R-CCS Cafe 第204回 第1部

Since the processing performance of FPGA devices have increased rapidly, FPGAs are becoming competitive for real-world high-performance computing applications, in terms of not only power efficiency but also processing power.
Therefore, a large-scale FPGA cluster, which combines power efficiency and computational performance, is expected to be one of the candidates for HPC systems in the near future.
However, due to the difficulty of designing an application as hardware and mapping it onto an FPGA, there is a need for a more user-friendly interface and development environment.
Furthermore, the previous studies of FPGA clusters, in which the entire hardware system including the network is optimized for the specific application, are not suitable for systems in which a lot of users execute a variety of applications.
To solve these problems, our research team is developing an easy-to-use platform system that maintains the performance of the FPGA cluster and improves its versatility, both in terms of hardware and software.
In this presentation, I will show our achievements and future work on the FPGA cluster system, especially the network that connects FPGAs and the communication system in the cluster.