Not all supercomputers are created equal
The latter concerns were probably unfounded, since China already claims the title of the world's most powerful supercomputer. But some Reg readers may recall that regulators raised similar security issues when Lenovo bought IBM's PC business in 2005.
"I know how to make 4 horses pull a cart - I don't know how to make 1024 chickens do it." --Enrico Clementi.
We've gotten a lot better at harnessing the chickens since this statement was made some years ago. But there is still something to it.
There is much more to high performance computing than assembling massive numbers of microprocessors. This shows up on the LINPACK data on Top500.org Web Site, in terms of the fraction of the theoretical peak performance, Rpeak, that can be extracted to solve the problem (the Rmax.)
For the Japanese RIKEN machine, designed for cutting edge scientific computing, this ratio is 93%. For the Chinese Tiahnu-2, this figure is 62%, comparable to what so-called "commodity clusters," achieve. (Commodity clusters are systems assembled from mass market microprocessors and LAN interconnect technology.)
This is not do denigrate the Chinese accomplishments. Simply to point out that a presumption that they translate into technological superiority or commercial success may be unwarranted. In November 2003, the No. 3 spot on the Top 500 List was held by a commodity cluster built at Virginia Tech--the best of some 23 such home-made supercomputers.
Today, there are two, both assembled by Amazon Web Services, , , an application for which commodity clusters are arguably ideally suited.