When the Tianhe-1A system took the top spot on the TOP500 last year, there seemed to be a lot punditry that the system wouldn’t scale on real applications. Well, put that notion to rest because researchers at the Chinese Academy of Sciences’ Institute of Process Engineering have completed the world’s highest performing molecular simulation to examine improved techniques for more efficient production and use of crystalline silicon, a key material used in solar panels and the semiconductor industry.
“Researchers at the Chinese Academy of Sciences’ Institute of Process Engineering (CAS-IPE) used Tianhe-1A, the world’s fastest supercomputer, to perform a simulation on NVIDIA Tesla GPUs that was five times the performance and more than twice the size of the previous highest-performing molecular simulation. The simulation modeled the behavior of 110 billion atoms at an unprecedented 1.87 petaflops of performance. The previous record for a simulation of this kind was 49 billion atoms at 369 teraflops of performance. For details on this type of simulaiton, read the white paper on from LANL and IBM.
To learn more, I caught up with Nvidia’s Sumit Gupta, Senior Marketing Manager for Tesla:
insideHPC: There are reportedly only a handful of codes running at sustained Petaflops today. Is this molecular simulation code “embarrassingly parallel?“
Sumit Gupta: No, this app is actually far from being embarrassingly parallel. In fact, due to the multi-body nature of the potential, its parrallelization is much more complicated than standard pair potential MD code.
insideHPC: Is Tianhe-1A often used by researchers as a single system, or is typically partitioned off for use by multiple users?
Sumit Gupta: Both are happening. Multiple user cases are more typical at this point
insideHPC: Is the study of new materials one of the primary uses of Tianhe-1A at this time?
Sumit Gupta: No, it is just one of many types of research going on Tianhe-1A
insideHPC: How far away are we from being able to engineer new materials at the molecular level?
Sumit Gupta: Insights from molecular simulations are used today to engineer new materials. In particular, quantum chemistry codes are used for insights into new chemicals like catalysts, better lubricants, etc.
insideHPC: Will these results be presented at the ISC’11 conference coming up in a few weeks?
Sumit Gupta: Six members of the CAS-IPE research team will present this research at the HPC in Asia workshop at ISC. It will also be presented at the upcoming GPU Tech Conference (GTC) in San Jose, CA, May 14-17, 2012, and at GTC China taking place December 15-16 2011 in Beijing.