China's Tianhe-2 Supercomputer Is Still the World's Most Powerful

Sorry, America, but the most powerful computer on the face of the planet still resides in China.

That’s the pronouncement of researchers who assembled the twice-a-year Top500 list of the world’s top supercomputers, released this morning at a conference in Frankfurt, Germany. The latest list marked the fifth consecutive six-month period that China’s Tianhe-2 computer has topped the list, and the sixth time overall since 2010 that a Chinese machine has taken the crown.

The machine was designed by China’s National University of Defense Technology and deployed at the National Supercomputer Center in Guangzhou. Its top performance is measured as 33.86 quadrillion floating point operations in a single second. A floating point operation, or flop, is a math problem that involves fractions, and when measured in quadrillions (which follows trillions) they are referred to as petaflops. That makes the machine about 5,000 times more powerful than the most tricked-out Mac Pro you can buy from Apple.

But it couldn’t have that kind of computing oomph without some American-made technology. The Tianhe-2 sports 16,000 computing nodes, each packed with two Intel Xeon Ivy Bridge processors and three Xeon Phi processors. Together they combine for 3,120,000 computing cores.

There’s a fair amount of domestically developed technology in it, too. Connecting the nodes and allowing them to work together is a Chinese-developed technology called a TH Express-2 interconnect network. It also runs a Chinese-made variant of Linux known as Kylin as its operating system.

Keep cool, America, you’re still holding second and third place. Titan, a previous world champ running at the Oak Ridge National Laboratory in Tennessee, held on for second. Sequoia, a machine at the Lawrence Livermore National Lab in California and also a previous number one, came in third.

Machines in the U.S. accounted for five of the top 10, and 233 of the top 500, making it by that metric the country with the most supercomputing muscle in the world, if not the brawniest single machine. And while the U.S. still dominates the list, it did so by the smallest margin since the list was first collected in 1993.

Supercomputers in European countries collectively came in second place, with 141 machines on the list. Japan qualified 39, an increase from 32 systems six months ago. China had 37 systems on the list, down from 61 six months ago, but in terms of raw power of its combined machines, it is the world’s second-highest consumer of supercomputing capacity.

The only new entry in the top 10 was in Saudi Arabia, with a machine called Shaheen II installed at King Abdullah University of Science and Technology, which reached 5.5 petaflops of performance. It’s the first machine from the Middle East to make it to the top 10.

The entry point to make it on the Top500 list also increased to 165 teraflops, or trillions of flops. The 500th system on the list is owned by an unnamed financial institution in the U.S.

So who builds these machines? Hewlett-Packard was the No. 1 vendor, responsible for building 178 of the systems on the list. IBM was second with 111 systems in total. Cray came in third with 71. IBM’s BlueGene/Q was the most popular supercomputing product on the Top 10.

But Cray won the list by another metric: Overall performance. The combined performance of Cray-built systems accounted for 24 percent of the total performance of the list. IBM and HP came in second and third respectively.

The one vendor that truly dominated the list was Intel: Its chips were used in 432 systems, or more than 86 percent of those on the list. IBM’s Power processors were No. 2, accounting for 38 systems. Advanced Micro Devices made the chips in 22 systems. Nvidia GPU chips, which are often combined with Intel or AMD chips to speed up the number-crunching associated with floating-point operations, were used in 52 systems. AMD’s Radeon GPU chips were used for the same reason in four systems.

China’s lead over the U.S. is likely secure for the next year or two. The U.S. Department of Energy has a pair of new systems on the drawing board, one of which, dubbed Summit, stands a reasonable chance of retaking the world supercomputing crown. When operational sometime in 2017, it will supposedly be about nine times more powerful than Tianhe-2, reaching about 300 petaflops, which is pretty close to the 363 petaflops that you’d get if you combined all 500 systems on the current list.

That sounds like a pretty impressive leap forward for the long march of computing power overall. But consider this long-term trend: Between 1994 and 2008, performance of the No. 500 machine on the list grew at an annual rate of 90 percent. Since 2008, it has grown by only 55 percent.

This article originally appeared on Recode.net.

Will you support Vox’s explanatory journalism?

Most news outlets make their money through advertising or subscriptions. But when it comes to what we’re trying to do at Vox, there are a couple of big issues with relying on ads and subscriptions to keep the lights on:

First, advertising dollars go up and down with the economy. We often only know a few months out what our advertising revenue will be, which makes it hard to plan ahead.

Second, we’re not in the subscriptions business. Vox is here to help everyone understand the complex issues shaping the world — not just the people who can afford to pay for a subscription. We believe that’s an important part of building a more equal society. And we can’t do that if we have a paywall.

So even though advertising is still our biggest source of revenue, we also seek grants and reader support. (And no matter how our work is funded, we have strict guidelines on editorial independence.)

If you also believe that everyone deserves access to trusted high-quality information, will you make a gift to Vox today? Any amount helps.

One-Time Monthly Annual

$5/month

$10/month

$25/month

$50/month

Other

$

Yes, I'll give $5/month

Yes, I'll give $5/month

We accept credit card, Apple Pay, and Google Pay. You can also contribute via