Microsoft DirectCompute is a set of DirectX 11 APIs that allow general purpose computing (typically performed by the CPU) to be performed on the GPU. This takes advantage of the enormous parallel computation capacity of GPUs that would be well suited for parallel, computationally-intensive operations. DirectCompute is fully supported on DirectX 11 GPUs and partially supported on DirectX 10 GPUs. DirectX 11 runtime libraries are required to run these tests. These libraries are only available for Vista and Windows 7.
There are four different DirectCompute tests available in PerformanceTest 8.
Bitonic sort is a parallel sorting algorithm that is suited for parallel systems such as GPUs. This algorithm is specially designed so that processing threads operate independently on parts of the unsorted list, allowing sorting to be done in parallel. This test can be performed on DirectX 10 and 11 graphic cards.
This test performs a fluid simulation using the MacCormack method to simulate a fire reaction in an enclosed volume. The computationally intensive nature of determining the solutions to partial differential equations used in fluid simulation is ideal for GPU processing. This test can be performed on DirectX 11 graphic cards only.
This test generates a four-dimensional Julia fractal set, which is then rendered in three dimensions and animated. Determining whether a point is in the fractal set is independent of other points, making it well suited to be processed by the GPU. This test can be performed on DirectX 11 graphic cards only.
This test continuously zooms onto a particular point in a two-dimensional Mandelbrot fractal set, generating various shapes and patterns. Like the Julia fractals, generating the Mandelbrot set is well suited for the GPU. This test can be performed on DirectX 11 graphic cards only.