Surface radars are cyber-physical systems that autonomously produce and process a large amount of sensor data, in the order of terabits every second. This data deluge prompts the need for a large amount of computing power to extract an air picture in real-time, and within tight latency constraints.

Due to the nature of the tactical field of operation of surface radars, streaming that amount of data to a remote computing facility (be it cloud, supercomputer or edge datacenter) is not practically feasible today, so the necessary computing power must be brought next to the actual sensor.

This talk explores the constraints of running real-time HPC loads on the far edge, the trends at this extreme of the compute continuum, how similar and different surface radars are to classical HPC systems and to the more challenging data-ingestion HPC systems.

It also shows where HPC already helps during systems development (simulation and digital twins, verification and validation of software-based systems, more recently in AI training) and where it is going (deeper integration of AI, simulation tools and operational software) and the challenges of using bigger systems such as large public HPC systems and AI factories.