Boffins demo 'memcomputer', plot von Neumann's retirement

Re: I slipped on muh snake oil

"but lady nature doesn't like her no-go areas lifted"

Thing is though, we don't know where the hard limits are. The history of science and technology is littered with definitive statements about what is and is not achievable. Most turn out to be wrong.

Re: I slipped on muh snake oil

Ye cannae change the laws of physics Jim

Re: I slipped on muh snake oil

Reading between the lines, I think that's probably "in theory, ignoring noise".

Allowing for noise, either this piece of magic will fail rendering the whole thing useless, or assuming that the boffins have more of a clue than the journalists, you'll be running a sort of physically accellerated annealing process that will give you an answer that's pretty close to THE answer to the NP-complete problem. However, it won't necessarily be that answer, and you may never be able to know whether it is or not. For a travelling salesman problem (route optimisation), that doesn't matter. Any close approximation is equally useful. For other (crypto?) problems, that most certainly does.

It's almost certainly a good idea to build one and study its operation. As with quantum computation, detailed information on what can and can't be achieved may lead to the next breakthrough in our understanding of our universe.

Re: a Fourier transform in real time but it couldn't run a simple spreadsheet with ...

Your requirements might be different, of course, but my use for fourier transforms is much greater than that for spreadsheets (although, to be fair, I did use a spreadsheet about three years ago).

It can be quite useful to dig up old ideas and see how much further you can push them with modern technology. Perhaps we should wait and see how far they can get before turning on the full-strength pessimism.

Re: Given the example shown:

That's right. The results are approximate and limited by the noise floor; this is not a general-purpose computer in the regular sense. Not to say that getting approximate answers quickly to things like Fourier transforms isn't useful; but getting answers to discrete problems, like integer factorisation, do not suit this architecture well.

For those not old enough to remember analogue computers:

http://blog.analogmachine.org/2012/03/15/analog-computers/

I once had an edition of a 1970's electronics magazine which had a project to build an analogue computer lunar lander game. Integrators were used to store your altitude, velocity and fuel remaining, and you controlled the rate of burn with a potentiometer.

@Paul A. Clayton

AFAIK whether or not NP-complete problems admit of solution in polynomial time remains an unsolved problem itself. I believe the weight of opinion says not, but that scarcely constitutes proof in the field of computability.

Re: @dESTROY aLL mONSTERS

Its not scalable or you do not understand or know how to make it scalable? They used to say that about valve architectures, then silicon came along. I don't think we are talking about fundamentals of the universe here but limitations in materials, fabrication and programming added to architectures.

Computers and computing is not some fundamental property of the universe unless you get really arcane in information theory that both leaves practicality behind and suffers from poverty of assumptions.

We also have and do have dedicated units in our computers. What is a graphics card but an optimised sub motherboard dedicated to one task? I remember the Mac SE 30 that had a maths co-processor added to the chip. I was doing 3D reconstruction (stacked dinnerplates, not solids) of mouse legs and the difference from a computer without vs with was in the former: input x.y,z go away and do something else, maybe get yet another coffee while you wait; the latter watch it in real time and cancel if you get the x,y.z wrong and go again.

So I don't see this as a motherboard architecture, I see this as a dedicated add-on unit for speeding up particular tasks or doing them more efficiently and thus freeing the CPU for other tasks. Imagine how much slower your device would be if the CPU was handling all the graphics.

Re: @dESTROY aLL mONSTERS

A scalable method of solving any NP-complete problem in polynomial time / space would take out most forms of cryptography

I don't think so - only the asymmetric-key ones, and other primitives that rely on trapdoor functions. I suppose you could argue that constitutes a majority of the "forms of cryptography", depending on how you count them, but "most" seems like a stretch.

Or do you know of an argument that if P=NP then symmetric cryptography is doomed? I'd be interested to hear it.

Re: P != NP....

"Thank you for dredging that up - it is an accessible explanation."

It is in my opinion a *very* accessible explanation, needing not too much more than an understanding of superposition and Fourier transforms.

In particular (if I've understood correctly) the point that to get a Fourier transform of infinite resolution will need an infinite length (ie infinite time) sample. Less precision permits shorter run times. Transfer my wages to an account numbered in the region of 08-00-28 please. Oh hang on, that's pretty useless as a concept. Which may slow matters down somewhat.

Why does anyone attempt to claim success with stuff like this?

It's not like the Italian "faster than light" stuff where they said "these are our measurements, what are we missing?" and ended up being somewhat misreported.