The emerging of new technologies in the last years continues with the availability of new building blocks for high-performance computing system-architectures. New generations of processing devices like many-core and vector CPUs, GPUs, and FPGAs accelerate compute-demanding workloads. A deeper memory hierarchy with integrated high-bandwidth memory (HBM) and node-local persistent memory (NVRAM) spans now a range from large-capacity on-node memory, over fast network-attached memory up to far memory on high-capacity parallel file systems. These innovations pose new challenges for application developers and resource providers.
To tackle these challenges, innovations at the software level are required. This includes the further development of advanced programming models and flexible runtime environments for complex application settings.
Further, we evaluate state-of-the-art storage technologies like scalable distributed object stores (DAOS) for future application scenarios.
In close cooperation with technology providers and scientists we develop new methods to make the potential of innovative architectures accessible for various applications, e.g., in material science and numerical mathematics. For that, we design new algorithmic approaches and support the application developers with the integration into their workloads.