The original 8800 GTS launched in November of 2006 and in contrast to the new version it was actually a cut-down version of the more powerful 8800 GTX rather than the more powerful version of a lesser card as is the case with the new GTS. It featured 96 stream processors divided into six clusters of 16 (the 8800 GTX and Ultra have eight clusters) each of which was accompanied by four texture address units and eight texture filtering units, making for totals of 24 and 48 respectively.
Conversely, the new card has a total of 128 stream processors divided this time into eight clusters with eight texture address and eight texture filtering units for each cluster. Combined with the significantly higher 1,625MHz shader clock speed, this makes for more shader processing power than the 8800 Ultra even, and significantly more than the old GTS.
However, as you should always know, things aren't so simple in the world of graphics and a large proportion of a card's eventual performance is determined by the pixel fillrate and memory bandwidth, figures which are now largely decoupled from the shader performance. Indeed, it was these two figures that really let down the underachieving mid-range parts we saw from both nVidia and ATI during the middle of last year.
With its four ROP partitions, each able to output four pixels per clock cycle, and a core clock speed of 650MHz, the new GTS has a total pixel fillrate of 10.4Gpixels/sec. Compare this to the old GTS, with its five ROP partitions running at 500MHz and you get 10.0Gpixels/sec. So, not far off. The interesting bit though, is when you compare to the Ultra that has eight ROP partitions and runs at 612MHz. This combination makes for a whopping 14.7Gpixels/sec pixel fillrate and goes a long way to showing why that card is still outperforming everything else.
The memory configuration of the new GTS is also different, with it using 512MB GDDR3 RAM running at an effective speed of 1,940MHz. This is accessed across a 256-bit wide memory interface, which provides a total memory bandwidth of 62.08GB/sec. The wider memory interface (320-bit) of the old GTS and larger amount of memory (640MB) would normally provide a greater memory bandwidth but, again the higher clock speeds of the new version enable it to compensate for these deficits. Though, once again, the Ultra shows its prowess with its 768MB (GDDR3 2,160MHz) memory accessed across a 384-bit interface giving it a whopping 103.7GB/sec.
So essentially on the performance front, the new GTS is a refined replacement of previous cards that should variably sit somewhere between the old 8800 GTS and the 8800 GTX with the 8800 Ultra still likely to retain its top spot in the majority of situations. Certainly, at high resolutions and with anti-aliasing enabled, the 8800 Ultra's high memory bandwidth and pixel fillrate will keep it above the competition.
Aside from these technical performance differences, the 8800 GTS 512 also includes the new VP2 video processing engine that debuted with the 8600 range. This brings with it hardware acceleration of HD video playback, offsetting the task from your CPU, as well as support for the all important HDCP content protection that is needed to playback HD discs. Video post-processing also ensures video looks its best by applying a number of artefact reducing filtering techniques. Of course, ATI still has nVidia beat on this front with its more comprehensive video offloading techniques and built in audio pass-through capabilities that mean you can pass video and audio from your computer along an HDMI cable to your TV. With nVidia you need to run a cable from your sound card to the graphics board to enable this and even then it's dependent on the board partners adding in this capability.