All processes in a software implemented inference machine are called fuzzy processes. These processes are fuzzification, fuzzy inference and defuzzification, executed in that order. The implementation itself is called a fuzzy inference machine or a fuzzy logic system. This thesis describes such a system in a parallel implementation with a specific approach, where complex operations are broken down into multiple simpler ones. We used the CUDA architecture, which allows us the usage of general purpose parallel computing on modern GPU’s. At the end we tested this implementation, in comparison with the sequential implementation on the CPU, by comparing the precision of the computational results and the needed times of operation algorithms.