There are two tests used here to measure the graphics card's raw fill rate. Fill rate is the measurement of how quickly a graphics card can "paint" textures onto a scene's polygons. In order to rule out other factors that might taint the results these tests use large polygons and textures to try to isolate fill rate performance. The two stages to this test are:
64 individual polygons are used and each requires a single texture. This forces the hardware to apply only a single texture for each pass
This test makes your hardware apply textures to the same 64 surfaces in the quickest way possible, so in the unlikely event that your graphics card is able to apply 64 textures per pass this would be completed in a single pass.
The test result is given in million texels drawn per second (MTexels/s).
High Polygon Count
This test measures the graphics card's ability to handle complex scenes comprising large numbers of polygons. By using very simple textures there is less chance that fill rate limitations will become a factor. The ability of a graphics card to handle large numbers of polygons becomes impeded when those polygons have to be lit so this test is run in two stages:
This test uses only a single light and the surfaces are generally less reflective which allows for a high throughput.
The same test but with eight light sources and an added shiny surface to the dragons on the carousel. This increases the calculations required due to the need for specular reflections. Of the eight lights one is directional and 7 are point lights
The test result is given in million triangles drawn per second (MTriangles/s).
In all honesty this is a slightly rough and ready test that measures frame rates for the same scene when bump mapped using two different techniques. Many factors could make the results of this test doubtful but it's okay as a rough guide and does at least demonstrate support (or lack of) for the two main forms of bump mapping.
Environment Mapped Bump Mapping:
Environment Mapped Bump Mapping (EMBM) was championed by graphics company Matrox which produced the world's first EMBM capable graphics chip, the G400. The advantage to this method of bump mapping is its suitability for use with highly reflective surfaces. Even though EMBM is the oldest method of bump mapping it has seen the least support since its introduction with DirectX 6. However, DirectX 8 increased support by requiring pixel shaders to perform environment mapped bump mapping.
Dot Product 3 Bump Mapping (DOT3):
This method of bump mapping was introduced in DirectX 7 and is a technically far more accurate way of creating a true bump mapped surface. Theoretically this method can only be used to bump map diffuse surfaces but there are workarounds that can be used to extend DOT3 to include specular surfaces.