backtop


Print E-mail del.icio.us 29 comment(s) - last by mvrx.. on Mar 5 at 8:10 PM


The new Optocard from IBM features wave guides smaller than a human hair. The end result is 8 Tbps data rates, at a meager 100 Watts.  (Source: IBM)
IBM's new data-wire is ultra-efficient and carries an extremely large amount of data

IBM is looking to combine both its proficiency for power and its environmental interests in a new effort.  IBM demoed an ultra-fast optical connection earlier this week which it feels will revolutionize the world of supercomputing and IT centers, as well as eventually trickling down to the consumer.

IBM's new technology can transmit at speeds of up to 8 trillion bits a second (8 Tbps), or as IBM puts it, "equivalent to about 5,000 high-definition video streams."  Better yet, the new technology only uses 100 Watts to transmit the massive amount of data -- about the energy required to power a lightbulb.

IBM plans on first deploying the technology commercially in the supercomputers market.  A typical 100 ft electrical wire datalink uses 100 times more power than IBM's prototype, while a typical 100 ft optical data link uses 10 times more power.  Thus there is significant savings in both energy and waste, making the technology doubly "green."

Better yet IBM is working hard to make the technology available in a single package, which includes the optical chips and the optical data bus and utilizes standard components.  This should help to make the technology affordable and ease installation troubles.  IBM Researcher Clint Schow, part of the research team working on the link describes IBM's rapid advances, stating, "Last year we unveiled an optical transceiver chip-set that could transmit a high-definition movie in under a second using highly customized optical components and processes.  Just a year later, we've now connected those high speed chips through printed circuit boards with dense integrated optical 'wiring.' Now we have built an even faster transceiver and have moved the optical components away from custom devices to more standard parts procured from a volume manufacturer, taking an important step toward commercializing the technology."

IBM's technology uses low-loss polymer wave guides on optically enabled circuit boards known as "Optocards."  The wave guides are smaller in diameter than a human hair in size.  IBM's "Optochip" manages the massive amount of raw data streaming along the bus.

The Optochip uses a 3D assembly process, which packs in the complex processing circuitry.  The data bus features a whopping 80 gigabits per channel over 32 channels.  This is historic first, described by IBM as the "first ever demonstration of an integrated module-to-module, 32-channel optical datalink on a printed circuit board."

IBM looks to bring a "scaled down" version of its high speed interconnect to consumer electronics such as cell phones, PDAs and MP3 players in the near future.  This will allow handheld manufacturers to build extremely small and low voltage devices capable of high definition video on incredibly small power envelopes.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Interesting...
By daftrok on 3/2/2008 2:21:10 PM , Rating: 2
If these energy efficient optical wires only cost 2 to 3 times as much as the power hungry optical wires than this has some serious cost-benefit ratios. Any idea how much these green wires cost when compared to the regular ones?




RE: Interesting...
By ImSpartacus on 3/2/2008 2:27:52 PM , Rating: 2
Yeah, I agree. Does this wire have any speed improvements, or is it simply power usage?

I don't think 2 ft of wire between my ipod and my pc would matter unless a substitute is faster. Supercomputer arrays are another story.


RE: Interesting...
By PlasmaBomb on 3/3/2008 2:12:17 PM , Rating: 2
A miniature version would be more useful for data transmission within the device, i.e. from storage to screen, whilst using less power than current methods. The end result would be that you get longer out of a charge :)


RE: Interesting...
By Nik00117 on 3/2/2008 2:28:37 PM , Rating: 2
More so then you'd like to imagine is my guess. I figure this technology should be commerially viable within 3-4 years and consumerable in 6 to 8.


RE: Interesting...
By eye smite on 3/2/2008 8:36:32 PM , Rating: 3
I have to wonder about that since they made all this happen over the last year. Should be interesting to see what happens by the end of 08 if they keep the same developement pace.


RE: Interesting...
By bjacobson on 3/2/08, Rating: -1
RE: Interesting...
By mvrx on 3/5/2008 8:10:09 PM , Rating: 2
According to a paper Intel submitted in April of last year, they want to jump to 40GbE, and put off any serious push for 100GbE until 2015 or so, with no heavy adoption until 2020 due to the price being far out of range for even enterprise IT.

