Intel Rolling Out 800Gbps Cables This Year

phmadore writes "10Gbps cables are what are commonly used in large server centers today, but very soon, according to Ars, 800Gbps cables will be available from Intel. From the article: 'The new cables are based on Intel's Silicon Photonics technology that pushes 25Gbps across each fiber. Last year, Intel demonstrated speeds of 100Gbps in each direction, using eight fibers. A new connector that goes by the name "MXC" holds up to 64 fibers ... The fiber technology also maintains its maximum speed over much greater distances than copper, sending 800Gbps at lengths up to 300 meters, Intel photonics technology lab director Mario Paniccia told Ars. Eventually, the industry could boost the per-line rate from 25Gbps to 50Gbps, doubling the overall throughput without adding fibers, he said.'"

Like these?

[Anonymous Coward]

http://www.amazon.com/Denon-AKDL1-Dedicated-Cable-Version/dp/B000I1X6PM

http://www.bestbuy.com/site/AudioQuest+-+Diamond+3.3%27+High-Speed+HDMI+Cable+-+Dark+Gray/Black/2383276.p;jsessionid=310CCC6FDFA4F4B48027114FF363F3FC.bbolsp-app04-32?id=1218324437192&skuId=2383276#BVRRWidgetID

http://www.geekosystem.com/funny-amazon-review/

Good to know.

Re:

[UnknowingFool]

Well if you just looked at the photos of the cables, you would see it is not a common consumer cable. It is a 64 optic fiber cable that does not exist in other fields. The Denon AKDL1 is simply an ethernet cable. AudioQuest Diamond is simply an HDMI cable. AudioQuest K2 is simply speaker cable. The Intel cable probably will be expensive but it will be used for back-bone usage and not consumer usage.

Re:

[AK Marc]

Why aren't backbones using 100 Gbps (an established standard) and WDM for up to 88 channels of 100 Gbps? Though WDM is usually not used for shorter distances, you can already get 8.8 Tbps using commonly available commercial gear on 20 year old fiber.

Re:

[UnknowingFool]

Why aren't backbones using 100 Gbps (an established standard) and WDM for up to 88 channels of 100 Gbps?

You mean besides cost? Yeah, that pesky factor that everyone is willing ignore because they want things now, now, now. Also just because the backbone is using high capacity fiber does not mean consumers will get it. Google is probably the biggest user of fiber out there and is certainly the biggest user of dark fiber. I'm sure Google is using the highest capacity they can get at a reasonable cost.

Re:

[AK Marc]

Backbones are using 100 Gbps. I've seen it and personally done it. But yes, at this time, 100 Gbps is very expensive. This isn't news for nerds, this is a sale on the Home Shopping Network. Yes I understand that it's a cost breakthrough for 1/10th of highest commercially available. But, when 400 Gbps becomes commercially availble (this year, maybe next), about the time Intel's hits the market, it'll be 1/40th of the best commercially available.

It trades cost and complexity in the wiring for cost. It'

Re:

[Bengie]

400gb/s is slow.
https://www.infinera.com/j7/se... [infinera.com]

delivering 8 Terabits per second (Tb/s) capacity using production ready super-channels across 800 kms of ITU-T G.653 Dispersion Shifted Fiber (DSF).

This is per fiber, uni directional of course.

Re:

[AK Marc]

Read two up of my posts (I mention 8.8 Tbps). Your link points out they are using WDM, same as I note. I can do 88*100 Gbps today, using existing standards, not some proprietary crap. Perhaps you should read your own link closer next time.

Re:

[Bengie]

The difference is this is using WDMD and can carve up that 8.8Tb/s into any combination of channels between 10Gb/s and 500Gb/s. Want 880 10Gb virtual channels? Sure! Want 17 500Gb/s channels? Sure! Mix and match, and you don't need to take down any channels to add new ones. It also uses about 10x-100x less power. Behold, the power of Photon Integrated Circuits(PICs)

Re:

[AK Marc]

You've magically discovered G.709. It will do all you describe, today, without proprietary crap. It even has an installed base. There's nothing compelling in this to get anyone to throw out a better network to adopt this. Yes, I'm sure it's lower power. Most 100Gbps DWDM is for longer distances, so you attenuate the signal for short distances. If it were optimized for shorter distances, it'd use 10-100x less power. We don't need a new standard to do that.

