Figure 3.

Typical GPGPU application flow. Input data for a GPGPU application must be copied to the GPU's memory along with a pre-allocated output buffer prior to invoking the GPU-based kernel. Output from the kernel is read back into main memory and reported to the user.

Schatz et al. BMC Bioinformatics 2007 8:474   doi:10.1186/1471-2105-8-474
Download authors' original image