Scalable Computing in a Hybrid System Architecture
The largest performance gains in current supercomputers come from increased parallelism on practically all levels. Within a processor, there are multiple functional units as well as multiple cores. In order to reach Petaflop scale, well over 100000 processor cores need to be employed in such a system. This trend is expected to continue for the next couple of years, thus the biggest challenge is making this computational power available to a wider spectrum of “real” applications, like fluid dynamics or complex weather models. In order to increase single processor performance, various concepts of hybrid computing are under investigation. This paper covers technology trends, an overview of the current Cray XT5h hybrid supercomputer, and provides an outlook to future Cray MPP architectures.
Unable to display preview. Download preview PDF.
- 1.Top 500 supercomputing sites. http://www.top500.org/
- 2.In Cray’s “Cascade”, the computer will adapt to the codes. HPCwire (2006). http://www.hpcwire.com/features/17883334.html
- 3.Dongarra, J.J., Bunch, J.R., Moler, G.B., Stewart, G.W.: LINPACK Users’ Guide. Society for Industrial Mathematics (1987) Google Scholar
- 5.Petrini, F., Kerbyson, D.J., Pakin, S.: The case of the missing supercomputer performance: Achieving optimal performance on the 8,192 processors of ASCI Q. In: Proceedings of the 2003 ACM/IEEE conference on Supercomputing, p. 55. IEEE Computer Society, Washington, DC, USA (2003) Google Scholar
- 6.Snell, A., Goldfarb, D., Willard, C.G.: Designed to scale: The Cray XT5 family of supercomputers. White paper (2007) Google Scholar