Re:

[Bengie]

The physical channels can also be tweaked for range instead of bandwidth. You can reduce the physical channel from 500gb to some slower speed, like 250gb/s and nearly double your range. These PICs have a 700km range with no signal regeneration, and over 1100km if you reduce the logical bandwidth of the physical channel. This has been testing in the real world. Stockholm to Frankfurt with no repeaters and can handle 8tb+, which can all be provisioned in less than 1 hour.

Re:

[AK Marc]

I work with that on a regular basis. It's a few years old, but works as you describe. Again, I don't see anything new in this, other than unsubstantiated claims of cost.

Re:

[davydagger]

no, but it will have a ripple effect.

If 800 per cable becomes the new standard, companies are going to start dumping their existing fiber connectors, and they will get cheap, and there will be less demand. The companies that make them, will most likely keep making the same or similx GBICs, because they already did the R&D, and they are cheaper to produce because the cost of design is already paid for.

So existing connections will get cheaper. The net effect is the total cost per/ MB/s is going to go down

Re:

[Bengie]

This is only temporary, IBM already demonstrated production ready integrated circuits that could be built directly into a CPU/Chipset/etc, that could handle 1tb/s-2tb/s with a 2km range. The process is also so cheap, they expect it to be integrated into low-end cell-phones and disposable electronics. It's really more like a 2tb/s fiber data bus with a 2km range. They actually claim that this new tech is cheaper than using copper traces, once you get past retooling your factory.

Re:Like these?

[petermgreen]

Some cables are legitimately expensive because they are expensive to make. Some cables are expensive because they are a niche product and there is only one vendor and then some cables are stupidly expensive simply to prey on idiots.

These cables undoutablly will not be cheap but they may well be cheaper than terminating and patching all those fibers seperately for those few niches that really need that much bandwidth between the same pair of devices.

Re:

[lsatenstein]

Some cables are legitimately expensive because they are expensive to make. Some cables are expensive because they are a niche product and there is only one vendor and then some cables are stupidly expensive simply to prey on idiots.

These cables undoutablly will not be cheap but they may well be cheaper than terminating and patching all those fibers seperately for those few niches that really need that much bandwidth between the same pair of devices.

Your comments remind me of the $40.00 six foot HDMI cables from that big box store. I went to the dollar store and bought equivalents for, you guessed it, two bucks. Have as yet (2 years) not had a problem with them.

And the new scam of the year is LEDs.

Before it became "wrong" to purchase incandescents, the LED lamps were priced 1/3 lower. I paid $7 per bulb, and now the price varies between $14 and $37. LED lamps may pose a fire hazard, not because of low wattage at the LEDS, but because the dropdown

Will we still talk ethernet over it?

[glasshole]

Or will it be a new protocol all together. I guess it depends if its suppose to be point to point or not.

Re:

[msauve]

The real question should be - why are they misleading us?

This is nothing but marketing. They're comparing 10 Gb Ethernet, which runs over a single fiber pair, to something which runs at 25 Gb, over 32 fiber pairs.

Meh. 100 Gb Ethernet is commercially available today, which is 4x faster that what the article is hyping.

Re:

[pLnCrZy]

And 100Gbps ethernet runs over 10 fiber pairs.

Your point?

Re:

[jabuzz]

Well there is 802.3ad of course so yes. This seems to be more about using silicon diodes and multimode fibre to be cheaper in the short term than using single mode fibre and other laser sources.

Re:

[msauve]

Uh, no. Take 100GBASE-LR4 as an example. It runs over a single fiber pair. It does send 4 "lanes," or frequencies of light, through the fiber using WDM, but it's a single fiber in each direction.

You're apparently trying to refer to 100GBASE-SR10, but don't know enough to say so.

Re:

[pLnCrZy]

... or I know plenty, and didn't feel the need to wave my nerd around to show you how big it is.

100GBASE-LR4 is still a multiplex. It runs over a single physical fiber pair. That doesn't mean it's a 100Gbps signaling rate.

My comment was to the one above mine, not to the one my magic hat predicted from you in the future. In the comment to which I replied, the poster was grumbling that 100Gbps ethernet is commercially available today in contrast to "something which runs at 25Gb, over 32 fiber pairs."

