Game Test 3: "Trolls Lair"

See here for details of this test.

Given the similarity in the rendering techniques employed by this test and Game Test 2, its hardly surprising that we see both similar scores and similar trends across all the boards.

As mentioned in our Introduction to 3DMark03, the major difference between this test and Game Test 2 is in the quantity of geometry used, in that this test a somewhat more geometry intensive. We can see that this appears to be the case with all the DX9 Radeon's, which have 4 Vertex Shaders each, as the FPS scores at lower resolution are all lower in GT3 than they are in GT2. Interestingly, all four of the other boards that have a lower vertex rate actually perform better at the low resolutions in GT3 than they do in GT2.

Game Test 4: "Mother Nature"

See here for details of this test.

With this test utilising DX9's PS2.0 Pixel Shaders, we are reduced to only using the four DX9 Radeon boards. As we can see the benchmark tells an interesting, but not entirely unexpected story.

This test uses both Pixel Shaders and high levels of Vertex Shading. The Vertex Shading in this test is used to produce and animate all the leaves and grasses, which is going to require plenty of calculation. The fill-rate graph shows quite a continuously steep graph for the 9700 PRO, which would tend to indicate that this test is quite bound by the geometry processing for much, with an increasing Pixel Shader limitation through the higher resolutions. This is further highlighted by how close the 9700 and the 9500 PRO are performing, given that both parts have the same Vertex and Pixel Shader rates but are only differentiated by their respective bandwidth.

However, when we compare the performances of the 9500 PRO against the 9500, which has half the Pixel Shading capabilities we can clearly see that the two times extra pixel shading rate the 9500 PRO has over the 9500 is a big boon, as the 9500 becomes predominantly fill-rate (Pixel Shading rate) bound at relatively low resolutions. What's also important to take away here is that clearly the test is also bound by the Pixel Shading rate and how little the difference makes between the 9700 and the 9500 PRO. Because the Pixel pipelines are tied up with plenty of maths calculations for the Pixel Shader programs it is neither addressing textures or writing pixels to memory as frequently, thus reducing the need for high bandwidths, traditionally seen with highly textured, 'fixed function' rendered scenes.