Abstract: CUDA is intended to give full play to the advantages of ultra-high computing performance of GPU in data processing, scientific computing and other fields of general purpose. This paper studied MD5 crack method based on Compute Unified Device Architecture and carried on hybrid compilation by using VS-2005 and NVCC. The experiment runs CUDA programs on GcForce-9600GT graphics card and runs Standard-C programs on Quad-Corc CPU-Q6600.The results shows that, under the environment of high computational load and huge amounts of data, the computing speed of mid and low end graphics card is 30 to 50 times higher than that of high end CPU. CUDA gives full play to the advantages of GPU Streaming　Multiprocessors Array and greatly improves the efficiency of the parallel computation　programs.

Key words: MD5 crack, CUDA, Parallel computation