Hullie wilde even kwijt dat je bij Deafening Silence een performance vergelijking kan vinden tussen een GeForce kaart met verschillende Detonator drivers (3.53, 3.62, 3.65 en 3.68). Het blijkt dat de 3.53 drivers het snelste zijn met zo'n 6% gemiddeld (gemeten onder 3DMark 2000). Dit verschil kan nu echter verklaard worden doordat dynamic lighting in die versie niet werkte.
It's surprising but the 3.53 drivers significantly outperformed the other detonator drivers (6% faster). After I completed all the tests, I retested the 3.53 drivers to make sure I didn't make a mistake. Nope, no mistakes...they are faster. A good portion of that increase may be attributed to the much higher CPU Marks. What did they do differently? All the drivers since then have performed about the same. Unfortunately, the Elsa drivers take another 6% performance hit. However, the 3.53 and Elsa drivers have much better 2D quality at the 1152x864 that I normally use. With the exception of the 3.53 drivers (which exhibited some tearing in Quake 3 demo), there were no game problems with any of the other drivers. Ultimately, every system is different and drivers may act differently on your system.
Update 1: It appears that the performance increase seen with the 3.53 drivers is because dynamic lighting was broken. When nVidia fixed dynamic lighting in the 3.6X drivers, there was a corresponding performance drop. However, in many games, dynamic lighting is crucial, looks great and therefore worth the hit.
Update 2: Ignoring the 3.53 drivers, with the tests that 3D Mark 2000 uses, there "appears" to be no performance improvements in each new driver release. I say "appears" because there may be things happening beneath the hood. I guess I'll have to test the drivers again and look at things like fill rate and polygon throughput.