Was my

Re:

[msauve]

Sorry to step on your tiny pee-pee, but it's obvious you don't have a clue how any of it works, and are relying on searching Wikipedia for your "knowledge."

Re:

[pLnCrZy]

That's cute.

You like to throw jabs and insults, yet you don't actually back any of it up. I'm impressed by you. Really.

Re:Will we still talk ethernet over it?

[AK Marc]

I work with 100 Gbps that runs over a single pair. And doesn't use DWDM, so you can run 88 channels of it, for 8.8 Tbps. On a single fiber. And the fiber can be 20+ year old single mode (though results depend on quality). Industry standard, no new cabling. Intel is solving a problem that doesn't exist. Or maybe coming up with something not-new in a cheaper price point. 88-channel DWDM is expensive. I think we are paying somewhere in the neighborhood of $1,000,000 per fully-populated 8.8Tbps link. But then, that's for optics capable of 1000+ mile transmission (with amplification). As the market is small for those speeds from server to switch or with any level of port density, so the price is not optimized for common use.

So from what I can tell, this is less than 1/10th the speed of common industry standard gear, but cheap. Since when did stories about price breakthroughs get front page (other than solar)?

Re:

[Lennie]

"cheaper price point"

That is the whole point of silicon photonics, mass production.

Re:

[AK Marc]

Like 1Gbps? When it came out, I paid more than $1000 per port for a Cisco switch for 1G ports. Now, $30 devices have 4+ 1Gbps ports. Just wait, and they will drop. Racing to the bottom with an inferior product isn't good for the customer or the companies involved. Working on cheaper 400 Gbps ports and standards would be better for all.

Re:

[Lennie]

You are talking about 1 Gbps ports, not fiber optic cables.

What I mean is, fiber optic cables can not be mass-produced (well), termination and testing of ready made patches is still a partly done by hand. That is what silicon photonics is trying to solve, mass production of fiber optic cables with connectors and all. This should drive down the price.

Re:

[AK Marc]

The story read like they were solving short distance issues. And I don't know whether they are "mass produced" but I can get fiber optic cables that are pretty cheap. Inventing a new standard for connectors, adaptors, and cards to "solve" an already solved problem with low cost that would require re-buying already owned things sounds like a really stupid idea. Do you expect datacenters everywhere to abandon the 100Gbps they already support for this new "cheaper" tech that will cost them lots of money to

Re:

[timeOday]

Look at the screenshot in the article - they packed 64 fibers into a cable about the same bulk as Cat5 copper - even the connector, which has 64 'pins'. Just like when somebody makes a new multi-core chip that replaces an entire cluster - that's progress.

Re:

[msauve]

It's like having a beowulf cluster of Commodore 64s instead of a modern desktop. That's not progress.

Re:

[ltskinol]

This is nothing but marketing. They're comparing 10 Gb Ethernet, which runs over a single fiber pair, to something which runs at 25 Gb, over 32 fiber pairs.

Meh. 100 Gb Ethernet is commercially available today, which is 4x faster that what the article is hyping.

It's not just marketing.

25 Gb/s per fiber x 32 fibers = 800 Gb/s. 8x faster than 100 Gb/s.

Re:

[msauve]

100 Gb/s per fiber x 32 fibers = 3200 Gb/s. 4x faster than this marketing of a promise for the future, and available today.

Re:

[ltskinol]

Ok. Do you have a source for "100 Gb/s per fiber?"

Re: Will we still talk ethernet over it?

[astar]

Yah. I can have 100Mbs to my house right now off the shelf. Since the fiber is there I expect that if I am needy with deep pockets I can get 1Gbs. And I am sure there are lots of dark Fibre at the road so I can get 100Gbs with lots and lots of money and lots of lead time. So I need the right corporate configuration to get Internet2 to be friendly but that is more just more money plus mission commitment. But my desktop OS does not support ATM/SONET. :-)

But let us have a little fun here. There are some

Excellent News, Now if only...

[bobbied]

We can get Netflix and Verizon together using this I'll be able to actually watch something now and then...

