New processor computes at light speed

[Larry_g1s]

lol...I just came to post this and see if I'd beat you! Alright you when this round. lol :p

Pretty amazing wouldn't you say? But who knows how long before it's availble to the public. :annoyed:

[slaughters]

Originally posted by Juggernaut
[color=orange] lol...I just came to post this and see if I'd beat you! Alright you when this round. lol :p Ehh... Hmm..
http://www.cgtalk.com/showthread.php?s=&threadid=99277

[My Fault]

Originally posted by slaughters
Ehh... Hmm..
http://www.cgtalk.com/showthread.php?s=&threadid=99277

Roberto is a posting machine, like a force of nature he can not be stopped!

To steal a blatant quote from the Mango:
"Can you know the mighty ocean? Can you lasso a star from the sky? Can you say to a rainbow... 'Hey, stop being a rainbow for a second'? No! such is Roberto!". :p

[gmask]

I remember hearing about this research back in the early 90's.. looks like it is finally maturing! I also remember hearing about holographic harddrives.. I wonder where those are at these days?

[kwshipman]

Originally posted by gmask
I remember hearing about this research back in the early 90's.. looks like it is finally maturing! I also remember hearing about holographic harddrives.. I wonder where those are at these days?

I also remember being told that we would all be driving flying cars by the year 2000, and look where we are.:D

[Array]

Originally posted by kwshipman
I also remember being told that we would all be driving flying cars by the year 2000, and look where we are.:D

actually i've heard some people say that we dont have flying cars because we dont need them anymore. with the internet, we can virtually go anywhere and retrieve any information we need. we can even work and go shopping from the comforts of our own homes.

back to the topic of the processor though...you guys need to realize that this isnt like the processor in your computer. it's a DSP (digital signal processor). That pretty much means that it's limited in terms of the types of algorithms it can execute.

[js33]

While DSP chips are cool maybe soon we'll see a general processor built this way that would run at 8000 Ghz!

Whooo Hooooo. Hehehehe

Cheers,
JS

[easy]

Does all this means that this is as fast as chips will go. Is Enlight going to be the last fastest processor. I mean that must be the limit? Mind i guess this is all you need?

[jdj]

Originally posted by easy
Does all this means that this is as fast as chips will go. Is Enlight going to be the last fastest processor. I mean that must be the limit? Mind i guess this is all you need?

The apparent speed of a processor - wich is aggresively used in marketing today - is not really connected to the signal speed nowadays. In the early days of consumer processors (before Pentium I believe), it was though.

Earlier if you wanted a faster processor, you increased the MHz of the clock and made sure that the chip coped, but with the Pentium family and onwards they started doing things in parallell as well, and all of a sudden they could say that they had doubled the MHz etc. This is however more or less only true as long as you have ideal conditions. So the current "MHz" rating is a measurement of the throughput of the processor, not the execution speed of a single instruction.

An analogy is a comparison with a Boeing 737 and a fighter jet: The fighter jet might be faster to go from A to B, but the Boeing will get more people from A to B than the jet during the same period of time. So when it was hard to make the jets go faster, they put in more seats. :)

So - you could in theory build a "8000 GHz" processor today given these conditions, but it would truly suck at doing anything at that speed since the parallellism it needs would not exist in a general purpose environment.

Anyhow - I guess that even with optics in the chip, the engineers would find new and better algorithms to increase the throughput of the processors without actually increasing signal speed, pushing the "MHz" rating even higher with time.

But in practice I believe that you will push the bottleneck away from the processor into memory oand other surrounding devices instead. At 8000 Ghz you would need some drastic changes to the memory structure if the processor should not waste 95% of its lightning fast cycles waitning for memory response ... :/

Blah. blah. blah... ranting again!

Thats my $0.02
Cheers!
/ Daniel

[trthing]

Originally posted by gmask
I remember hearing about this research back in the early 90's.. looks like it is finally maturing! I also remember hearing about holographic harddrives.. I wonder where those are at these days?

