K Computer claimed the top position on the Top 500 list back in June with a peak performance of 8.8 PFlops, which was achieved with 68,544 8-core Spark VIIIfx processors. The updated system includes 88,128 CPUs in 864 racks and scored a LINPACK benchmark result of 10.51 PFlops, which makes it the world's first known supercomputer to exceed the 10 PFlops barrier.
10 Pflops translates to 10 quadrillion floating point operations per second, which means that all 7 billion people in the world would need about 16.5 days to achieve the same number of calculations that K does in 1 second - if we assume that those 7 billion people can post one calculation result every second.
It is unclear how much power this computer system consumes, but if we consider the fact that the 8.8 PFlops system was rated at 9.9 MW, we can safely assume that this updated system will consume close to 13 MW.
The Oak Ridge National Laboratory recently said that it will be completing its Titan supercomputer, which is estimated to hit a performance of 20 PFlops in late 2012 or early 2013. Titan will integrate 18,688 16-core Opteron processors as well as 7000 to 18,000 Nvidia Kepler GPUs. Another 20 PFlops supercomputer will be Sequoia, a BlueGene/Q-based system that will integrate about 100,000 16-core PowerPC A2 processors, which will go online sometime in 2012 at the Lawrence Livermore National Laboratory.
The next milestone will be the 100 PFlops mark, which IBM recently mentioned in a patent filing. IBM's proposed BlueGene/Q system will include 524,288 16-core PowerPC A2 processors with a total core count of nearly 8.4 million. IBM estimates that this system can deliver up to 107 PFlops at a power consumption of about 15.7 MW.
Scientists believe that the first Exascale supercomputer will be possible by about 2020.