Doesn't matter

[Anonymous Coward]

Here in the States, our ISPs will keep their shitty service and infrastructure. Regardless of Intel's products, I'll be stuck with my 1.5/0.25Mbps ADSL from HellSouth (an AT&T company at $42/mo) - unless, I pony up for their Uverse shit and get landline and TV shoved up my ass then I can get a whole 3.0Mbps - woohoo.

Re:

[alen]

$uck$ to be you

in NYC time warner is upgrading my 20/2 internet to 50/5 later this year for the same price i'm paying now. they doubled speeds last year as well.

later this year the top tier for TWC in NYC and LA is going to be 300/50 or so

Re:

[Anonymous Coward]

I get something like 40/8 but Netflix doesn't care. I still get lots of buffering. Good luck with your 50/5.

Re:

[alen]

i can stream netflix and HBO Go or time warner cable live TV at the same time in HD with no problems on 20/2

of course almost everything i have is on cat 5 and i'm not streaming into different rooms via wifi

Re:

[Jason Levine]

Here in upstate NY, I pay for 15mbps but actually get closer to 8mbps. Since there's no FIOS or other high speed Internet service where I live, Time Warner Cable has no incentive to upgrade their network by me. They might eventually get around to it, but they'll take their time. (I'm not in a rural area so they don't even have that excuse.)

Re:

[bobbied]

Hey now.. I get 25Mbps both ways for $90/month... Well, I get that to any of the speed test servers out there. Now if I actually want to watch Netflix or something, all bets are off.

I prefer Monster cables

[NotDrWho]

The guy at Best Buy told me they were the best.

Re:

[noh8rz10]

dude, don't be a sucker! they just push it on you because the gross margins are huuuge. my favorite cables come from monoprice [monoprice.com]. You can get a 3 foot HDMI cable for $3, and a 6 foot cable for $4.

Re:I prefer Monster cables

[H0p313ss]

One of you managed to make my eyes bleed, but I'm not sure which one.

Re:

[Redmancometh]

I'm not usually one of "those guys"
buuut

WHOOSH

Re:

[bobbied]

Oh yea, got to love the $200 HDMI cables they sell... What part of "digital" do people not get? And why does the signal transfer quality matter as long as it is "good enough" to get the logical one or zero to the destination?

Always felt sorry for the folks that fell for the "monster cable" thing, even back when it *might* have mattered in speaker cables, not that I knew anybody who could actually hear the difference between 12 AWG and full on monster cables of 10' length. Now days, with digital being the

Re:

[iggymanz]

I must say these 800GB fiber bundles give very molten and sonorus bass notes below 64 hz.

Re:

[bobbied]

You must be using the mufti-mode fiber cable, switch to the single mode version and you can extend your sonic impact below 15Hz, but only if your Tube amp can handle the increased power required and the magnaplaners are big enough.

Re:

[iggymanz]

but that fibre aren't yet available with iridium plating on the connectors for crisp bright treble

Re:

[wiredlogic]

High speed digital cables do require more expensive manufacturing processes. They all use differential signaling and require length matching and controlled impedance within wire pairs. The manufacturing process is a little more demanding than slapping some wires together willy-nilly. That being said, on high volume products like HDMI the added costs are negligible on a per unit basis and the gold plated $200 fluff is just gouging the ignorati.

Re:

[bobbied]

It's not that hard to build an HDMI cable if you have the right wire with close to the right twist. Length variances isn't all that important, it's the twisting of the various pairs that matters. Even that isn't critical until you start talking about really long cables that most people don't need anyway.

Further there is a standard that all "HDMI" cables must comply with to use the HDMI lable. If it meets the standard, it will work. So, if it is advertised/labeled as "HDMI" and not manufacturer reject

Re:

[Ravaldy]

If you buy a Monster HDMI cable over a no name HDMI cable you aren't getting the same thing. For 1, the connectors are of much better quality, the copper used is capable of handling more bending without breaking and it is capable of higher frequencies and longer distances. Would I pay as much money as they want for this cable? No but I would pick a better choice than the crappy $5 cable.

I only partially agree with the argument of "why do we care if it's a digital signal". In the case of HDMI instead of gett

Re:

[bobbied]

I would concur with your conclusion, just not with your reasons.

