At face value and in terms of appearance, the GeForce2 GTS is essentially a GeForce 256 DDR die-shrunk from .22 micron to .18 micron with the clock speed increased from 120MHz to 200MHz and the DDR SGRAM speed increased from 150MHz to 166MHz. Our board used the rather reliable and familiar 6ns SGRAM from Infineon (a division of SIEMENS).

Mounted in 564 PBGA packaging, the GeForce2 GTS sports 25 million transistors on a .18 micron, six-layer process. This can be compared to the GeForce 256's 22-23 million transistors on a .22 micron process. The smaller process gives several advantages. It reduces power usage or dissipation (halves it to 10W), which also reduces heat, thereby letting you set the chip to a higher clock speed.

Less power usage combined with more efficient board power usage also solves what may be the GeForce 256's largest difficulty, incompatibility with some motherboards due to high power demands. Also, the smaller transistors switch faster, which helps in reaching higher clock speeds. With the chip taking up less silicon space, more chips can be produced at once, which reduces per chip manufacturing costs.

The GeForce2 GTS uses a quad-pipeline architecture capable of drawing two textures per pixel per clock per pipeline, a theoretical fillrate of 1.6 Gigatexels/second. The T&L engine has been optimized and some marketing material now includes a "C" for clipping to keep up with ATI's latest feature announcement. The chip can now process up to 25 million polygons per second - though, to be a bit picky about terminology, there is a condition for this. The triangles must be attached to each other in strips or fans, thus requiring only one new vertice for each new triangle. At the other extreme, if a developer chose to separate each triangle, an admittedly unlikely prospect unless you are making "Attack of the Killer Triangles, part deux", each triangle would require three vertices thus dropping the triangle count to one-third of the 25 million per second.