Aiming a tenfold increase in computing power using the same power consumption, Japan is aiming to put their next supercomputer online by 2017.
Japan's planned supercomputer has a processing capacity of 130 petaflops outperforming the current leader in China's Sunway Taihu Light that delivers 93 petaflops per second. By building this system, it will catapult Japan to the top of the world's computer rankings by year-end 2017, PCWorld reported.
If plans push through, the project would cost a reported 19.5 billion yen or $173 million.
Satoshi Sekuguci, director general at Japan's National Institute of Advanced Industrial Science and Technology (AIST), said that as far as they know, there is nothing out there that is as fast, referring to their planned system. He adds saying it leaves to the imagination if there are secret hidden supercomputers out there plowing through data in hollowed-out mountains, according to Gizmondo.
Aside from aiming to create the fastest supercomputer, Japan also aims to make the system efficient. They aim for power consumption fewer than 3 megawatts. Their older Oakfest-PACS running at 13.6 petaflops consumes the same amount of power. China's TaihuLight consumes over 15 megawatts.
AIST plans to use liquid cooling for their system similar to that used by the French Atos supercomputer used by the French Alternative Energies and Atomic Energy Commission (CEA). Atos system plans to have a performance power of one exaflop, but the system will not be ready to go online until 2020.
AIST plans to use their supercomputer for deep learning applications directly competing that of Google's DeepMind AI program. The yet unbuilt supercomputer will be called ABCI, an acronym for AI Bridging Cloud Infrastructure.
As of present time, Japanese corporations had to rely on outsourcing-intensive processing needs to companies such as Microsoft and Google. AIST intends to build ABCI at the University of Tokyo's Kashiwa Campus. Bidding for the project has already commenced and will end on Dec. 8, with work expected to commence early 2017.
