SoylentNews is people
SoylentNews
As the cost of sequencing a person's genome plummets, demand for the computing power needed to make sense of the genetic information is growing. Nicholas New hopes some of it will be worked on by a processor that parses data using laser light, built by his U.K. startup, Optalysys.
New says his company's exotic approach to crunching data can dramatically upgrade a conventional computer by taking over some of the most demanding work in applications such as genomics and simulating weather. "The grunt work can be done by the optics," he says.
Researchers have worked on the idea of using optics rather than electronics to process data for decades, with little commercial traction. But New argues that it is now needed because manufacturers such as Intel admit that they can't keep improving conventional chips at the pace they used to
takyon: Optalysys is a company that claims it will scale to petaflops and exaflops in the near future using desktop amounts of power. However, their technology has been described as a coprocessor so it isn't clear what kind of operations it can perform. Much has been made of its partnership with The Genome Analysis Center (TGAC).
The article also mentions LightOn, which published this paper. Optical and quantum computing are both being examined for their potential to enhance machine learning. Google is researching the quantum approach.
(Score: 0) by Anonymous Coward on Friday November 04 2016, @10:34PM
But if you could just make matrix multiply about 1e3 times faster, rather than some obscure quirky operation, that'd be nice.
(Score: 1) by Ethanol-fueled on Saturday November 05 2016, @12:31AM
C on Windows 98 [numerical.recipes] can already do that compared to whatever Win10 shitbox you're running now, can.
(Score: 2) by tynin on Friday November 04 2016, @10:37PM
If you need to open it up for maintenance, I imagine even in a clean room, you'd invite some contamination into its innards which would lead to computational errors due to messing with the optics. Perhaps you could make modular parts and have them forever sealed and if part breaks, swap out the entire sealed portion.
Could a million sharks with lasers on their head reproduce Shakespeare?
(Score: 2) by takyon on Friday November 04 2016, @10:46PM
That's interesting. Don't smudge the computational LCD screen!
In the initial videos of this thing, it looks like it was designed to be able to sit around like a desktop. But there just isn't enough information to evaluate the effectiveness of Optalysys and its claims. Hopefully we will know more and see some shipping products by next year.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 1) by Ethanol-fueled on Saturday November 05 2016, @12:01AM
It looks rather gimmicky. And until we move to all-optical architectures (fat chance), then we're still left at the mercy of the O-E converters. I was horrified to read in the article that the light was being read by a "camera sensor," because even traditional laser diodes (GaN or the GaAs family) are capable of high-speed multiplexing [wikipedia.org] and have been for quite awhile.
What I'm getting out of this article is that this device is comparing snapshots, or something. But hopefully somebody could explain it better.
(Score: 3, Informative) by takyon on Saturday November 05 2016, @01:58AM
There's no point of even thinking about it right now, which is why I've mentioned Optalysys in a few comments but never subbed a story yet. They claimed that they could have a 9 petaflops machine ready to be sold by 2017 and 17.1 exaflops by 2020. Even if they are a couple of orders of magnitude off, it would still be a vast improvement in FLOPS/W. Until then, it's vaporware.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 2) by opinionated_science on Saturday November 05 2016, @11:09AM
Well if they can so sub 1us fft's , I have a problem for them....;-)
To be precsise, 128x128x128 in less than 1us would be quite an advance, assuming:
a) they can do convolution too
b) reverse the result.
c) the result could be distributed to the rest of the calculation.
Otherwise I'll wait and see...though I might go see them in they come to Utah in two weeks...