I think with all of IBM's latest breakthru's in the optical field, they could make it a reality sooner.

I want to see PCIe skipped and replaced with 40GbE or 100GbE as an interconnect technology. Massively scalable home computers - something Microsoft and Intel actually don't want you to have.


tos link vx
By tastyratz on 3/3/2008 10:23:38 AM , Rating: 2
So how long until this technology filters down to home theater practical applications. Right now TOSlink bandwidth limits it from carrying next gen format audio. I wouldn't mind a wider band cable to carry all my audio needs.




RE: tos link vx
By sporr on 3/3/2008 1:02:05 PM , Rating: 2
HDMI bandwidth is 10.2Gbps and is capable of streaming DTS Master Audio and Dolby TrueHD at the same time as 1080p material.


RE: tos link vx
By tastyratz on 3/3/2008 4:19:21 PM , Rating: 2
This is true,
I still personally prefer optical as a format however due to its naturally inherent EMI resistance. a backwards compatible toslink upgrade would be pleasantly accepted by me.


RE: tos link vx
By sporr on 3/4/2008 1:58:01 PM , Rating: 2
Good point, EMI has no effect over optical signals.

Unfortunately this technology will be, if it is an electrical signal, rather than light.


RE: tos link vx
By mvrx on 3/5/2008 8:04:50 PM , Rating: 2
TOSlink should have been updated into the 1gbit range long ago. And it still can be, even tho most toslink cables are cheap plastic, they technically could still carry a good 100+ gigabit :)

I'm still no fan of HDMI, I would have much rather seen 10GbE (10 gigabit ethernet) become the standard for home AV routing. At least that would have had a simple cheap upgrade path to 40GbE and 100GbE for when we need to carry 30 super-high-def uncompressed video and 45 channel surround sound audio around the house ;-)


Energy consumption
By myslivec on 3/2/2008 4:24:50 PM , Rating: 5
quote:
Better yet, the new technology only uses 100 Watts to transmit the massive amount of data -- about the energy required to power a lightbulb.

IBM plans on first deploying the technology commercially in the supercomputers market. A typical 100 ft electrical wire datalink uses 100 times more power than IBM's prototype,


Either I'm missing something or this is just PR talk. New technology requires 100W, old 100x that much. This is 10 kW. Unless the article means per bandwidth, it does not make any sense.
And unless you can scale it down (in speed and energy consumption), it is useful only for backbones.




RE: Energy consumption
By Duwelon on 3/2/2008 4:51:35 PM , Rating: 2
IBM's tech is the "new" tech.


RE: Energy consumption
By PlasmaBomb on 3/3/2008 2:34:02 PM , Rating: 2
The facts are from IBMs press release although slightly garbled (bad Jason!) foot =/= metre ;)

quote:
For a typical 100 meter long link, the power consumed by the optical technology is 100 times less than today's electrical interconnects, and offers a power savings of 10 times over current commercial optical modules.


You are correct though, they are talking in PR...
quote:
Compared to current commercial optical modules the transceiver provides 10x greater bandwidth in 1/10 the volume while consuming comparable power.


Thinking green...
By dflynchimp on 3/2/2008 2:44:53 PM , Rating: 2
is unfortunately a secondary concern to most companies. Ultimately they will see this as a cost reduction benefit, and will probably end up using the money saved to buy other energy guzzling hardware...




RE: Thinking green...
By Duwelon on 3/2/2008 3:36:09 PM , Rating: 3
The more "green" catches on, the more companies will look to it for marketing purposes. Banks like Citibank for awhile now have pushed paperless statements under the marketing of 'green', also saying they plant a tree for every customer that switches.

Personally I don't believe in all the global warming crap but I am a huge, huge fan of finding more energy effient ways of doing things as long as it's practical.


RE: Thinking green...
By dever on 3/3/2008 1:43:04 PM , Rating: 2
Please, Please, Please, I beg... stop using the silly "Green" moniker for technology.

Usually by definition, if a technology is more advanced than an older technology, it can do a task better, faster, etc. This means that it's more efficient at performing the task. Is this "green" or just another advance in technology?

Some would like to call this "green" because it gives them warm fuzzies for some reason. Recognize that the "green" label is part of an agenda to demonize our wonderfully advanced style of living.

