In the interview we conducted with ATI's engineering regarding R580's configuration of 16 texture units and 48 pixel shader units ATI stated that at the bandwidths available to X1900 presently, 16 texture units is optimal, and moving beyond this would have limited gains. Here we'll run some our Doom 3 and SC:CT test at 1600x1200, varying the core clock rates, then the memory clock rates to see how the performance drops in order to see how balanced the current configuration is.

Throughout the core clock speed changes we see that SC:CT has a larger performance deficit from dropping the core speed than Doom 3 does, which may not come as much surprise seeing as it utilises more complex shader code, which should be more limited by the computational processing capabilities of the core than bandwidth bound.

Next we'll run the same tests but by varying the memory clock speed. Note that the memory speed is altered at the same ratio's the core clock speed was in the previous test - because the memory clock is higher than the core clock to begin with, each step is larger than the core steps.

When varying the memory speed, oddly its Splinter Cell that takes the larger performance hit as well. Although in this case neither of them are having as large a performance hit as varying the core clock speed.

When looking at both of these sets of results in conjunction with each other it can probably be assess that card is less bandwidth bound in this configuration than it is limited by operations on the core, given that there is a larger performance hit in both tests for varying the core clock rate than there is for varying the memory bandwidth. However, overall there is a fairly close performance degradation in both cases, suggesting that adding more texture units may make some difference, but not necessarily that much.