Intel developing optical chip-to-chip interconnects

[Idontcare]

Intel developing optical chip-to-chip interconnects

Intel Corp. is studying optical interconnects with an eye toward replacing chip-to-chip electrical interconnects in order to overcome looming bandwidth issues as microprocessors with an increasing number of cores usher in the era of tera-scale computing.

http://www.eetimes.com/news/se...l;?articleID=213900581

I'm pretty sure Intel made it public knowledge they started developing optical interconnects some 4 or 5 years ago if not even longer than that.

So, yeah I know this is actually not new news, and EETimes has taken a flair for the sensational headlines lately, but it is nice to see the efforts at Intel continued onwards toward eventual productization.

I wonder how much latency between the IMC and ram dimms could be shaved off if they used optical interconnects instead of PCB copper traces.

The best thing about optical interconnects is that they use photons instead of electrons, i.e. no charge involved with the signal carrier. That means no capacitance issues with the board design. Nice.

 

[aigomorla]

isnt this the PLX chip?

Which is also located on a ATI X2 card?

 

[SunnyD]

Originally posted by: Idontcare
The best thing about optical interconnects is that they use photons instead of electrons, i.e. no charge involved with the signal carrier. That means no capacitance issues with the board design. Nice.

And much lower thermal characteristics.

 

[aka1nas]

Originally posted by: aigomorla
isnt this the PLX chip?

Which is also located on a ATI X2 card?

Nope, there aren't any actual commercial products based on optical interconnects yet, AFAIK.

 

[polarbear6]

this i think would only result in better core to core communication but what about the cpu to mother board like
the fsb
they should probably have optical interconnects where it is more useful
more like northbridge to cpu links
and i wonder how would the price factor be more like how many k's(1k = 1000$)

 

[Cogman]

Originally posted by: SunnyD

Originally posted by: Idontcare
The best thing about optical interconnects is that they use photons instead of electrons, i.e. no charge involved with the signal carrier. That means no capacitance issues with the board design. Nice.

And much lower thermal characteristics.

Not to mention fewer problems with induction and magnetism interference. It will be interesting to see if this takes off.

 

[Kraeoss]

lol @ 1k = $1000 lol wow that's like 2 x2's xfired/slied lol

 

[aigomorla]

Originally posted by: aka1nas

Originally posted by: aigomorla
isnt this the PLX chip?

Which is also located on a ATI X2 card?

Nope, there aren't any actual commercial products based on optical interconnects yet, AFAIK.

so an optical PLX chip?

becuase itsn the PLX chip a interconnect chip?

 

[Idontcare]

Originally posted by: aigomorla

Originally posted by: aka1nas

Originally posted by: aigomorla
isnt this the PLX chip?

Which is also located on a ATI X2 card?

Nope, there aren't any actual commercial products based on optical interconnects yet, AFAIK.

so an optical PLX chip?

becuase itsn the PLX chip a interconnect chip?

Yeah you are getting the idea. Basically anything that involves two discreet IC's communicating with one another could in theory eventually be interconnected with optical interconnects.

Ram, SB, hard-drives, PLX, hydra, GPU's, multi-socket CPU's, etc.

Naturally it won't be a one-size fits all, and cost/performance will be at a premium when these debut. I'm just happy someone is researching it. It gives me warm fuzzies.

 

[taltamir]

mmm... intel makes SSDs, GPUs, North bridges, South Bridges, and CPUs... Imagine them all with optical interconnects

 

[soccerballtux]

Originally posted by: taltamir
mmm... intel makes SSDs, GPUs, North bridges, South Bridges, and CPUs... Imagine them all with optical interconnects

Microsoft will have a tough job on their hands using up all this processing power.

 

[Cogman]

Originally posted by: soccerballtux

Originally posted by: taltamir
mmm... intel makes SSDs, GPUs, North bridges, South Bridges, and CPUs... Imagine them all with optical interconnects

Microsoft will have a tough job on their hands using up all this processing power.

