The TOP500 supercomputer tabulators announced yesterday (18 November) at the SC13 conference in Denver that the world most powerful number-crunching machine is…drum roll please…the very same computer that was leading the pack in June: the Tianhe-2 (“Milky Way 2”) supercomputer developed by China’s National University of Defense Technology. (See below for the latest listing of the top 10 supercomputers.)
As was reported in June, Tianhe-2 completed an astronomical 33.86 petaflops in benchmark tests. That’s 33.86 x 1015 floating point calculations per second, making it the almost twice as powerful as the next runner up, the Titan supercomputer at Oak Ridge National Laboratory (ORNL), which clocked 17.59 petaflops on the same tests. So the Tianhe-2 must be twice as useful, right?
At the end of May, the University of Tennessee’s Jack Dongarra (who helps conduct the TOP500 rankings) attended International High-Performance Computing Forum in Changsha, China, which was organized by China’s National University of Defense Technology. During that trip, he and other conference participants were given a tour of the Tianhe-2 supercomputer, which he described in a brief report.
The hardware Dongarra saw during that tour was duly impressive, as was the performance demonstrated in benchmark trials. But the section of his report on applications raises some doubt about the how useful the Tianhe-2 actually is, at least for the moment. That section leads off with the statement, “[The National University of Defense Technology] claims to have a number of applications that are being ported to the TH-2.”
Notice Dongarra’s choice of the word claims. That this very tentative statement was written very shortly before the computer was declared operational in June makes you wonder whether the makers of this computing colossus had given adequate thought and energy to preparing the applications to be run on it.
Contrast this with ORNL’s efforts to have five target applications ready when its Titan supercomputer came online for researchers early this year. ORNL had set up a Center for Applications Readiness to help researchers prepare code that could utilize Titan’s computing architecture. The targeted applications were:
- S3D, for simulating combustion
- LSMS, for studying magnetic materials
- LAMMPS, for studying molecular dynamics
- Denovo, for studying neutron radiation in nuclear reactors
- CAM-SE, a version of the community atmospheric model for studying climate change
So it would seem that the folks at ORNL were trying very hard to live up to the tag line on the “Introducing Titan” website: "Science on Day 1."
Tianhe-2’s makers also have some applications in mind, but the description they shared with Dongarra in May was suspiciously vague. Two of those applications are the kind of things you'd expect a supercomputer to tackle: computational fluid dynamic modeling (in particular for the design of China’s upcoming C919 airliner) and gyrokinetic toroidal code for studying magnetic-confinement fusion. But the two other applications areas Dongerra reported--"business opinion analysis" and "security e-Government Cloud"--could mean almost anything and in any event strike me as things that could be handled readily enough using more typical data centers.
The only concrete evidence I can find of a real-life application now running on the Tianhe-2 supercomputer is from September, by a group doing atmospheric modeling at Tsinghua University in Beijing. Perhaps there’s more being run on the Tianhe-2 that’s not easily accessible at the moment unless you’re in the know (or speak Mandarin). If you are aware of any applications that have been successfully ported to the Tianhe-2 or have used this supercomputer yourself to advance the frontiers of science or engineering, please share a little about what you've been doing in the comments section.
Top 10 Supercomputers of the 18 November TOP500 Ranking
Computer (performance on benchmark tests)
- Tianhe-2 (MilkyWay-2) (33.86 petaflops)
- Titan (17.59 petaflops)
- Sequoia (17.17 petaflops)
- K computer (10.51 petaflops)
- Mira (8.59 petaflops)
- Piz Daint (6.27 petaflops)
- Stampede (5.17 petaflops)
- JUQUEEN (5.01 petaflops)
- Vulcan (4.29 petaflops)
- SuperMUC (2.90 petaflops)
David Schneider is a senior editor at IEEE Spectrum. His beat focuses on computing, and he contributes frequently to Spectrum's Hands On column. He holds a bachelor's degree in geology from Yale, a master's in engineering from UC Berkeley, and a doctorate in geology from Columbia.