Hans Meuer Award for Paper on In-Memory Processing

For modern high-performance computing architectures, it is becoming increasingly difficult to balance compute performance and data bandwidth. This challenge could be mitigated by moving compute capabilities closer to the data. This approach has yet another advantage: it has huge potential for improving energy consumption, as data movement is, in most cases, more expensive than data processing. One example of moving compute capabilities closer to the data is known as in-memory processing. A team at IBM Research recently designed a new version of such an architecture, which exploits new options for the three-dimensional stacking of logic and memory dies and is therefore called the Active Memory Cube (AMC).

Researchers at JSC are collaborating with IBM at the Exascale Innovation Center to explore the capabilities of this new technology. Performance-critical code sections of relevant applications have been implemented for the new architecture, such that performance can be investigated on the basis of cycle-accurate simulations. In order to explore performance at system level, a performance model approach was applied.

The results were presented in the paper "Accelerating LBM & LQCD Application Kernels by In-Memory Processing" which was not only accepted by the ISC High Performance 2015 conference, but also received the Hans Meuer Award. This award was introduced in memory of Hans Meuer, who was a co-founder of this conference, as well as of the TOP500 list.
(Contact: Prof. Dirk Pleiter)

JSC News No. 232, July 2015