Slashdot Mirror
'Inexact' Chips Save Power By Fudging the Math
Barence writes "Computer scientists have unveiled a computer chip that turns traditional thinking about mathematical accuracy on its head by fudging calculations. The concept works by allowing processing components — such as hardware for adding and multiplying numbers — to make a few mistakes, which means they are not working as hard, and so use less power and get through tasks more quickly. The Rice University researchers say prototypes are 15 times more efficient and could be used in some applications without having a negative effect."
http://tech.slashdot.org/story/09/02/08/1716235/sacrificing-accuracy-for-speed-and-efficiency-in-processors
Of course, you might've been sacrificing speed for accuracy in that 3 year estimate.
(and for all of the nay sayers -- I could see this being great for monte carlo simulations or other modeling where you're dealing with so much imprecise inputs that minor error's not going to be significant)
Build it, and they will come^Hplain.
I haven't RTFA yet, but I strongly suspect that there would be different instructions when accuracy matters (ie program flow control), from where it's not as important (ie signal processing).
I have a MS in AI, and I'm working on a PhD (although not in the imprecise math area). Let me see if I can explain using as few points as necessary:
1 - We are interested in machine which make decisions
2 - Machines can only perform several _very_ basic operations (add, subtract, multiply, divide, move to/from memory)
3 - Probabilistic models are used to compare the relative 'goodness' of several decisions in order to determine the best
3.1 - statistical models, bayesian networks, voting systems, complex adaptive systems, neural network models, etc.
4 - Because of (1), we are interested less in the _model_ than the _decision_. When deciding among 3 decisions ranked 3.43122323, 4.12312455, and 1.85730385, we only care that 4>3, but had to take all of the factors into account to determine this
4.1 - Frequently, you decide between 20+ alternatives, so you don't even care if you make the second-, or third-best decision
5 - Among the limiting functions to making a cyborg is the large power requirement.
6 - If you can do imprecise math, and result in a reasonable decision at 1/15th the power, you could come closer to cyborg-AI.
If someone is doing structural engineering they are already aware of how much precision they actually need, and probably are not going to be reusing some 'hobby' application to do those calculations... crow, they probably are not even going to use one of the common languages like C/C++ since floating point operations in them are already unpredictable past a certain point (the chips will do the work to great precision, but the language is sloppy)... if they REALLY need the precision they will probably use specialized libs or a more audit-able language like Ada or FORTRAN.