By 2020, you could have an exascale speed-of-light optical computer on your desk

Optalysys, a UK technology company,  says it’s on-target to demonstrate a  novel optical computer, which performs  calculations at the speed of light, in  January 2015. If all goes to plan,  Optalysys says its tech — which is really  unlike anything you’ve ever heard of  before — can put an exascale  supercomputer on your desk by 2020.
When we talk about optical computing ,  we’re actually referring to a fairly large  number of different and competing  technologies. At its most basic, optical  computing refers to computing that uses light instead of electricity. When we’ve  previously written about optical  computing, we’re usually referring to  chips and computers that have replaced  their internal wiring with optical  waveguides, and some kind of optical  transistor that is controlled by photons  instead of electrons. There are also  optoelectronic devices, which use a mix  of the two (usually optical interconnects  and electronic transistors).
In the case of Optalysys, optical  computing is something else entirely. At  this point, because the Optalysys tech is  rather complex, you should probably  watch the video embedded below — not only will it probably do a better job than  me at explaining it, but it’s also narrated  by the adorable Heinz Wolff. If you can’t  watch the video, read on and I’ll try my  best.
It goes something like this. You start  with a low-power laser. This laser is then  directed through a massive liquid crystal  grid. This grid works in much the same  way as a liquid crystal display. By  applying electricity to each “pixel,” the  laser light passing through it is affected.  Complex calculations would turn  hundreds or thousands of these pixels  on or off. After the laser has passed  through this grid, the beam is picked up  by a receiver. By analyzing the beam’s  diffraction and Fourier optics, matrix  multiplication and Fourier transforms can be combined to perform complex maths. You can also have multiple pixel grids in  sequence or parallel, significantly  boosting the complexity and parallelism  of the optical computer. There’s a little  more technical info on the Optalysys  website , but not much.
Moving away from the technical nitty- gritty, Optalysys’s optical computer is  exciting for two main reasons: It  consumes very little power, and there’s  essentially no limit on how parallel you  can make it. There’s no direct analogy to transistor-based logic, but you could  almost think of every liquid-crystal pixel  as a tiny processing core (or at least a  tiny transistor). In a normal computer  chip, while there is some parallelism,  most things happen very sequentially,  with each core (and each transistor)  working mostly in serial. In an Optalysis  optical computer, the laser beam hits  every single pixel at the same time — it  essentially performs hundreds or  thousands (or millions?) of small  computations in parallel, at the speed of  light.
An Optalysys optical computer, on a  desktop
In terms of power consumption, HPCwire says the estimated running costs of an  Optalysys computer would be in the  region of $3,500 per year. I think this is  based on the assertion by Optalysys that it will be able to deliver a desktop-size  computer with exascale performance.  Compare this to the world’s fastest  supercomputer, which would cost  around $21 million in energy costs if it  was run at peak (~34 petaflops)  performance for a year.  [Read: HP bets it all on The Machine, a new computer  architecture based on memristors and  silicon photonics .]
Optalysys says that its technology is  already at NASA Technology Readiness  Level (TRL) 4, meaning it’s ready for  full-scale testing in a laboratory  environment. By January 2015, Optalysys says it will have a 340-gigaflop prototype ready to go. By 2017, the company  wants to have two commercial systems  in place: A big data analysis system that  will add 1.32 petaflops of grunt to an  existing, conventional supercomputer —  and a standalone Optical Solver  supercomputer, which will start at 9  petaflops. Optalysys thinks its Optical  Solver could scale up to 17.1 exaflops by  2020. This seems like a very bold  statement for an entirely novel and  untested method of computing — but  given how conventional computing has  mostly stagnated by this point , I hope  the folks at Optalysys can follow  through.
source : Extremetech.com