[Beowulf] DARPA issues 20 MUSD grant to nVidia to go from 1 GFLOPS/Watt to 75 GFLOPS/Watt
Lux, Jim (337C)
james.p.lux at jpl.nasa.gov
Tue Dec 18 08:16:42 PST 2012
>>7 nm gives a factor 16 boost over 28 nm, in theory. So the derived truth from the article points me to double precision.
I don't think there's a heck of a lot of "truth" to be derived from that short article. The DARPA MTO site gives more meaningful details on what they're looking for.
$20M over 5 years (from the article) is $4M/year, which is enough to fund a small team (maybe a dozen people) to do some interesting research for a few years, depending on how much hardware they're buying. If it stops being interesting, the funding probably evaporates. That is, the funding document probably says the maximum possible award is $20M, it's certainly not a firm fixed price, but is probably a Cost Plus Award Fee or Incentive Fee type contract, with periodic deliverables (annual reports), subject to renewal each year during the 5 years.
Bear in mind that DARPA program managers (the guy or gal who ultimately monitors the contract and has a lot of influence over deciding what is "interesting" at any given time) rotate roughly every 3 years, and in any case, things change as far as what DARPA wants(or more properly, what DARPA thinks DoD needs). A few years ago, there was lots of interest in detecting, disabling, etc. IEDs... not so much any more (it's a very hard problem to solve, and has been for decades, and with the change in casualty rates/causes over the past few years, it's kind of dropped off the priority list) (for an excellent summary of the big picture problem, Augustine's paper "Land Warfare" from 70s, reprinted in IEEE AES, Sep 1986)
More information about the Beowulf