The Top 5 Supercomputers More Power Hungry Than Ever
As expected, Japans K Computer has extended its lead in the prestigious Top500 Supercomputer list.
The updated K system now has 705,024 processing cores, delivers 10.5 PFlops and consumes about 12.7 MW. For the first time, the list also includes a supercomputer that integrates a Chinese processor architecture.
The current Top500 list now ranks four Asian systems among the five fastest supercomputers in the world. NUDT YH MPP follows the K Computer with a performance of 2.6 PFlops. A 1.8 PFlops computer at the Oak Ridge National Laboratory is in third place, followed by a Dawning TC3600 system in China (1.3 PFlops) and a HP ProLiant SL390s supercomputer (1.2 PFlops) at Tokyo's GSIC Center. Despite greater power efficiencies in microprocessors, the overall power consumption is expanding at a rapid pace: K Computer consumes 12.7MW. The five fastest supercomputers are estimated at a consumption of 27.3 MW, up from 14.8 MW just three years ago.
The November 2011 list also includes, for the first time, a computer that uses China's Shenwei SW1600 CPU. Clocked at 975 MHz, the processor has 16 cores and debuts in a system installed at the National Supercomputing Center in Jinan. The system is ranked at position 14.

kilowatt=thousand
megawatt=million
http://en.wikipedia.org/wiki/Megawatt#Megawatt
If they had a x86 license and made said architecture CPU's, I wonder how they would perform, a.k.a price/performance.
Was I the only one who asked themselves that question?
On topic now..
Damn that's a lot of power, like a power station dedicated just to the top supercomputer. :\ I wonder if indeed they did dedicate one to it.
You know this isn't talking about enthusiast PCs right?
The average i7 PC is putting out around 60-70 GFlops, which is about a million times less than a PFlop...
(1 PFlop = 1000000 GFlop)
12.7 mega(yes mega)watts actually seems a bit low for 10.5 PFlops, but maybe I'm missing something
Well, this does seem more like an Enthusiast PC, only it's a very enthusiastic corp.
They alone should be held accountable for such a massive waste of electricity and it is a waste if the rigs can't use up their PCIe Bandwidth.....
For each 1$ spent on computing for research the gain is about 10 times more saved by the results of the research.
@Pasoleatis if only that where true. I would love to see the actual figures for that. I've been a long term folder but after a few years I've came to wonder is such projects actually producing valuable results. Not that I expect to wake up and suddenly x, y , z is cured but has folding actually improved understanding in those areas? F1 is a prime example of how moving to modelling systems can't produce the results of actually getting a car out on track and physically running the parts.
for simulating directors and ceos (people with money) pr0n!
Boards, RAMs, etc. So it's actually less than 18W per core
You do realise that this particular supercomputer was the 6th most energy efficient as of June 2011? It's probably shuffled a few places now as it is using ~30% more power, but efficiency is about performance/watt, not just how much power is consumed.
I didn't ask. I looked.
The SW1600 looks like it is based on the Alpha 21164
http://laotsao.wordpress.com/2011/10/29/sw1600-and-alpha-21164/
which begs the question: did the Red Chinese license or pirate the technology?
What the results would be and what researchers would find.
It also be interesting if they used all that computer power to make a crazy A.I system for a video game.
I work at a commercial nuclear plant. We have 2 units. Each unit produces just under 1250 MW, for just under 2500 MW total put on the grid. Coal plants range from 500 MW to thousands of MW's also, depending on how many boilers they have.