Java on CUDA architecture

Abstract

This paper presents technologies, programs and Java libraries which allow usage of a CUDA capable GPU device within Java programming language. All these approaches are described with their simplified usage guide. Then, we compare performance of these methods, as well as we compare their friendliness to a programmer, documentation or their maturity. For performance tests, we used matrix multiplication and Gamma correction. We recommend to use JCuda library as currently the best available method, followed closely by JNI.