RIKEN Institute's Next Supercomputer: EXAFLOPS by 2020

RIKEN New Supercomputer - AkihabaraNews.com

November’s SC13 Supercomputer Conference confirmed China’s Tianhe 2, clocked at 33.86 PETAFLOPS, as the world’s fastest. Japan’s K Computer comes in at #4, but the RIKEN Advanced Institute for Computational Science has plans to smash both rankings, and it’s impressive, but also relative.

• • •

Goin’ all FLOPSy
Now, as previously mentioned here, it really doesn’t do much for the average citizen to understand the minutiae of what exactly FLOPS are (but if you must, it’s an acronym: FLoating-point Operations Per Second). And for most of humanity, ignorance of what exactly a “floating point calculation” is can probably be taken to the grave (but if you also must, it’s a big-ass mathematical expression with a wildly variable, multi-digit exponent).

In a layperson’s nutshell summary, the pursuit of hardcore FLOPS involves doing long, complicated math with what is effectively incomprehensible speed and accuracy. This is something at which silicon and plastic and semiconductors excel; for humans and their meat-based, shoulder-mounted computers, well, that kind of math is exactly why we need computers.

Now, in discussing such things, because saying “kajillion quintillion quadrillion billion thousand” over and over again is inefficient and lame, we substitute a set of Greek-derived prefixes: mega, giga, tera (most computery people know those three), peta, exa, zetta, and yotta; after yotta we just kinda shrug.

RIKEN Upping the FLOPS, but it’s Relative... 
Digression aside, the news at hand is Japanese computational research organization RIKEN’s plans for an as of yet unnamed machine slated to be 100 times faster than their K Computer and 30 times faster than China’s Tianhe 2, the current world leader. The plan is to bring the machine online in 2020, and this iteration will be the first to break the EXAFLOPS barrier, i.e., capable of at least one quintillion (a million trillion) FLOPS. The project is funded by MEXT, Japan’s Ministry of Education, Culture, Sports, Science, and Technology.

It sounds impressive - and it is - but here’s the relative part: the petaflop range of capability (that’s at least a quadrillion FLOPS), currently inhabited by all the world’s top supercomputers, was only breached in 2008, and the title of world's fastest has changed hands several times since. Sure, unless someone beats RIKEN to it, their exa-scale machine will be the first to cross that threshold, but if Moore’s Law and similar trends continue to hold, their 2020 supercomputer won't be the fastest for very long. Moreover, the next order of magnitude in performance, the zetta-scale, is likely to arrive pretty quickly thereafter. So, well done - but don't sleep.

Why Supercompute?
Regarding the utility of RIKEN’s K Computer successor, as with most of the top machines certifed by the SC13 Supercomputer Conference, the usual purposes are lined up: environmental simulation, modeling the universe and stuff, super-advanced chemistry, and the latest fashion, something RIKEN’s K Computer already ponders: baby steps toward human brain simulation.

One has to wonder, when will supercomputers be tasked with designing supercomputers? That’ll be… interesting.

• • •

Photo: RIKEN, specific source unavailable