Actually, I did research on logical optical ports (which would be the building bricks for an optical processor) in 1990: there was a special edition of IEEE magazine about them. By that time there was a lot of discussions about the path to be taken regarding the technology used to integrate the working table-size prototypes on optical ICs: some used piezoelectric materials to drive light polarization through optical fiber, others used electrically induced optical indexes changes to drive the light path and others dealt with the holographic memory/devices.
It is amazing to see that it is now been implemented, 13 years afterwards. I most certainly expect something for PCs sooner than later.:eek:

[PhilOsirus]

I'm sure we will make processors that transfer data faster than the speed of light sooner or later, but that would involve crazy intradimensional-physics-something-something:)

[jdj]

Originally posted by Phil "Osirus"
I'm sure we will make processors that transfer data faster than the speed of light sooner or later, but that would involve crazy intradimensional-physics-something-something:)

If they get these quantum computers up and running I guess they literaly will blow our minds. :surprised

/ Daniel

[trthing]

Originally posted by Phil "Osirus"
I'm sure we will make processors that transfer data faster than the speed of light sooner or later, but that would involve crazy intradimensional-physics-something-something:)

I would totally disagree: the "intradimensional" part is probably mostly settled.

Obstacles now would remain, mainly, on the engineering/manufacturing/profiting side of the project, rather then on physics/research... :thumbsup:

[NUKE-CG]

Originally posted by RobertoOrtiz
Quote:
"A superfast processor that uses light, not electrons, to perform calculations goes on sale for the first time"


Sweeet
Originally posted by RobertoOrtiz
Quote:

The device is called Enlight and can perform 8000 billion arithmetic operations per second, about 1000 times faster than a standard processor.


Sweeeet..
Originally posted by RobertoOrtiz
Quote:

He believes EnLight will be useful across a broad range of applications, from military projects to compressing high definition video images.


Sweeeeeet..
Originally posted by RobertoOrtiz
Quote:

Sariel acknowledges that Enlight "is not a general purpose processor like a Pentium". Instead, each processor will be custom-built to perform a specific set of tasks, and will not be programmable.


Noooooooooooooooo!
Ah well, it was a nice 5 minutes of me thinking I'll have one in the next 20 years ;\ :annoyed:

[trthing]

each processor will be custom-built to perform a specific set of tasks...

...like 3D rendering...

Sweet :buttrock:

[Thalaxis]

Originally posted by jdj
The apparent speed of a processor - wich is aggresively used in marketing today - is not really connected to the signal speed nowadays. In the early days of consumer processors (before Pentium I believe), it was though.


It wasn't then, and it isn't now, but we're getting close. The
signal propgation delay is non-trivial, and that's one major reason
for the optical computing research.


Earlier if you wanted a faster processor, you increased the MHz of the clock and made sure that the chip coped, but with the Pentium family and onwards they started doing things in parallell as well, and all of a sudden they could say that they had doubled the MHz etc. This is however more or less only true as long as you have ideal conditions. So the current "MHz" rating is a measurement of the throughput of the processor, not the execution speed of a single instruction.


No, the MHz rating is a measure of the number of clock ticks per
second, nothing more. It has very little to do with the processor's
throughput directly, in fact. The simplest way to explain it is that
for each clock tick, all of the execution units do their one task.
There could be a dozen execution units, there could be just two...
but it's still all timed by a single clock.


An analogy is a comparison with a Boeing 737 and a fighter jet: The fighter jet might be faster to go from A to B, but the Boeing will get more people from A to B than the jet during the same period of time. So when it was hard to make the jets go faster, they put in more seats. :)


That's a good analogy, you're just comparing the wrong
measurements. :)


So - you could in theory build a "8000 GHz" processor today given these conditions, but it would truly suck at doing anything at that speed since the parallellism it needs would not exist in a general purpose environment.


Actually, you probably couldn't; I don't think that anyone has a
transistor design that can switch at that speed on any process
anyone can fabricate today.


But in practice I believe that you will push the bottleneck away from the processor into memory oand other surrounding devices instead. At 8000 Ghz you would need some drastic changes to the memory structure if the processor should not waste 95% of its lightning fast cycles waitning for memory response ... :/


Well, we've already reached a point where the bottleneck isn't
the processor, so I think that pretty much proves that you're right
about that :)

[jdj]

Originally posted by Thalaxis
No, the MHz rating is a measure of the number of clock ticks per
second, nothing more. It has very little to do with the processor's
throughput directly, in fact. The simplest way to explain it is that
for each clock tick, all of the execution units do their one task.
There could be a dozen execution units, there could be just two...
but it's still all timed by a single clock.


