FYI.

This story is over 5 years old.

Tech

The Fastest Supercomputers Are Getting Slower at Speeding Up

But at the end of the day, it’s not about how many flops you have, but what you can do with them.
Image: Dario Lo Presti/Shutterstock

Supercomputers are continuing to get faster—but they’re not getting faster as fast as they used to. That’s one of the takeaways from the latest Top500 list of the most powerful supercomputers, which was released today at the International Supercomputing Conference in Leipzig, Germany.

As you’d expect, the supercomputers on the list are all still improving in performance. In the top spot is the Tianhe-2 at China’s National Super Computer Center in Guangzhou, a machine that’s hogged the limelight in the list’s last three iterations. It has a performance of 33.86 petaflops per second, aka a hell of a lot. Flops is a measuremnt of the number of floating-point operations a computer can do per second; the “peta” prefix means quadrillion, i.e. there are 15 zeroes after that number, if you were to write it out.

Advertisement

Even by supercomputer stakes, Tianhe-2 is mighty powerful. To put it in perspective, the second supercomputer on the list, the Titan at the Oak Ridge National Laboratory in the US, comes in at a (now) comparatively slim 17.59 petaflops.

But while the numbers are boggling, they’re not as high overall as you might expect by looking at the trend over previous lists. Combined across all the 500 computers on the list, the sum performance is now at 274 Pflop/s, from 250 six months ago. So it’s climbing, but at a rate much reduced than in previous years. As Top500 wrote in their release, “For the second consecutive list, the overall growth rate of all the systems is at a historical low.”

This shows the growth in performance of the entire list (purple), the top supercomputer (blue), and the 500th (orange). Image: Top500

So why is that? Given that the observation is based on the cumulative performance of the entire list, it’s not as simple as surmising that each machine isn’t improving as fast as expected. Rather, it’s a reflection of the market as a whole. In the past, the performance of the overall list has been bolstered by some real kickass supercomputers (like Tianhe-2) pushing into the top of the list.

When Tianhe-2 comes in at 33.86 petaflops, hogging over ten percent of the entire list’s combined performance, it makes a huge impact on the overall performance of the list. But meanwhile, lower down, performance growth has been slowing for a while. It’s not as noticeable in the overall counts because the amounts you’re talking about down in place 500 are much smaller; the 500th supercomputer “only” comes in at 133.7 teraflops, orders of magnitude less powerful than the top dogs. But with fewer new superstar systems in recent years, the dwindling growth at the bottom of the list is no longer being counteracted so much.

“This offers an indication that the market for the very largest systems might currently behave differently from the market of mid-sized and smaller supercomputers,” wrote the Top500 compilers.

Over at HPC Wire, Nicole Hemsoth broke down this trend further, and discussed what it could mean. She notes that the trailing off in performance growth would be even more noticeable if the top part of the list wasn’t still propping up the overall count, but suggests this could be down to the demands of real-world applications. “When examined as a whole, we’re falling off except at the highest end…but what does this mean for end user applications? Is high end computing getting smarter in terms of efficiency and software to where, for real-world applications, FLOPS no longer matter so much?”

There’s no question that we could use more powerful supercomputers than what we have now for some applications. Take the Human Brain Project’s mission to build a model of an entire human brain—that will take an exascale supercomputer, which means we’re talking not just in quadrillions but quintillions (18 zeroes). But at the end of the day, it’s not about how many flops you have, but what you can do with them.

Meanwhile, even as the overall top 500 see a dip in performance growth, they’re still increasing at a strong rate, especially over at the top end. If things go according to Top500’s projected graphs, we could be seeing that exaflop performance in the number one spot by around 2020.