Tomi Erlih (2010) Parallel computing on graphics processing unit. EngD thesis.
This diploma shows how to solve a compute-intensive problem using a graphics processing unit. Current graphics cards have a highly sophisticated and powerful processing unit, best suited for parallel computing where one operation is performed on multiple data. To test the performance of the graphics card, a test application was developed. It simulates the motion of stars in the universe, where each star pulls on all other stars using only its gravitational pull. N-body simulation is used to solve this astrophysical problem. Two separate development libraries are used to utilize the computing power of the graphics card. One library is used to work with the CUDA architecture while the other library implements the OpenCL standard. For the best estimate of the actual computing power of a graphics card, its performance is compared with the parallel capabilities of a common computer processor and with the performance of this processor when using the advanced algorithm Barnes-Hut. The goal of this diploma is to find out whether or not the use of this technology is feasible and to compare its performance with the parallel computing technologies that are used today.
Actions (login required)