Lets not get ahead of ourselves here. The only thing that will increase as a result of this is (probably) bandwidth. While higher bandwidth has some effect on the total speed of the system, we aren't really in a situation yet where it is the major limiting factor.

People equate light with fast, but when speaking about speed with computers we are more often concerned with operation speed then how fast our messenger is.

Probably the real beauty of this (in my mind) is the fact that we could potentially have one line for all our bandwidth, so rather then sending 1 bit per line, we would be able to send multiple bits per line using different light spectrum. That translates into mother boards that is able to have a heck of a lot more devices on it then we now have (much less lines to fight with.)

 

[aigomorla]

Originally posted by: soccerballtux

Originally posted by: taltamir
mmm... intel makes SSDs, GPUs, North bridges, South Bridges, and CPUs... Imagine them all with optical interconnects

Microsoft will have a tough job on their hands using up all this processing power.

windows 95.

problem solved.

 

[chizow]

Sounds like it'd be a perfect fit for some of Intel's future SoC (system on a chip) designs, since power and temperatures would be of particular concern there.

 

[Hacp]

Hp's doing the same thing.

http://bits.blogs.nytimes.com/...hunk-of-burning-light/

 

[Nemesis 1]

Originally posted by: aigomorla
isnt this the PLX chip?

Which is also located on a ATI X2 card?

What. Did you just say . Ati has optics on board. LOL LOL.

 

[Nemesis 1]

Originally posted by: Idontcare

Originally posted by: aigomorla

Originally posted by: aka1nas

Originally posted by: aigomorla
isnt this the PLX chip?

Which is also located on a ATI X2 card?

Nope, there aren't any actual commercial products based on optical interconnects yet, AFAIK.

so an optical PLX chip?

becuase itsn the PLX chip a interconnect chip?

Yeah you are getting the idea. Basically anything that involves two discreet IC's communicating with one another could in theory eventually be interconnected with optical interconnects.

Ram, SB, hard-drives, PLX, hydra, GPU's, multi-socket CPU's, etc.

Naturally it won't be a one-size fits all, and cost/performance will be at a premium when these debut. I'm just happy someone is researching it. It gives me warm fuzzies.

Actually Intel says its cheaper than what is used presently.

 

[Idontcare]

Originally posted by: Hacp
Hp's doing the same thing.

http://bits.blogs.nytimes.com/...hunk-of-burning-light/

Cool. Thx for the link :beer:

Originally posted by: Nemesis 1
Actually Intel says its cheaper than what is used presently.

Well why wouldn't they? That's marketing 101. Claim something that doesn't exist (so cost claims can't be quantified, substantiated as being false) is cheaper than something that does exist.

There is no harm in saying it will be cheaper, no one comes back in 5 yrs from now when it finally debuts and says "hey! you said it would be cheaper, but you are charging me and arm and a leg for first-gen hardware!?"...Intel's response would (rightly) be "yes, we said it wouldn't cost us any more, of course it'll cost you more when you buy it, that's called mark-up, what kind of idiots do you take us for?"

 

[Nemesis 1]

Oh . OK its been so long sense hearing intel do double talk that I forgot they to market. Its just Intel has been straight forward with us I forgot about the P4P days. Ya what you say is possiable . 5 years ya say. So that does out past my time.

 

[Idontcare]

Yeah think about how long Intel took to replace their FSB topology with QPI. QPI has an equally long life ahead of it before it will be replaced, either with an all the more superior electrical bus interface or potentially with an optical interconnect interface if it is ready for productization at that time.

I highly doubt Otellini will still be CEO of Intel when this technology debuts. This is long-range research just now getting to the inflection point where it is transitioning into the development phase. Hence the refresher on it's existence to the public domain.

I have no confidential insight into the development and productization timelines for this technology, but given the typical pace that this stuff moves along at I'd say we won't see it debut in commercial products for another decade.

