Tuesday, December 17, 2013

Supercomputing Processing Speed Nearly Doubles in One Year

The Top500 November 2013 biannual list of the world’s fastest supercomputers shows China's Tianhe-2 still at nearly twice the capacity of the second fastest, with virtually no change since the machine vaulted onto the list in June 2013.

Tech Specs: Tianhe-2 (MilkyWay-2) - TH-IVB-FEP Cluster, Intel Xeon E5-2692 12C 2.200GHz, TH Express-2, Intel Xeon Phi 31S1P. The machine was constructed by the National University of Defense Technology (NUDT).

Tianhe-2's maximum processing power is 33.8 petaflops per second (and peak processing speed 54.9 petaflops per second). The nearest competitor is the US DOE's Titan, a Cray XK7 Opteron 6274 16C 2.200GHz with a maximum processing power of 17.6 petaflops per second. Processing power is firmly on a steep growth curve, accelerating since the 5 petaflops per second mark was surpassed in June 2011 by RIKEN Japan (Figure 1).

Figure 1: Supercomputing Processing Capacity (Source: Top500)

Supercomputers, although processing faster than the human brain for some time now, process in a massively linear parallel manner which is not at all how the brain functions. However, while not mimicking the human brain, trying to understand it is a key use of supercomputers, along with other traditional prediction problems like weather forecasting, and physics phenomena and energy modeling. Now firmly in the big data era, processing astronomical data too is a key use for supercomputing. Maybe Tianhe-2 will compute findings from lunar data processed through China's recently landed Jade Rabbit. In other astronomical applications, astronomers expect to be processing 10 petabytes of data every hour from the Square Kilometer Array telescope under development in Australia and South Africa with a total collecting area of one square kilometer.

blog comments powered by Disqus