World supercomputing capability more than triples

In a breath of good news for Japan this year, RIKEN's supercomputer "K Computer" vaulted to the top slot in world supercomputing in June 2011 as tracked by Top 500 Supercomputer Sites.

Remarkably, capability more than tripled to over 8 petaflops per second (8 quadrillion calculations per second, measured as the Maximal LINPACK performance achieved), after supercomputer performance had been asymptoting at close to 1 pf and 2 pfs for the last three years (Figure 1). China's Tianhe-1A at the National Supercomputing Center in Tianjin was in second place, and the US's Jaguar Cray at Oak Ridge National Lab in third.

The capacity tripling constitutes obvious potential benefits to scientific computing, the realm of applications for which supercomputers are used. It is hoped that these kinds of quantitative changes may eventually lead to qualitative changes in the way other problems are investigated, for example how the brain works.

Figure 1. Source: Top 500 Supercomputer Sites

Supercomputing and human intelligence

As of November 2009, the world’s fastest supercomputer was the Cray Jaguar located at the U.S. Department of Energy’s Oak Ridge National Laboratory, operating at 1.8 petaflops (1.8 x 10¹⁵ flops). Unlike human brain capacity, supercomputing capacity has been growing exponentially. In June 2005, the world’s fastest supercomputer was the IBM Blue Gene/L at Los Alamos National Laboratory, running at 0.1 petaflops. In less than five years, the Jaguar represents an order of magnitude increase, the latest culmination of capacity doublings each few years. (Figure 1)

Figure 1. Growth in supercomputer power

Source: Ray Kurzweil with modifications

The next supercomputing node, one more order of magnitude, at 10¹⁶ flops, is expected in 2011 with the Pleiades, Blue Waters, or Japanese RIKEN systems. 10¹⁶ flops would possibly allow the functional simulation of the human brain.

Clearly, there are many critical differences between the human brain and supercomputers. Supercomputers tend to be modular in architecture and address specific problems as opposed to having the general problem solving capabilities of the human brain. Having equal to or greater than human-level raw computing power in a machine does not necessarily confer the ability to compute as a human. Some estimates of the raw computational power of the human brain range between 10¹³ and 10¹⁶ operations per second. This would indicate that

supercomputing power is already on the order of estimated human brain capacity, but intelligent or human-simulating machines do not yet exist.

The digital comparison of raw computational capability may not be the right measure for understanding the complexity of the brain. Signal transmission is different in biological systems, with a variety of parameters such as context and continuum determining the quality and quantity of signals.