In this paper, a multi-server resource loss system with resource requirement-dependent service times is considered. We develop a convolutional algorithm for the evaluation of the stationary probability distribution and various system characteristics, such as the loss probability and the average volume of occupied resources. In the case study, we calculate these characteristics under the assumption of the Erlang distribution for the service time and several distributions for the resource requirements. We also compare the execution time of the developed convolution algorithm with direct calculations of probabilistic measures.