HPC Processors and Accelerators
As part of the European Processor Initiative (EPI), we are working with our partners to develop high-performance architectures for High Performance Computing (HPC) processors and application-specific accelerators. The goal of these developments is to enable demanding computing applications for exascale computers in a highly energy-efficient manner (1 ExaFLOPS=10^18 floating point operations per second).
STX stencil and tensor accelerator as chiplet
In cooperation with the Fraunhofer Institute for Industrial Mathematics (ITWM), ETH Zurich, EXTOLL and other partners, we are realizing highly efficient SoC architectures for 2D and 3D stencil-based applications. For this purpose, a specially optimized stencil processing unit has been developed with the support of RISC-V processors. This architecture is not only scalable in terms of computing power, but also consumes only a fraction of the energy compared to commercially available HPC processors or GPUs with identical computing power.
Based on such architectures, we develop scalable, customer- and application-specific accelerators that are easy to program, offer extremely high performance and energy efficiency, and represent a high degree of technological sovereignty as purely European solutions.
Applications
- Machine Learning with Convolutional Neural Networks (CNNs)
- Current and wave simulations
- Weather and climate simulation
- Big data analytics
- Data center flow optimization
- Materials research
- Seismic exploration