Like all intelligent beings, I will seek out more efficient products and services, because it will benefit me through lower costs and my own higher productivity and efficiency. Am I being "green" or just human?

We can, of course, label every little advance as "green," but it's tiring, especially to those of us who don't subscribe to the theory that human activity is evil and freedom is overrated.


Keep an eye on this.
By mvrx on 3/2/2008 3:19:08 PM , Rating: 4
If IBM truly masters this as a low cost solution, you could easily say goodbye to PCIe 3.0. Imagine your motherboard coming with a bunch of optical interface ports (perhaps using some kind of high density 4 to 8 strand miniplug), each running at 100gbit/sec. No more need for PCIe 2.0 16x slots for each GPU card. You need to expand? You buy a matching PC case that is nothing but a bunch of slots, a power supply, and an optical bridging chip. Drop 12 GPU's in the new case, optically interconnct them to your core PC, and off you go.

While this has tons of applications for the enthusiast, there is one that really excites me. Think of the possibility of buying two PCs, interconnecting them at native bus speeds via cheap optical interconnect, booting up the HyperVisor (VMWare, IBM, Pheonix Hyperspace, etc) and configuring your two or more PCs to act as a single resource pool. You boot up Vista and see all the resources of both machines as a single PC image.

This could be a reality pretty easily if IBM got some of the industry together and went for it.

Modularize PC resources, combine them with an insanely fast optical bus, and the door is opened for enthusiasts to build core home servers that do it all - including play multiple games at once on thin clients on your HDTVs.

I predict that by 2010, IBM finds some way to break into the graphics processing world like Nvidia wants to do with Tesla. 32nm Cell B.E. processors with 32-64 cores, native h.264 encode/decode capable of super-high-def processing, etc - interconnected by nothing more than fiber to a host computer or better yet a cluster. IBM isn't just developing this for DARPA.

Sony wants optical interconnection for the playstation 4. Imagine if the PS4 came with a 10GbE, 40GbE, or even 100GbE interface with full TCP/IP Offload Engine (TOE) and a 32nm Cell processor with a 4+ core Power Processing Element (PPE) that more resembled an actual POWER6 core. I think plenty of HPC entities would be attracted to building TOP500.org beating clusters at <$500 each node.

Yea I'm a dreamer... but as I said. Stay tunned.




RE: Keep an eye on this.
By EricMartello on 3/2/2008 3:57:59 PM , Rating: 3
I'm all for the idea of modular computing, in fact, I've often wondered why computers aren't moving in that direction more than they already are.

I think the key problem with switching from local bus to serial bus is latency and its effect on the efficiency of the system. A serial bus uses a single access point, like a chip, to control all activity on the bus. IN other words it has a "middle man". While high bandwidth is capable, the speed at which the bus responds to commands is limited to the speed of the controller chip AND latency will always be present due to the fact that all commands are being funneled thru a controller chip.

A local bus, on the other hand, gives up convenient expandability and compatibility for raw speed. Local buses still have a controller chipset, but the CPU can bypass this chip (bus mastering) and communicate directly with the device connected to the bus - and there is zero latency added to the system from the local bus. When you are doing millions of calculations per second, even minuscule increases in latency can sap the overall computing power of a system quite a bit.

Like many things with computers there are always trade-offs between compatibility, performance and price...but I do think that we will have systems that are "decentralized" and modular in the near future. Whether or not we are using optical interconnects and circuitry is unclear, and I'd say unlikely. PCIe 3.0 is still 'safe' in the near future, and eventually we may see an optical PCI spec being drafted and put into action.


Ha!
By murphyslabrat on 3/2/2008 7:36:34 PM , Rating: 2
quote:
This will allow handheld manufacturers to build extremely small and low voltage devices capable of high definition video on incredibly small power envelopes.


OK, who the hell wants to watch 300 in 1080i/p on a 3" screen? Worse yet, who could possibly be silly enough to clog up a 30GB HDD with an HD movie?

I would, howeever, be a fan of a cost-effective consumer-level 10Gbit Ethernet-over-Optical spec. 1.25 GBytes per second far exceeds current HDD transfer rates, but the 1GB Ethernet spec isn't maxing HDD speeds...