Thinking Tech

More speed, fewer watts: the next race in supercomputing

More speed, fewer watts: the next race in supercomputing

Posting in Architecture

Homegrown components will make China's Dawning 6000 unique, and more importantly extremely efficient. Is this the opening salvo in a new supercomputing war?

The Dawning 6000 will be the first truly Chinese supercomputer, in that its core components--thousands of processors--will be designed and built in China. This processor, called the Loongson Godson-3b, is based on an unusual architecture and is expected to be relatively cheap. What makes it truly interesting, though, is the fact that it's coming from a company other than Intel, AMD or IBM.

The successful completion of a top-500 supercomputer with homegrown Chinese technology would go a long way toward changing China's international image as a mere manufacturer of technology developed elsewhere. But according to a report on HPCwire (via Technology Review), it could also signal a new direction for supercomputing as a whole:

As one might expect from a MIPS architecture, the Godson delivers outstanding performance per watt numbers. The new Godson-3B achieves 128 gigaflops with a power-sipping 40 watts. The relatively slow clock speed (1.05 GHz) is the key to the low-energy use... [T]he Godson-3B appears to be a very power-efficient design, and the upcoming Dawning machine could rival even Blue Gene/Q systems for performance per watt supremacy.

Performance per watt considerations are hugely and obviously important for battery-powered devices like cellphones and laptops. Supercomputers don't operate from batteries, but they do consume tremendous amounts of electricity, which in turn cost their operators quite a bit of money. (Not to mention the implicit environmental concerns.)

So, let's assume that these new Loongson processors are less powerful than their Intel or AMD counterparts--a safe assumption, according to prior reports. If the processors are intended solely for supercomputing, this may not matter. If two Loongson processors can do the work of one Intel processor using less energy, the Loongson processors are a better option in most cases, assuming their initial cost isn't too high, and their energy advantages are more than negligible. In other words, what matters in supercomputing isn't how much processing a single CPU can carry out; it's the cost of that work.

Last November I spoke with Mark Barnell, the Air Force Research Laboratory high performance computing director, about the future of supercomputing. He explained that the future of supercomputing will probably be a bit more frugal, both in terms of component cost and energy use:

We see that a lot of resource centers have the same problems: cost, size, power consumption, heat. What will up happening in the next few years is that with commodity products like this, you’ll be able to get a lot more power and computational throughput without increasing power demands. That’s something we’re paying a lot of attention to.

The processors in our iPhones and our Androids have the potential to be the next game-changer a few years down the road.

The Dawning 6000 won't break any records for processing power, nor is it likely to blow other supercomputing systems away in terms of efficiency--though that remains to be seen.

What it will do is serve as an exemplar of the next era in supercomputing, where success isn't just measured in petaflops.

Pictured: An older Loongson processor, the 450MHz 2C.

Share this

John Herrman

Contributing Editor

Contributing Editor John Herrman is a freelance writer based in New York City. He is also contributing editor at Gizmodo. He holds a degree from the University of Edinburgh. Follow him on Twitter. Disclosure