Programmable Near-Memory Acceleration on ConTutto


Abstract to be Presented at Summit 2016

Our team at the IBM Zurich Research Laboratory is investigating hybrid computer systems that offload part of the processing to programmable near-memory accelerators that tightly integrate computation with main memory operation. Besides exploiting a data-centric concept to limit data transfers, this research targets to further improve power efficiency by 1) letting the memory system play a more important role in workload optimization by enabling different ways to adapt the memory system operation to the workload behavior, and by 2) exploiting a novel architecture and programming model to reduce programmability and related overhead compared to traditional architectures.

At the OpenPOWER Summit, I will present an initial implementation of our work on the ConTutto card (introduced at last year’s OpenPOWER Summit) that allows programmable near-memory acceleration to be performed directly on the POWER memory bus. One of the applications that will be demonstrated is FFT.

Speaker Bio

Jan van Lunteren has been with the IBM Zurich Research Laboratory, Switzerland, since 1994. His current interests include near-memory processing, deep learning architectures, regular-expression engines for pattern scanning at Terabit/s speeds, and high-performance programmable accelerators. He is also involved in the Dome project, which is a research collaboration between IBM and the Netherlands Institute for Radio Astronomy (Astron) to investigate Exascale computer systems targeted at the Square Kilometer Array (SKA), which will be the world’s largest and most sensitive radio telescope when it becomes operational in the next decade. Jan has an M.Sc. and a Ph.D. degree in Electrical Engineering from the Technical University of Eindhoven, The Netherlands. He holds more than 60 issued and pending patents.