This paper presents a systematic investigation of optimization strategies for the convolution algorithm. Special attention is given to features relevant for the creation of virtual room acoustics, where the source signal is convolved with a room impulse response signal which has a length of several seconds. Examined were optimizations for the discrete convolution in the time domain and for the partitioned fast convolution in the frequency domain. Applied technologies were usage of AVX instructions, and GPU computing with the OpenCL framework. The results of the various algorithms are evaluated in terms of sample throughput. Various influence factors on the measured performance were identified. It turned out, that even ambitious projects with more than 10 channels and filter response lengths of several seconds may be rendered in real-time with the GPU version of the discrete convolution.