High Performance Computing (HPC) technologies have been evolving fast during recent years. Despite the legacy x86 CPU, still the most widely adopted processing units, GPUs and accelerators are becoming more and more relevant. In the next future, the scientific & technical community will be confronted with an even larger diversity of architectures feeling the impact at all levels: hardware, operative system, compilers, libraries and applications. This presentation details the different hardware technologies dedicated to HPC with a special emphasis on the ARKA compute nodes, designed by E4 which are based on ARM 64 bit System on Chip (SoC) combined with GPU accelerators and low latency network infiniband.
In order to validate the platform several tests are showed spanning from the synthetic benchmarks up to real applications, such as: fluid-dynamics, molecular modeling, deep learning, proving that low-power solutions can effectively be implemented in industrial environments benefiting performances and productivity.