We occasionally discuss photonic computers as news is announced, because we're starting to reach “can count the number of atoms with fingers and toes” sizes of features. For instance, we reported on a chip made by University of Colorado Boulder and UC Berkeley that had both electric and photonic integrated circuits on it.
This announcement from Optalysys is completely different.
The Optalysys GENESYS is a PCIe add-in board that is designed to accelerate certain tasks. For instance, light is fourier transformed when it passes through a lens, and reverse fourier transformed when it is refocused by a second lens. When I was taking fourth-year optics back in 2009, our professor mentioned that scientists used this trick to solve fourier transforms by flashing light through a 2D pattern, passing through a lens, and being projected upon film. This image was measured pixel by pixel, with each intensity corresponding to the 2D fourier transform's value of the original pattern. Fourier transforms are long processes to solve algebraically, especially without modern computers, so this was a huge win; you're solving a 2D grid of values in a single step.
These are the sort of tricks that the Optalysys GENESYS claims to use. They claim that this will speed up matrix multiplications, convolutions (fourier transforms — see previous paragraph), and pattern recognition (such as for DNA sequencing). Matrix multiplications is a bit surprising to me, because it's not immediately clear how you can abuse light dynamics to calculate this, but someone who has more experience in this field will probably say “Scott, you dummy, we've been doing this since the 1800s” or something.
Image Credit: Tom Roelandts
The circles of the filter (center) correspond to the frequencies it blocks or permits.
The frequencies correspond to how quick an image changes.
This is often used for noise reduction or edge detection, but it's just a filter in fourier space.
You could place it between two lenses to modify the image in that way.
From a performance standpoint, their “first demonstrator system” operated at 20Hz with 500×500 resolution. However, their video claims they expect to have a “PetaFLOP-equivalent co-processor” by the end of the 2017. For comparison, modern GPUs are just barely in the 10s of TeraFLOPs, but that's about as useful as comparing a CPU core to a digital signal processor (DSP). (I'm not saying this is analogous to a DSP, but performance comparisons are about as useful.)
Optalysys expects to have a 1 PetaFLOP co-processor available by the end of the year.
Nice, funny tho how this
Nice, funny tho how this feels like analog brought back to the digital world
It probably is analog. Then
It probably is analog. Then again, so is MLC (and higher) flash, just rounded into states.
But yeah, intensity will probably be converted from analog to floating-point decimal.
Please elaborate on how that
Please elaborate on how that works? Math is not my area.
I’m not sure what you’re
I'm not sure what you're referring to (with "that") specifically. If it's what I think it is, then voltage in the real world is continuous, whether it's in a NAND memory cell or on a pixel in a camera's sensor. When it's converted to digital, we then bracket it into ranges based on the precision of the tool that we can measure it by.
The world is analog, an in
The world is analog, an in PC’s we try very hard to forget that, with obvious advantages. My poorly made point was that, these guys are actually applying analog algorithms, which strikes me as both innovative and old fashioned at the same time. I like they way they are going and hope they succeed.
Stagnating Moore’s law creates a bit of space for these more niche tools which is great.
Analog is not “old
Analog is not “old fashioned”.
Not when you think about the atom as a computer that has some analog and digital like aspects to the various electron and proton, neutron/other particle interactions. Atoms interact with electromagnetic radiation and throw off light at defined wavelengths. Just modeling a single hydrogen atom requires some supercomputer assistance. So the atom as an analog and digital like(quantum) computer. Analog computers can be much more efficient than any digital version and produce at analog fidelity something that digital to analog loses in the conversion process.
No transmissions over the “airwaves” are purely digital to begin with.
I’m sorry you confuse me,
I’m sorry you confuse me,
There is a third image in
There is a third image in that “Tom Roelandts” underlined hot linked article, and more information about applying convolution filters and reversing the process.
This linked to article(1) is titled “Why is Deconvolution Difficult?”. And journalists should not link to an article/reference with an untitled refrence by the Author’s name as simply using the Author’s name as a hotlink is really not good from an academic/journalistic standard point! Links to articles/Refrences need to have the full article’s title and the Author’s name before any hotlinking. It’s the Article that is refrenced and not just the Author of that article “Tom Roelandts” it’s his article by Title and Name that must be included so the reader can see what subject matter that this Untited link “Tom Roelandts” actually concerns, which is: “Why is Deconvolution Difficult?”.
(1)
“Why is Deconvolution Difficult?”
https://tomroelandts.com/articles/why-is-deconvolution-difficult
Also: Performimg optical
Also: Performimg optical operations/calculations using light quanta requires a whole lot less power and has a whole lot less leakage/waste than using transistors, and the electromagnetic quanta frequencies can go way up there as opposed to pure electronic circuit frequencies(oh the leakage and heat).
