Benchmarks
Theoretical Rates
Before going on to look at any actual benchmarks scores we'll take a look at the theoretical metrics of the GammaChrome S18 Nitro along with the rest of the boards we are using for comparison.
| Core Clock (MHz) | Fill-rate (Mp/s) | Texture Fill-rate (Mt/s) | Triangle (Mtris/p) | Memory Clock (MHz) | Memory Bandwidth (GB/s) | |
| S18 Nitro | 450 | 1800 | 1800 | 450 | 450 | 14.4 |
| X600 PRO | 400 | 1600 | 1600 | 200 | 300 | 9.6 |
| 6600 | 300 | 1200 | 2400 | 225 | 250 | 8.0 |
| 6200 | 300 | 1200 | 1200 | 225 | 275 | 8.8 |
| S18 % Faster Than: | Core Clock | Fill-rate | Texture Fill-rate | Triangle Rate | Memory Clock | Memory Bandwidth |
| X600 PRO | 12.5% | 12.5% | 12.5% | 125.0% | 50.0% | 50.0% |
| 6600 | 50.0% | 50.0% | -25.0% | 100.0% | 80.0% | 80.0% |
| 6200 | 50.0% | 50.0% | 50.0% | 100.0% | 63.6% | 63.6% |
Ostensibly all the boards here have the same number of pixel pipelines (or ROPs - Render Output Units) so their pure pixel fill-rate difference correspond to their clock rate differences, which results in the S18 Nitro having a theoretical pure pixel fill-rate advantage over the others. The same follows for the texture fill-rate, except on the 6600 which has twice as many texture sample units, so on balance the S18 Nitro has a 25% disadvantage to the 6600 here. With the organisation of the pipelines on these boards their texture rate also corresponds to the number of shader pipelines each board has, so all have 4 pipelines with shader math capabilities, except for the 6600 which has 8 - the performance characteristics, as well as capabilities, of each shader pipeline will differ greatly from one architecture to another as well.
Each of the manufacturers are implementing differing numbers of vertex shaders for this class of boards as well, with the S18 having four, the X600 having two and both the GeForce 6's having three, so with these and the clock rate differences in mind it would appear that the S18 Nitro should have at least twice the vertex processing abilities of the other boards as well. However, as with Pixel Shaders, the composition of the math ALU's within each vertex shader can affect the performance so testing will reveal if the theory matches to the reality.
Finally, all the boards utilise a 128-bit memory interface so the bandwidths differ in-line with the memory clock rates, which again the S18 Nitro has the highest.
Fill-Rates
For the first test we'll take a look at some of the key fill-rate characteristics of the boards on test here:
| M Samples/Sec | S18 Nitro | X600 PRO | 6600 | 6200 |
| Color Fill | 1811.9 | 1205.4 | 1190.3 | 1197.9 |
| % Diff from Theoretical | 0.7% | -24.7% | -0.8% | -0.2% |
Looking at the pure pixel fill capabilities on all the boards we see that the S18 is actually mildly overachieving from its theoretical fill-rate, which may just be down to a slight wavering in clock rate above 450MHz.
| M Samples/Sec | S18 Nitro | X600 PRO | 6600 | 6200 |
| Single Texture Alpha Blend | 1228.1 | 911.0 | 833.0 | 906.0 |
| % Diff from Theoretical | -31.8% | -43.1% | -30.6% | -24.5% |
Looking at the Alpha blend rate we see that all of them are blending fewer pixels than their pure theoretical fill-rate capabilities, but this is expected to be the case due to memory bandwidth restrictions - the S18's blend rate is still high enough to suggest that all the ROP's have blending capabilities.
| RightMarkD3D - Texture Fill-rate (No of Textures / Mpixels/s) | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
| S18 Nitro | 530.7 | 265.4 | 179.5 | 137.1 | 106.8 | 88.0 | 74.1 | 62.2 |
| X600 PRO | 450.7 | 325.8 | 235.8 | 154.9 | 113.5 | 89.5 | 73.8 | 62.8 |
| 6600 | 418.1 | 328.2 | 194.9 | 137.4 | 104.8 | 86.4 | 72.2 | 61.8 |
| 6200 | 447.7 | 235.2 | 124.6 | 86.4 | 65.8 | 53.1 | 44.6 | 38.3 |
| S18 % Faster Than: | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
| X600 PRO | 17.7% | -18.5% | -23.9% | -11.5% | -5.9% | -1.7% | 0.4% | -0.9% |
| 6600 | 26.9% | -19.1% | -7.9% | -0.2% | 1.9% | 1.8% | 2.6% | 0.7% |
| 6200 | 18.5% | 12.8% | 44.1% | 58.7% | 62.4% | 65.6% | 66.2% | 62.3% |
The single texture texturing performance of S18 Nitro is higher than the other boards in the test here, which is to be expected given the fill-rate and bandwidths of it in relation to the other boards. The S18's performance drops off in relation to the X600 and 6600 at two texture layers, which certainly suggests that there is only a single texture unit per pipeline. As the number of layers increase the S18 Nitro gradually falls back in-line with the X600 PRO and 6600.
