Illustration of GPU parallel computing in FamSeq.

<p>The program can be divided into two parts: a serial part and a parallel part. The serial part is processed in a CPU and the parallel part is processed in a GPU. The program: 1. Prepare the data for parallel computing in a CPU; 2. Copy the data from CPU memory to GPU memory; 3. Parallelize the 3ⁿ jobs computing in the GPU, where n is the pedigree size; 4. Copy the results from GPU memory to CPU memory; and 5. Summarize the results in the CPU.