high performance computing on graphics processing units: hgpu.org

Development of Central Processing Unit (CPU) is approaching its physical limits, struggling at the same time with problems like high temperature and increasing need of extra power. There are mainly two alternatives to relieve those problems – Grid computing and Multi Core processors. Both of those approaches are able to give us more computing power than Single Core processors, though it means new algorithms and programs must be developed to be able to use all possibilities of new systems. Although Grid seems to be a very sophisticated system which seems to need a room full of personal computers and Multi Core processors are expensive – all those options are available almost in every modern machine in the form of Graphic Processing Unit (GPU). Generally there are many processors in GPU and they can act like Grid – run in parallel way or same program on different data, or by Flynn’s Taxonomy [1] it is Single Instruction, Multiple Data (SIMD) stream. The rest of this paper introduces technologies used in GPU programming, algorithm of AES block cipher, implementation of this algorithm on two different graphics cards, problems that rose during creation of working code and the results.