Active noise control (ANC) is a technology which lowers the noise level by using the principle of destructive interference of sound wave. Performance of active noise control, generally, is highly related to the number of reference transducers, secondary speakers, and error microphones. However, the more number of transducers, microphones and speakers are used for improving ANC performance, the higher computational requirements of such an ANC algorithm would be. Currently, limited computational efficiency of conventional DSP systems hinders the possible maximum noise reduction performance of ANC. In the previous research, for figuring out the limited ANC performance issues due to a lack of computational efficiency, graphics processing units (GPU), which possesses significant parallel computational efficiency, is utilized for ANC algorithms. However, ANC algorithms with GPU computations in the previous researches have limitations on the fact that they are designed in frequency domain structure due to a block data transfer between central processing units (CPU) and GPU, which suffers from a block delay in control signals. Therefore, throughout this research, we proposed a CPU-GPU architecture for ANC algorithms, which does not suffer from the problems of frequency domain algorithms and take advantage of GPU computation. Also, we suggested important factors needed to be concerned when an ANC algorithm based on CPU-GPU architecture is being developed. The feasibility of the proposed CPU-GPU architecture for ANC algorithms is shown with an example of implementing the proposed CPU-GPU architecture on multichannel delayless subband ANC algorithm. (C) 2019 Elsevier Ltd. All rights reserved.