I like to track various measures of our civilization’s scientific and technological capabilities, from the number of exoplanets discovered to the amount of data collected by LHC experiments. In computing, I use the Top500 list of supercomputers, ranked by their ability to run the LINPACK benchmark. The list is compiled every June and November, and today the June 2012 list landed, with the most pertinent details contained in this poster.
The IBM BlueGene/Q Sequoia supercomputer at Lawrence Livermore takes the top spot from the Japanese K computer, with a peak performance of 16.33 PFLOPS (that is, 16,330,000,000,000,000 floating point operations per seconds) while the K remains at about 10 PFLOPS. Total aggregate performance has gone from 74.2 PFLOPS to 123.02 PFLOPS, and was 58.7 PFLOPS a year ago. The number using coprocessors (mostly nVidia GPUs) has gone from 17 to 39 to 57 over the past three lists. The trajectory still seems on course to break the exaflop barrier around 2019, around the same time the #500 computer on the list should be about as powerful as the present #1.