Unless you are moving your HDMI cables a lot, I doubt you are going to have issues as you suggest. A 2 Meter HDMI cable for $5 can be replaced 5 times over the $45 prime super duper name brand, and I'm not so sure the name brand is going to actually be usable at 5x the number of connect/disconnect cycles. Go with the cheep one. The tolerances for 6 foot cables are EASY to meet, even with cheep materials. Even my ElCheepo E-Bay cables work and

Re:

[petermgreen]

but on HDMI it's likely just not going to work if the error rate is anything you'd notice

Not true, when a HDMI connection gets poor you get "sparkles" on the screen.

Remember, the issue is that you MUST meet the same specifications no matter how long the cable is.

Afaict (granted I got this information from a cable vendor so take it with a pinch of salt but what they say seems to make a lot of sense)

1: Apparently the HDMI guys are lax on enforcing the rules with many noncompliant cables on the market.
2: Many real setups are likely to have multiple cables between source and sink (wallports, passive switch boxes, whatever), even if all the individual cables are compliant the combination may not

Impressive?

[Anonymous Coward]

I don't think this is impressive. I know this is different http://tech.slashdot.org/story/99/10/12/1835225/nortel-gets-64-terabits-on-a-single-fibre but Nortel did 80gbits/s on a single wavelength over 480 km in 1999. They had to have multiple rack of equipment to generate all wavelength to get to their 6.4 tbits/s but I don't see why we could not just use one unit. Did it really take 15 years to adapt the tech and integrate it in a server for server to server communication in data centers?

It is all a matter of cost and size

[hamjudo]

These will be used in data centers where it is common to have redundant systems connected with redundant cables, in order to maintain really high uptimes. Say a hypothetical system has a cluster which consists of 16 compute nodes and 2 storage nodes, Each of CPUserver01 through CPUserver16 will have two of these cables going to storageServerA, and two going to StorageServerB. For a total of 64 of these cables, for that one little compute cluster. Which would leave it an island, so of course there will be m

Re:

[Kjella]

Probably for the same reason most of us still have just gigabit ethernet, cost. This is not scale up, just scale out where they can bundle a bunch of "cheap" fiber channels together and presumably get the same effect cheaper. I can't imagine the saving being that much over 64 single channel solutions though, you still need the same number of transmitters, receivers and strings of fiber-optic, it only takes slightly less space and you can save a tiny bit on insulation but you'd be 90% of the way there with c

Re:

[Pinky's Brain]

And when their patents run out I'm sure we'll see some great thing. I'm pretty sure things like partly optical domain QPSK encoding/decoding could be miniaturized and commoditized right now ... yeah the necessary photonic devices are complex and bleeding edge, but not as bleeding edge as the processes Intel uses for it's commodity processors.

Patents are probably the biggest reason why Intel uses this many fibers rather than more intelligent signalling.

25 Gbps is the fastest the industry can do?

[Anonymous Coward]

http://www.newscientist.com/article/mg21028095.500-ultrafast-fibre-optics-set-new-speed-record.html

At the Optical Fiber Communications Conference in Los Angeles last month, Dayou Qian, also of NEC, reported a total data-sending rate of 101.7 terabits per second through 165 kilometres of fibre. He did this by squeezing light pulses from 370 separate lasers into the pulse received by the receiver. Each laser emitted its own narrow sliver of the infrared spectrum, and each contained several polarities, phases

Great

[Chrisq]

At last something that can keep up with my online porn feed

Re:

[Anonymous Coward]

He's got 209 screens.

Let's assume he has very expensive tastes in porn, and likes to watch it at 4K. A 4K video has a bitrate of 477.76 MB/second. Since this stream is equivalent to 100 GB/sec, that works out to a theoretical limit of 209.31 simultaneous streams possible. Rounding down to account for audio streams (209 CD-quality audio streams works out to 36.86 MB/sec), that works out to 209 simultaneous streams of porn.

I conclude that he must be the Architect from the nonexistent Matrix sequels, living

How dare you!

[Chrisq]

I conclude that he must be the Architect from the nonexistent Matrix sequels, living with his mother. And he is a lonely, lonely man.

How dare you begin a sentence with "And".

MXC

[pulski]

Most Extreme Data Transfer Challenge?

All of your data will get there, however half of it will be broken and the other half poorly translated.

