I attended this year’s Supercomputing Conference (SC13) a few weeks ago and have finally had a chance to sift through all the news announcements, technical sessions, and meeting notes to provide you with a few highlights.
Our very own Todd Farrell gave an overview of HMC technology to conference-goers while Micron and Fujitsu announced the integration of our HMC technology into a supercomputer! I think this is a game-changing system design, which will provide flatter, more powerful memory hierarchies for large-scale, high-performance computing (HPC) applications. This is the first public supercomputer design of its kind, and it begins to directly address the memory wall—the holy grail of HPC.
At the Micron booth, we demonstrated an FPGA-based platform for flexibly evaluating the potential of HMC-based systems. This high-performance computing challenge (HPCC) random-access platform initially demonstrated a cube capable of nearly 1 giga-update per second (GUPS), which is over 5X more capable than four DDR3 memory modules! HMC is setting the most challenging memory benchmark in the industry.
I also had the privilege of moderating a panel on the future of memory systems, for exascale and beyond, with a terrific set of panelists: Shekhar Borkar from Intel, Bill Dally from NVIDIA, Andreas Hansson from ARM, Doug Joseph from IBM, Peter Kogge from University of Notre Dame, and Troy Manning from Micron all provided a lively discussion. More than 275 people attended, and it was exciting to see such an interest in memory! Stay tuned for more insights from SC13 from my Micron colleagues…