GA-Source heeft een editorial gepost over de recente discussies met betrekking tot de T&L prestaties van de GeForce. Twee weken geleden bleek uit benchmarks van HardOCP dat de GeForce met een snelle processor en hardware Transforms & Lighting in 3DMark 2000 trager presteerde dan met software T&L. Of er harde conclusies uit deze bevindingen getrokken kunnen worden hangt uiteraard af van de vraag of 3DMark 2000 representatief is voor de werking van de gemiddelde 3D engine:
With a fast enough processor (over 700mhz), does the GeForce do T&L faster with its GPU turned off? All of the evidence we can find looks that way. However, just the same, with a slower processor and the GPU turned on, the T&L specific numbers hardly change at all even if the CPU is significantly slower. As the first generation of the GPU, and one existing before there is any significant amount of software that supports it, I am not surprised if its performance looks to be less than stellar. Something to remember is that S3's Virge chip was one of the first chips to support 3D acceleration, and in the first synthetic benchmarks it looked faster than anything. Only a few months later, when the software titles that actual game developers wrote appeared we all realized that the Virge was not the "screamer" it appeared to be. Does this mean the GeForce sucks? Absolutely not, but it may be a little early to buy it "for hardware T&L support".
By the same token, 3D Mark 2000 implementation of T&L testing can be viewed as flawed to an extent, as it is using a game engine that has not actually been used in any game release as of today. Its all well and good to say that it uses a game engine, but it would be nice to see a game based on that engine. Perhaps in the future they can do some licensing with id and Epic to license the Quake III and Unreal Tournament engines to run the same type tests through the various engines. Until that time comes, I think 3DMark is a pretty valid OVERALL benchmark, so long as you are using the overall 3DMark score. Trying to go deep into any one specific test out of the bunch doesn't necessarily show much of anything other than HOW WELL IT PERFORMS ON THAT TEST.