Deep Learning Seminar  /  29. Oktober 2020, 10:00 – 11:00 Uhr

STX: Next Generation Stencil and ML Training Acceleration

Referenten: Fabian Schuiki (ETH Zürich) und Dr. Jens Krüger (Fraunhofer ITWM)

Abstract: 

[nur in Englisch verfügbar]

Modern supercomputers are not anymore limited by the number of operations they could possibly achieve but by the power they consume.

With the rise of heavily compute intensive applications like deep learning being applied in many different domains the power- challenge even worsened. Within the European Processor Initiative (EPI) the ETH Zürich and Fraunhofer form a collaboration to design the next generation of extrem energy efficient compute architectures called the STX.\\STX stands for »stencil and tensor accelerator« which describes the main focus it's been developed for in a strict hardware software co-design effort. It enhances the energy efficiency of deep learning application all the way to HPC simulation workflows with novel programming models for the next generation European supercomputers.