however, takes top slot once again, but not by a significant margin.