Re:

[alen]

strange,

for the last 10 years i've had cable internet service the price has been about the same and risen only about $5 per month and my speed has gone from 5mbps or so to 20mbps down. and will go up to 50mbps for the same price by the end of this year

Re:

[petermgreen]

Over long distances your main costs are the fiber and optical amplifiers, so fancy tricks at the transmitter and receiver to get more out of said infrastructure make a lot of sense.

Over short distances the fiber is a smaller part of the overall cost and so it's often cheaper to just lay more fiber than to bother with the fancy tricks.

Writing data

[blandcramration]

I'll be impressed when we can write data this fast, not when we can transfer it

Who cares?

[DaMattster]

They can have all the speed in the world but it honestly doesn't amount to a scrap heap of shit. I'd be more interested in better, more reliable broadband for consumers and small business. Our broadband is still late 1990s, early 2000s technology while the rest of the world can boast gigabit to the home. I have a friend in England who has a full 100MB symmetric line for what I'm paying Verizon 50 down and 25 up and he even gets an, OMG wait for it, a static IP without bandwidth caps and port blocking. I'm s

Re:

[petermgreen]

I have a friend in England who has a full 100MB symmetric line for what I'm paying Verizon 50 down and 25 up and he even gets an, OMG wait for it, a static IP without bandwidth caps and port blocking.

Where I am (in a suburban area in england) the best broadband services are openreach FTTC (up to 80mbps down, up to 20mbps up, subject to the condition of your phone line) and virgin media cable up to 152mbps down, difficult to find the upload speeds (last I checked the top upload speed they were offering to new customers was 5mbps but upload speeds are not something they like to talk about), also a shitty provider in other ways. Where my parents are (also in suburbia but slightly further out) the openreach

Bandwidth is easy, latency is hard

[Dr. Spork]

When we hear these impressive bandwidth numbers, usually a prophecy that "the future is on the cloud" is not far behind. Once our connection to the server is faster, we will get everything we could want without doing any of the computing on site. But people forget that a very low latency is also very important to the cloud experience, and there is very little that we can do about latency. At some point, we just run into fundamental laws of nature. I have a feeling that in my lifetime, consumers will basical

Won't Affect Provisioning

[Scot Seese]

So, data centers are going to realize a > 8x increase in speed. Awesome. Do you think Time Warner, Comcast, AT&T, and every regional carrier along the way are going to cheerfully provision more bandwidth to their customers? Or will their pencil pushers continue to view bandwidth as a scarce resource to be jealously guarded and sold for a kings' ransom?

We've had cable and DSL modems out in customers' basements for years now that are capable of > 10-20 megabit speed, yet according to a recent Net

Improvements for more modest uses?

[tji]

This tech looks cool. But, it's a bit surprising to me that we've not had any leaps in basic networking for a long time. Everything is gigabit ethernet. I thought 10Gbps Ethernet would have trickled down to some home usage by now.

A 10Gbps connnection to my NAS, hypervisor, or server would be very useful. Or, just an uplink between switches.. But, I've not seen anything available.

Re:

[Deathspawner]

I thought 10Gbps Ethernet would have trickled down to some home usage by now.

I'm with you. It drives me nuts that nothing better than 1Gbps has come along yet. Heck, I'd be thrilled with 5Gbps... it doesn't -have- to be 10Gbps.

Re:

[Lennie]

That is what silicon photonics is promising.

Affordable fiber.

Re:

[Bengie]

Intel stated that 10gb becomes simple, cheaper, and power efficient once you get to 22nm. Once the prices of 22nm gets affordable to NIC manufacturers, expect to see 10gb copper NICs doing to 1gb what 1gb did to 100mb.

Intel won't be using 22nm for non-CPUs until 14nm takes off. It will be a bit before we start to see chipset integrated 10gb, but not too far off. Would be nice to see these paired with PCIe4.0

Re:

[Deathspawner]

Very good points. I can't wait.

We already have 3 100 Gbps ports

[WillAffleckUW]

Actually, at the UW, we already have three 100 Gbps ports - two in the 4545 building and one in the basement of the UW Tower. And a bunch of 40 Gbps ports around the Seattle campus.

The surprising thing is there aren't any down at the UW Medical Center. Where you'd expect more demand.