The approach ATI has taken to address the bandwidth issue is more than just incorporating high speed DDR memory. As we’ve seen on the GeForce2 GTS, throwing fast RAM at the problem is a very temporary, very costly solution. In addition to the 183MHz memory, ATI’s Radeon features a technology called HyperZ, a three-pronged strategy to eliminate much of the unnecessary data that has traditionally “clogged” the memory bus.

After an application draws a 3D scene, the depth buffer (or Z-buffer) has to be cleared so that the next frame can be drawn. Traditionally, the processor would write 0’s straight across the buffer, consuming just as much bandwidth as an actual depth value would. To combat this expensive waste, ATI developed a method of clearing the Z-buffer without writing values to each of the memory locations called Fast Z Clear. We can do the math to get a better idea of just how much memory is being saved using the first component of HyperZ.

Let’s say we’re running Quake III at 1600x1200 in 32-bit color. Using the result from our Quake III MAX quality test, 35.6 frames per second, we derive the following equation:

(1600x1200x(32/8)x35.6)=273MB/s

In addition, since the processor already knows the value written to each memory location, there is no need to execute the first fetch for the Z comparison. This effectively doubles the result of our equation, meaning the Radeon saves 547MB/s of Z data from being transferred.