This Video has more info:
This Video has more info:
“Optalysys Optical Computing – Explained by Professor Heinz Wolff”
https://www.youtube.com/watch?v=T2yQ9xFshuc
P.S. This link is the much
P.S. This link is the much more the Definitive Explanation by the good Optalysys PHD himself, as opposed to that Marketing Video presented in PCPer article! Marketing + Technology discourse equals confusion!
Ah, I see. Yeah.
Ah, I see. Yeah.
Don’t worry Jann5s, I got
Don’t worry Jann5s, I got your point. And you’re right, it is an interesting application of analog computing. Modern computing has become dominated by digital computing, and seeing it go back to analog in this case is really interesting.
https://en.wikipedia.org/wiki/Analog_computer
>4-slots
>4-slots bracket
Tsk-tsk-tsk…
It’s 4 slots to fit the
It’s 4 slots to fit the optics, not necessarily because of cooling.
Where did I say anything
Where did I say anything about “cooling”?
It strikes me that, for a
It strikes me that, for a large enough matrix, you can just convert values into a wave on a 2D surface (basically, any digital image is a matrix… so any matrix is a digital image). I wonder if they are using that as a processing trick?
Fourier Analysis With A
Fourier Analysis With A 100-year-old Mechanical Computer
http://www.synthtopia.com/content/2014/11/17/fourier-analysis-with-a-100-year-old-mechanical-computer/
Harmonic analyser analog to square wave conversion Mechanical Computer!
Video 3 of 4
https://www.youtube.com/watch?v=9vawtvAUnN4
Scott please ask them about
Scott please ask them about any potential for Ray Tracing and Ray interaction calculation acceleration on this device. Obviously it using electromagnetic Rays for computation! So I’d imagine it may be useful for accelerating Ray Tracing workloads, maybe more so in real time compared to purely digital methods.
P.S. be careful with using
P.S. be careful with using that picture of Lena, it is officially owned by playboy and thus may infringe some (copy) rights. They condone usage for science applications but I’m not clear about the limits.
Its fair use, and not the
Its fair use, and not the objectional parts of the complete image! It’s used to show an image effect and that’s it. The subject matter may be debatable but the copyright question is answered.
“Image Credit: Tom Roelandts”
“Image Credit: Tom Roelandts” is a misattribution of image source, though the Tom Roelandts Graphic/Example that utilizes the Image may actually Tom Roelandts! But Tom Roelandts actually links in his article, linked to by Scott, to another of his(Tom Roelandts) article with a proper attribution of source for the image used, famous Lena image(1).
Hopefully Getty Images will not attempt any more foolishness and try to sue everyone on the planet over images they do not actually own the rights to!
As this article by Tom Roelandts explains the famous Lena iamge:
(1)
“Lena”
https://tomroelandts.com/articles/lena
I can’t wait for them to get
I can’t wait for them to get this down to the isolinear chip size or isolinear rod size, and maybe some semblance(visually only) to the holodeck can be achieved!
Here is a list of CPU’s for
Here is a list of CPU’s for comparison with the 1 PetaFLOPS Optalysys co-processor for computing power – for example, a Pentium i7 has less than 100 GigaFLOPS:
http://www.overclock.net/t/947312/how-many-gflops-does-your-processor-have
In 2020 Optalysys plan to have a processor that is at least 1 ExaFLOPS.
What I wonder is whether these PCIe cards could be inserted into robots. I have been reading “The Second Intelligent Species” by Marshall Brain, creator of the How Stuff Works site, and he talks about robots being able to see and then they will be able to perform physical tasks that we now do so intuitively and easily, like cleaning out washrooms and tossing burgers. I have checked, these cards will be able to be used for pattern recognition.
Does anyone know if these PCIe cards will be capable of being used in robots for machine learning and deep learning, perhaps in tandem with digital processing? Will they advance the introduction of robots capable of human physical work by decades?
I am hoping that they will, and with the Sun Cell also being manufactured, currently a prototype, by Brilliant Light Power in 2017, robots will have an uninterrupted portable inexhaustible and powerful energy supply. (See the seven videos of “Industry Day” on 26th October 2016 on You Tube or on http://brilliantlightpower.com )
Comparing the FLOPS value is
Comparing the FLOPS value is a little dodgy, as this is a single purpose processor, rather than a general purpose processor. Their website also seems fairly light (pun not intended) on what their coprocessor actually DOES: is it an accelerator specifically for the Fourier Transform? Does it just do correlation and autocorrelation (the ‘matrix transforms’ they are probably referring to is correlation using a kernel, such as the one shown in the image)? Does it contain function blocks to do both, or must you use an outboard FFT?