The 16-bit scores recorded at the lower resolutions were not exactly 'smoking' but nevertheless very acceptable and would outpace a Voodoo3 or TNT2 Ultra under the same conditions. The GeForce 256 is, by our own estimation, faster at the lower resolutions, with smaller texture sizes in use during benchmark runs. Where the texture size is set at the maximum of 512x512, S3TC will come into play since Quake III supports S3TC. It even edges close to the performance level of a GeForce 256 DDR-SGRAM.
To give you an idea, at 1024x768x32 in Quake III, with 512x512 texture sizes, the GeForce 256 hits 54.7fps (on an Athlon 600MHz) which compares almost identically to the Savage 2000's 54.8fps. There will be certain instances where S3's architecture is being made full use of and it can maximize its potential. The benchmark comparison we just made is indeed such a case. But at 800 x 600 x 16, when the texture size is NOT maximized, the GeForce 256 DDR races in just over the 100fps mark with the Savage 2000 lagging behind at 82.9fps.
The benchmarks also demonstrated that S3's new Dual Pixel/Dual Texture Pipeline architecture was at work and behind some of the impressive scores. 32-bit vs. 16-bit performance was another strong point. Rarely was the drop in performance more than 10-15%, which just goes to show that 32-bit color is here to stay.
The 3DMark Max 99 scores show just why T&L is important. Without T&L support and a conservative graphics clock frequency it's no surprise that the results were not all that amazing. DirectX 6 games will not benefit from T&L and will most likely run in a similar fashion to being run on current generation video cards. On the flip side, it's good to see S3 improving upon their 16-bit to 32-bit performance within D3D. The hit, whilst apparent, is far from being significant.
