Of course, NVIDIA didn't throw out the old GeForce 256 design when they built the GeForce2 GTS. They included all the features of the old design. AGP 4x with fast writes, cube environment mapping, vertex blending, per-pixel mip-mapping and more. They're all there, and here is a breakdown of the important ones.

When the GeForce 256 was released, no motherboards were shipping with AGP4X support. Since then, both Intel's i820 and some VIA chipsets have hit the street, both supporting AGP4X. NVIDIA has stated that the GeForce 256 and the GeForce2 GTS both support Fast Writes under AGP 4X. To understand what that means, here is a little primer on AGP4X on i820 motherboards.

In order for the CPU to get high-bandwidth information to an AGP card, the CPU first has to send information to the chipset through the 1.06GBps FSB. Second, the chipset sends the data to main memory through the 1.6GBps memory bus. Third, the chipset then accesses main memory to get at that data through the same 1.6GBps memory bus. Fourth, the chipset sends the data to the AGP card through the 1.06GBps AGP4X bus. This is obviously very inefficient.

With AGP4X Fast Writes, the CPU can send data to the chipset, which can then send the data straight to the AGP card. According to NVIDIA, AGP4X Fast Writes improve performance on certain calculations from the CPU to the AGP card, which include these four main areas:

• 2D operations
• Operations involving writing data to the frame buffer or sending data to the graphics chip
• Loading textures in D3D into local memory
• Writing push buffers to graphics local memory (should be where the greatest benefit of the benefit will be had).

This all sounds great on the theoretical level, however, the reality is a bit different. In our tests using AGP4X with a GeForce 256 on the i820 chipset, turning Fast Writes off (they were on by default) yielded up to a one-third increase in performance.