Design and implementation of computing module based on GPU

With the development of modern underwater acoustic technology, the demand for signal processing capabilities of underwater equipment continues to increase. In addition, from the perspective of hot topics in the signal processing industry, signal processing computing is evolving from specialized signal processing computing to intelligent signal processing computing, and from a single computation intensive approaching to a combination of multi-pipeline parallel computing intensive, memory access intensive, and communication intensive approaches. Overall, there is a trend of segmentation in current signal information processing and computation: neural network intelligent computing (NPU, TPU), parallel accelerated computing (GPU), and general-purpose computing (CPU, DSP) are all on par, while also cross fusing in application fields, making heterogeneous computing a popular trend in current signal processing. This article proposes a computing module based on domestic GPU, which is designed to address the current development of chips both domestically and internationally, as well as the application characteristics of the underwater acoustic industry. The module is defined based on the VPX bus interface and provides communication interfaces for 10G network and RapidIO channels. The module has stable and reliable performance.