It seems likely that the National Security Agency is spending about ten times more money and using a generator that is ten times larger to bring an exaFLOP supercomputer into existence 4-5 years earlier than a 2019/2020 expectation. $896 million and 60 megawatts instead of $100 million and 6 megawatts.
A similar increase in budget and effort and power could bring a zettFLOP supercomputer into existence 8 years earlier than the 2030 expectation. $9 billion and 600 megawatts instead of $900 million and 60 megawatts.
Is this a misallocation of resources and effort ?
If it was a one and done effort probably. However, if it was part of a long term strategy to step up to greater developments in high performance computing then it could be justified.
The grand supercomputer that is at the top of the heap is like an aircraft carrier that needs support ships. A supercomputer needs to be surrounded by smaller mini-supers so that algorithms and programs can be tested and developed. By stepping up a couple of orders of magnitude on a sustained basis then a nation could achieve certain advantages over nations that were less committed to achieving computational advantages. There would need to be success in driving economic growth or military advantages that offset the costs.
The $9 billion supercomputer could be compared to other large science and technology projects like particle colliders and large space telescopes. The mastery of successfully delivering supercomputer projects on that scale could accelerate scientific development and discovery.
The superlarge supercomputer strategy needs to be compared against against buying more smaller supercomputers on a frequent basis and constantly refreshing them with the latest technology.
A larger commitment to bigger supercomputer projects could also be a bet that the transition to the post-CMOS computing world will have delays. Since we are nearing the end of non-heroic computing architecture improvements, it could be a good strategy to ramp up over the next 1000X and 1000000X improvements. When the wall is hit the lead country could then have an extra 8 to 12 years at the plateau of performance. The lead country could also have extra resources and capabilities for the transition to the new computational architecture.
A Presentation from Supercomputer center that is trying to drive faster economic development
However, the NSA having a bigger supercomputer for code breaking does not translate to efforts to specifically use supercomputers to drive economic growth.
You could use more powerful supercomputers to find ways to improve logistics in an economy. There are other projects and applications. Applications and issues related to zettaFLOP supercomputers are discussed here
Zettascale and Exascale sample Applications
* Climate modeling.
* Biology: spread/analysis of diseases; phylogenetics.
* Nano/materials: simulate from first principles a transistor/quantum computer.
* Chemistry: beyond Born-Oppenheimer.
* Economic/behavioral models:Baghdad in faster than real-time.
* Cognitive models of brain: vision; mapping out functions.
* Medical applications: personalized medicine.
* Engineering: crash modeling.
* Decision support.
* Calibration of large instruments (LHC, LIGO).
* HEP and nuclear physics.
* Real-time weather: storms, tornados, icing; disaster planning.
* Nuclear energy: fission; fusion.
* Stockpile stewardship (ASC).
* Oil&gas exploration.
* Earth tomography.
* Math research.