High compute capabilities are one the key factors for many computational engineering research projects. In order to fulfill this requirement hardware vendors developed new heterogeneous parallel architectures over the last decades. On the one hand, these architectures improve the energy efficiency to lower the power consumption and thus to decrease the total cost of ownership (TCO) for the computing centers. On the other hand, new programming approaches and optimization techniques are required to benefit of these highly parallel systems.
In 2016, RWTH Aachen University has chosen the company NEC as supplier for the new supercomputer CLAIX. As part of this procurement, NEC agreed to fund a scientific-technical collaboration, which has the goal to improve the efficiency, reliability and productivity of High Performance Computing on the NEC CLAIX system. This includes the identification of suitable codes within the user base of CLAIX with respect to the use of SIMD and vector programming models. Furthermore, the project evaluates an OpenMP-based offload programming model for the upcoming NEC Aurora architecture. A prototype implementation based on the LLVM compiler infrastructure is one of most important aspects of this part of the project, because this extends the use cases for the NEC Aurora architecture and enables all users of CLAIX to develop reliable standard-compliant applications.
More information will be made available after the product launch of the NEC Aurora architecture.