LOSTCIRCUITS

3DMark'06

With all the GPU's specs being as confusing as they are made by ATI's marketing, the very first thing we were naturally interested in was the theoretical performance in some high-level benchmarks, in particular, the fill rate and shader performance. For comparison purposes, we are using the X800 Pro with the same number of pixel processing units (on paper). The X800 Pro accesses the memory through a 256-bit memory interface, which is partly offset in comparison to the X1600 Pro by the lack of the ring-bus .. on paper.The X800 Pro is actually running at a lower clock speed (475 vs. 500 MHz core) but uses slightly faster memory (2.0 ns at 900 MHz data rate). Overall, it is not exactly an apples and apples comparison but the specs are close enough to warrant a direct face-off. Keep in mind that the X800 Pro does not have support for SM3.0, therefore, the HDR / SM3.0 benchmark suite of 3DMark'06 will not even run.

   

Sapphire X1600 Pro (left) vs. ATI X800 Pro (right). Note also that the X1600 Pro uses DDR2 whereas the older X800 Pro uses DDR3 as memory components.

The X1600 Pro is in green, the X800 Pro is in rose. Higher is better

Green: X1600 Pro; rose: 800 Pro. Higher is better

In theory, the X1600 Pro with its 4 four texturing units should be able to deliver roughly 2000 Mpixel / sec and it does so in a multi-texturing task. In the case of single textures, however, it misses the mark quite dramatically. This is where the 3 to 1 funnel-architecture pursued by ATI with the X1000 series shows its clear weakness, the old 800 is more than twice as powerful in single texturing and almost 3 times as potent when it comes to multitexturing.

Next Page:    => 3DMark'06 - More shaders =>