For comparison, we've brought out the TNT2 troops. Along with the 3D Prophet, we have the Falcon Northwest/Guillemot Xentor TNT2 Ultra SE, which is a Guillemot TNT2 Ultra default overclocked at 195/235MHz. But for the main comparison (as very few of us actually own one of these Falcon hotrods), we've included Guillemot's own TNT2 Ultra offering, the Maxi Gamer Xentor 32, default clocked at 175/183MHz. For another perspective, we've also included the Voodoo3 3500 though be warned, its performance using the OpenGL full ICD as recommended by 3dfx is not what you are accustomed to seeing from the Voodoo3 series.
To get some depth, we ran all benchmarks with each board on the Athlon 650MHz system, the Intel Pentium III 600MHz and a Celeron 300A overclocked to 450MHz.
The first noticeable trend is how far ahead in the Quake 2 Timedemo1 scores the 3D Prophet actually sits. At 800 x 600 in Timedemo1, the 3D Prophet is over 52 frames per second (fps) faster than the TNT2 Ultra on the Athlon. At 1024 x 768 on the Celeron, it is over 44 fps faster than the TNT2 Ultra. Even at 1600 x 1200, the 3D Prophet is about 20 fps faster then the TNT2 Ultra across all three CPUs. All told, the 3D Prophet scores for Quake 2 Timedemo1 are from 23% (Athlon 640x480) to 65% (Celeron 1024 x 768) faster than the same scores on the TNT2 Ultra.
To get an idea of how the 3D Prophet performance compares to that of the TNT2 Ultra we've calculated the difference between scores for each of these cards and then represented that increase as a percentage of the TNT2 Ultra score (graph below). It is possible to gather from these graphs what percentage of increase in performance can be expected at different resolutions and on different CPUs.
So, for example, if you are currently running a TNT2 Ultra on a Celeron 300A overcranked to 450MHz, you can expect a 65% increase in speed running Timedemo1 at 1024 x 768 (graph below, in RED), taking your frame count score from 68.7 fps to 113.5 fps (graph above, in RED). While not too many of us actually play Timedemo1 (an appalling lack on interactivity, really) it has become such a classic benchmark that all the guys writing card drivers target it.
The most interesting thing about this graph is that (with the acception of the highest resolution on the lowest CPU), for a 'previous generation' OpenGL game (Quake 2), the performance percentage gain over TNT2 Ultra is increased as resolution is increased.
