Re:When It Comes to China, Google’s Experience Still Says It...
by walrus01 on Wednesday August 03, 2016 @01:42PM
(#12219899)
Building the world's fastest supercomputer is a financial and construction problem, not so much engineering. Assuming we are talking about a system built out of thousands of distinct nodes.
If I wanted rank #1 on the top500 list, had $900 million sitting around and wanted to build something faster, I could, by hiring some of the people responsible for building massive datacenter facilities in the US (750,000 sq ft+) with their associated cooling, generator, UPS systems, and buying several metric shitloads of 1U supermicro dual xeon systems.
All of that also assumes that it's a supercomputer designed for compute tasks that can be neatly divided up into small chunks and doesn't require a totally proprietary bespoke interconnect (rather, it can use something COTS).
It is a budget problem, not technical: Are you prepared to fork out the money for a hundred 200 kilowatt air conditioners all running in parallel? Do you have the real estate for them and their air handlers?
If I wanted rank #1 on the top500 list, had $900 million sitting around and wanted to build something faster, I could, by hiring some of the people responsible for building massive datacenter facilities in the US (750,000 sq ft+) with their associated cooling, generator, UPS systems, and buying several metric shitloads of 1U supermicro dual xeon systems.
All of that also assumes that it's a supercomputer designed for compute tasks that can be neatly divided up into small chunks and doesn't require a totally proprietary bespoke interconnect (rather, it can use something COTS).
It is a budget problem, not technical: Are you prepared to fork out the money for a hundred 200 kilowatt air conditioners all running in parallel? Do you have the real estate for them and their air handlers?