GRAPHICS PROCESSING UNITS APPLIED TO EXECUTING INTENSIVE NUMERICAL ALGORITHMS

Grafična procesna enota se je v zadnjih letih razvila iz preprostega procesorja v ekstremno zmogljiv paralelni procesor. Zmogljivost grafične procesne enote lahko poleg grafične obdelave izkoristimo za reševanje zahtevnih in numerično intenzivnih algoritmov. V diplomskem delu je predstavljena arhitektura CUDA in programski jeziki za programiranje grafičnih procesorjev. Za primerjavo zmogljivosti in učinkovitosti grafičnih procesnih enot je predstavljena primerjava med izvajanjem numerično intenzivnega algoritma na grafični procesni enoti in izvajanje istega algoritma na centralni procesni enoti.In the last few years the graphics processing unit has developed from a simple processor into an extremely powerful parallel processor. The performance of a graphics processing unit can be put to good use to do graphics work as well as to solve difficult and numeric intensive algorithms. The following diploma dissertation presents the CUDA architecture and program languages for programming graphics processors. To compare the performance and the efficiency of graphics processing units, I made a comparison of executing numerical intensive algorithms on a graphics processing unit and executing the same algorithms on a central processing unit