It's first use will be in a leading edge flagship supercomputer for the DOD. When we see that system announced we'll be 2-3 yrs from hardware, when we see hardware installed and operating we'll be another 3-4 yrs away from seeing it debut on the consumer market. Just my opinion, I don't know anything.

 

[taltamir]

Originally posted by: Cogman

Originally posted by: soccerballtux

Originally posted by: taltamir
mmm... intel makes SSDs, GPUs, North bridges, South Bridges, and CPUs... Imagine them all with optical interconnects

Microsoft will have a tough job on their hands using up all this processing power.

Lets not get ahead of ourselves here. The only thing that will increase as a result of this is (probably) bandwidth. While higher bandwidth has some effect on the total speed of the system, we aren't really in a situation yet where it is the major limiting factor.

People equate light with fast, but when speaking about speed with computers we are more often concerned with operation speed then how fast our messenger is.

Probably the real beauty of this (in my mind) is the fact that we could potentially have one line for all our bandwidth, so rather then sending 1 bit per line, we would be able to send multiple bits per line using different light spectrum. That translates into mother boards that is able to have a heck of a lot more devices on it then we now have (much less lines to fight with.)

electricity moves at the speed of light, electricity just has problems with:
1. heat
2. interference
3. insulation

The extra bandwith will improve fsb speeds will reducing temperature... so I expect it will allow for better overclocks.

 

[Idontcare]

Originally posted by: taltamir

Originally posted by: Cogman

Originally posted by: soccerballtux

Originally posted by: taltamir
mmm... intel makes SSDs, GPUs, North bridges, South Bridges, and CPUs... Imagine them all with optical interconnects

Microsoft will have a tough job on their hands using up all this processing power.

Lets not get ahead of ourselves here. The only thing that will increase as a result of this is (probably) bandwidth. While higher bandwidth has some effect on the total speed of the system, we aren't really in a situation yet where it is the major limiting factor.

People equate light with fast, but when speaking about speed with computers we are more often concerned with operation speed then how fast our messenger is.

Probably the real beauty of this (in my mind) is the fact that we could potentially have one line for all our bandwidth, so rather then sending 1 bit per line, we would be able to send multiple bits per line using different light spectrum. That translates into mother boards that is able to have a heck of a lot more devices on it then we now have (much less lines to fight with.)

electricity moves at the speed of light, electricity just has problems with:
1. heat
2. interference
3. insulation

The extra bandwith will improve fsb speeds will reducing temperature... so I expect it will allow for better overclocks.

Velocity of propagation

Velocity of Propagation (VoP) or velocity factor (VF) is a parameter that characterizes the speed at which an electrical signal (e.g. radio) passes through a medium

http://en.wikipedia.org/wiki/Velocity_factor

Propagation Delay and Its Relationship to Maximum Cable Length

Electrical signals in a copper wire travel at approximately 2/3 the speed of light.

http://www.wildpackets.com/sup...rnet/propagation_delay

Pushing out the limits of copper - Components

Copper is still an excellent medium for high-speed signal transmission. In theory, it should be capable of signal transmission near the speed of light. Structures and materials have been created that have demonstrated signal propagation in copper at nearly 80 percent of the speed of light and development continues.

Optoelectronics technology does offer unique advantages in signal transmission, and it is unexcelled in many ways. High-speed, low-loss, excellent signal integrity over long distances, security and nonexistent electromagnetic emissions (thus virtually no cross-talk) are among the most commonly cited advantages.

http://findarticles.com/p/arti...F/is_43_48/ai_93230614

 

[Cogman]

Originally posted by: taltamir

Originally posted by: Cogman

Originally posted by: soccerballtux

Originally posted by: taltamir
mmm... intel makes SSDs, GPUs, North bridges, South Bridges, and CPUs... Imagine them all with optical interconnects

Microsoft will have a tough job on their hands using up all this processing power.

Lets not get ahead of ourselves here. The only thing that will increase as a result of this is (probably) bandwidth. While higher bandwidth has some effect on the total speed of the system, we aren't really in a situation yet where it is the major limiting factor.

