Today is that twice-a-year day when the world pays attention to big iron. The latest edition of the Top 500 list of the most powerful supercomputers is released, coinciding with a conference being held in Denver.
The main thing you need to know is that the most powerful system in the world is the same one that topped the list in June: The Tianhe-2. It’s a system developed by China’s National University of Defense Technology, and it is capable of running at 33.86 petaflops. (A petaflop is a quadrillion* calculations per second.)
This supercomputer may be in China, but it’s packed with a lot of American-made chips. Specifically, Intel chips. Tianhe-2 has 16,000 nodes, each of them containing two Intel Xeon Ivy Bridge-generation processors and another three Xeon Phi processors, which adds up to a combined total of 3.12 million computing engines all being harnessed to work at once.
Unless you work with one of these machines, there’s not much reason to give a lot of thought to them in daily life. But they’re performing a lot of important functions from which you probably derive some indirect benefit. One of the systems on the list is involved in predicting the weather for the U.S. National Weather Service. Others may be helping a bank keep track of your money, or mapping genomes, designing drugs, or using complex mathematical algorithms to simulate all manner of complex things, from the planet’s climate to nuclear explosions.
There was no change among the Top 5 systems on the list from June.
Titan, a Cray XK7 system at the U.S. Department of Energy’s Oak Ridge National Laboratory, remained the No. 2 most-powerful machine. Capable of 17.59 petaflops, it uses 261,632 Nvidia-made K20x accelerator cores as its computing backbone. And, while it may have only about half the computing oomph of its Chinese rival, it’s the second most power-efficient system on the list, consuming only 8.21 megawatts to Tianhe-2 s 17.8. Titan was the reigning world computing champ before Tianhe-2.
At No. 3, again, is Sequoia, an IBM-made BlueGene/Q system installed at the Department of Energy’s Lawrence Livermore National Laboratory. No. 4 is Riken, a Fujitsu-made machine in Japan that topped two years ago this month. No. 5: Mira, an IBM machine at the DOE’s Argonne National Lab.
The highest-debuting new entry on the list is Piz Daint, at No. 6. A Cray XC30 installed at the Swiss National Supercomputing Centre in Lugano, Switzerland, it’s the most powerful machine in Europe, and is the most-energy-efficient one in the top 10. It, too, has a lot of Nvidia’s K20x accelerator chips powering it: 5,272 of them to be exact, making for a total of 73,808 cores.
At No. 7 is a Dell-made machine called Stampede, installed the University of Texas Advanced Computing Center in Austin. In June, it was No. 6. The next three machines rounding out the Ttop 10 – two of them in Germany, and one in the U.S. – were all made by IBM.
By way of measuring the march of supercomputing progress in the last six months, here are a few other highlights from the overall list to chew on. There are now 31 machines that can boast top performance of one petaflop or better, up from 26 on the list in June. And the entry point – the minimum performance required to make it onto the list – is 117.8 teraflops, up from 96.3 six months ago.
Intel chips are by far the most popular computing engine used in the systems on the list, showing up in 412 of the 500, or 82 percent. Opteron chips from Advanced Micro Devices were in 43 systems. IBM’s Power chips were in 40 systems. Nvidia’s GPU-based accelerator chips show up 38 systems.
Hewlett-Packard sold more of the systems on the list than anyone, accounting for 195, or 39 percent, of the 500. IBM was second, with 166 systems. If you added up the total computing performance of all the systems from each vendor, HP would rank fourth, while IBM would rank first.
Geographically, the U.S. is still the supercomputing leader, and is home to 265 of the systems on the list, up from 253 six months ago. Europe was second with 102, down from 112. China had 63, and Japan 28.
This is the 42nd time this list has been put out. It’s a joint project run by Hans Meuer of the University of Mannheim in Germany, Erich Strohmaier and Horst Simon of Lawrence Berkeley National Laboratory, and Jack Dongarra of the University of Tennessee, Knoxville. You can see the full list here.
The list pretty much covers the waterfront in listing the supercomputers publicly known to exist around the world. For the most part, the universities and other entities that have them, like to brag about their position on the list when they can. What the list doesn’t cover are the secret machines that might be used by government agencies like the U.S. National Security Agency and similar government spy shops around the world. One wonders if there’s information about just such a machine in the files of Edward Snowden.
* Quadrillions come after trillions, in case you hadn’t been keeping track.
(A small correction: I initially said Titan – number two – was the most energy-efficient system. It’s actually Piz Daint, number six.)