Ok - if you say so. I was under the impression that the processor was timed by the clock on the motherboard wich still runs in the MHz range and that one machine cycle (time to advance a instruction one step down the pipeline) was equal to one or in rare occasions two clock cycles. If multiple pipelines were running that would decrease the average cycles per instruction and hence the apparent frequencey increase of the processor.

But I can be wrong of course - I do not work with this and I bow down to superior expertise :bowdown:

This would imply though that we have small microwave ovens just at our sides... :eek:

/ Daniel

[Thalaxis]

Originally posted by jdj
Ok - if you say so. I was under the impression that the processor was timed by the clock on the motherboard wich still runs in the MHz range and that one machine cycle (time to advance a instruction one step down the pipeline) was equal to one or in rare occasions two clock cycles. If multiple pipelines were running that would decrease the average cycles per instruction and hence the apparent frequencey increase of the processor.


The processor's clock is usually a multiple of the clock on the
motherboard (there are some benefits to this, in terms of
complexity in transferring data between devices), but the CPU's
core clock really is what the rated clock speed is.

The parallelism is a bit more complicated than that. Adding
pipelines doesn't make one instruction move through its pipeline
more quickly, it just allows the processor to do other stuff in the
meantime.

At each clock tick, the functional units do their thing. Usually, that
means passing a result for their "job" onto the next stage. In
some cases, doing their job takes more than one clock cycle
(that's what they mean by latency), and in some cases it can be
tens of cycles. I think some of the more complex arithmetic
functions like arctan can take as many as 150 cycles, in fact.
(Ouch.)

If MHz measured throughput, then for one thing it would vary by
workload; for example, the 1.5 GHz Itanium2 is the fastest
number cruncher on the market -- in floating point. In integer
math, the Opteron246 flattens it (running 64-bit OS, anyway).


But I can be wrong of course - I do not work with this and I bow down to superior expertise :bowdown:


How else do you think I learned this? :)


This would imply though that we have small microwave ovens just at our sides...


Not that I like it any more than you do, but you're right.

[Omnicrola]

Originally posted by Phil "Osirus"
I'm sure we will make processors that transfer data faster than the speed of light sooner or later, but that would involve crazy intradimensional-physics-something-something:)

What a coincidence....

1 Bit Quantum Memory (http://www.sciscoop.com/story/2003/10/31/7845/5789) :eek: :D
In the experiment, designed by Gao Hong, a post-doctoral student in Batelaan's lab, 20-microsecond pulses of polarized light were beamed into a tubular, 4-centimeter-long cell containing rubidium vapor, where the pulses were captured before being released intact. Light normally moves through space at about 186,000 miles per second and a microsecond is one-millionth of a second, so a 20-microsecond light pulse normally would be about 3.72 miles long. But in Gao's experiment, that 3.72-mile-long light pulse was captured and stored in a tube about 1 1/2-inches in length.

[Thalaxis]

Thanks for the link -- there's some REALLY cool stuff on the
horizon these days :)

Here's a more short-term item of coolness:
Antelope Technology (http://www.antelopetech.com/en/Index.aspx)

[Array]

Originally posted by trthing
...like 3D rendering...

Sweet :buttrock:

nope...not rendering. if you used this thing to render out an image, it would look no better than quake2 (ie before we had programmable pixel/vertex shaders on GPU's). you need a programmable cpu to run custom shaders.

[Self-Designer]

In a matter of fact it's "old news":

- Here's a photo i posted no one replied about the perfectly same thing! - (http://www.cgtalk.com/showthread.php?threadid=94584&highlight=israel)

Were did i go wrong in posting that no one replied, yet here - party everyone!?? :hmm:

Well, no hard feelings :)

[Cyberdigitus]

Hmm, wait until you see Moletronics (http://www.wired.com/wired/archive/8.07/moletronics.html?pg=1) take off...

[trthing]

Originally posted by Array
nope...not rendering. if you used this thing to render out an image, it would look no better than quake2 (ie before we had programmable pixel/vertex shaders on GPU's). you need a programmable cpu to run custom shaders.

Heh, actually was replying to a post right above mine there...

Nonetheless, at least theoretically, if you have that kind of technology under you armpit it is possible to build a dedicated renderer "helper", as to say: it would be a matter of stepping down one or two hyerarchical levels...
:wavey:

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.