People equate light with fast, but when speaking about speed with computers we are more often concerned with operation speed then how fast our messenger is.

Probably the real beauty of this (in my mind) is the fact that we could potentially have one line for all our bandwidth, so rather then sending 1 bit per line, we would be able to send multiple bits per line using different light spectrum. That translates into mother boards that is able to have a heck of a lot more devices on it then we now have (much less lines to fight with.)

electricity moves at the speed of light, electricity just has problems with:
1. heat
2. interference
3. insulation

The extra bandwith will improve fsb speeds will reducing temperature... so I expect it will allow for better overclocks.

Last I checked, the temperature of the bus wires wasn't a big issue. This isn't going to reduce (by much) the thermal output of a chip. These are the wires coming out of the chip that would be replaced. However, interference and insulation are definitely issues that would be solved by this.

I don't know about initial bandwidth ratings though. It might be higher, it might be lower. Definitely it has the potential to become much higher then copper, but I don't know if that will be the case initially. (well, I guess they probably won't release it until it is faster then copper. So we'll see when that point gets here)

But again, this isn't going to affect the thermal properties of the chips all that much (it may even increase them ever so slightly). So higher overclocks aren't really a possibility. And as I said before, we aren't quite to the point where our CPUs are starved for bandwidth. We will get there, but we aren't there today. So increasing the bandwidth by 2x won't yield too many performance benefits (Think about the difference of going from DDR2 800 to DDR2 1066, or even DDR3 1200. Most reviews I've read show no notable differences in many applications).

 

[JackyP]

Originally posted by: Idontcare
Yeah think about how long Intel took to replace their FSB topology with QPI. QPI has an equally long life ahead of it before it will be replaced, either with an all the more superior electrical bus interface or potentially with an optical interconnect interface if it is ready for productization at that time.

Are you sure QPI will be necessarily replaced with a completely new optical interconnect and it can't just replace our current QPI implementation without major changes?
Dkanter said: "Clock encoding and clock and data recovery are prerequisites for optical interconnects, which will eventually be used to overcome the limitations of copper. By specifying CSI in an expansive fashion, the architects created a protocol stack that can naturally be extended from a parallel implementation over copper to optical communication." http://www.realworldtech.com/i...82807020032&mode=print

So there is a possibility we will see optical interconnects during the lifetime of QPI, no?

 

[Idontcare]

Originally posted by: JackyP

Originally posted by: Idontcare
Yeah think about how long Intel took to replace their FSB topology with QPI. QPI has an equally long life ahead of it before it will be replaced, either with an all the more superior electrical bus interface or potentially with an optical interconnect interface if it is ready for productization at that time.

Are you sure QPI will be necessarily replaced with a completely new optical interconnect and it can't just replace our current QPI implementation without major changes?
Dkanter said: "Clock encoding and clock and data recovery are prerequisites for optical interconnects, which will eventually be used to overcome the limitations of copper. By specifying CSI in an expansive fashion, the architects created a protocol stack that can naturally be extended from a parallel implementation over copper to optical communication." http://www.realworldtech.com/i...82807020032&mode=print

So there is a possibility we will see optical interconnects during the lifetime of QPI, no?

Oh yeah, I didn't mean to imply the topology will be replaced just because the data transfer medium goes from electrical/copper to photons/waveguides. Optical interconnect could have been implemented on 486 mobo's 20yrs ago if the optical interconnect technology existed then, it is entirely invariant to the bus topology. (but some bus topologies will make better use of optical interconnects, multiplexing being what it is and does)

What I meant to convey was that I'd expect optical interconnects to be targeted for productization and introduction alongside whatever internal roadmaps exist at Intel for debut of their next bus topology.

Kind of like how the IMC debuted simultaneously with QPI, whereas an IMC on FSB topology was possible they simply chose to align the roadmaps of the two major changes to coincide with each other.