As always, additional synthetic tests are included in this version. There are six of them, and like all synthetic tests their purpose is to isolate one aspect of the performance or the architecture of a GPU. They don’t affect the overall score.
As usual, the first two tests determine the fill rate of the graphics subsystem; the first tests involve the texturing fill rate while the second measures color fill rate.
More interestingly, the third test measures performance on a heavy pixel shader using parallax mapping in order to simulate a complex geometry (12 flags in the wind) starting with a simple rectangle (there are only two triangles present in this scene). Ray tracing is used on a large depth map to determine the intersections of the geometry with the camera point of view, and the surface is rendered using Strauss shading. This is also the test that produced the most surprising results:
Here, while the results remained more or less coherent when you limit them to each manufacturer, the results for Radeon were incredibly weak compared to those of the GeForce. That result can have a thousand different explanations depending on the exact nature of the shader being executed, and especially on the drivers, as we’ll later see.