It takes about 1,000 times more energy to move a data byte around than it does to do a computation with it once it arrives. What is more, the time taken to complete a computation is currently limited by how long it takes to do the moving.
Air cooling can go some way to removing this heat, which is why many desktop computers have fans inside. But a given volume of water can hold 4,000 times more waste heat than air.
Until recently, the supercomputer at the top of that list could do about 770 million computational operations at a cost of one watt of power. The Aquasar prototype clocked up nearly half again as much, at 1.1 billion operations. Now the task is to shrink it.
Japan's next-generation Supercomputer (under construction and expected operation in 2012) will use the ultra-high-speed SPARC64™ VIIIfx processor developed by Fujitsu. Each of these processors possesses a computational performance of 128 gigaflops, and has a degree of reliability inherited from Fujitsu's mainframe technology. The CPUs are also highly energy efficient, with a world-class processing power of 2.2 gigaflops per watt, a reduction of power consumption by 2/3 compared to previous levels.
Nvidia has a roadmap for its Maxwell GPGPU chip to get to 15 gigaflops per watt in 2013
DARPA is funding about $100 million so far for its exascale supercomputer projects and is targeting at least 50 gigaflops per watt.
There is an issue that the US approach to supercomputers has come to a dead end and that reinvention of computing is needed to push to the exascale and beyond.
If you liked this article, please give it a quick review on ycombinator, or Reddit, or StumbleUpon. Thanks
Ocean Floor Gold and Copper
Ocean Floor Mining Company