backtop


Print 38 comment(s) - last by GTMan.. on Mar 28 at 5:12 AM


IBM's optical chipset is not only fast, but also tiny
DARPA helps fund IBM's blazing optical chipset

IBM scientists said they will demonstrate this week at the 2007 Optical Fiber Conference a prototype optical transceiver chipset capable of reaching speeds at least eight times faster than optical components available today. The chipset is able to move information at blazing speeds of 160 Gigabits per second by using light pulses to speed the flow of data, instead of sending electrons over wires.

The transceiver is fast enough to reduce the download time for a typical high definition feature-length film to a single second compared to 30 minutes or more, according to IBM. Furthermore, the technology could be integrated onto printed circuit boards to allow the components within an electronic system – such as a PC or set top box – to communicate much faster, dramatically enhancing the performance of the system itself.

To achieve this new level of integration in the chipset, IBM researchers built an optical transceiver with driver and receiver integrated circuits in current CMOS technology, the same standard, high-volume, low-cost technology used for most chips today. They then coupled it with other necessary optical components made in more exotic materials, such as indium phosphide (InP) and gallium arsenide (GaAs), into one, integrated package only 3.25 by 5.25 millimeters in size.

This compact design provides both a high number of communications channels as well as very high speeds per channel, resulting in an amount of information transmitted per unit area of card space taken up by the chipset (the ultimate measure of viability for practical use) that is the highest ever. This transceiver chipset is designed to enable low cost optics by attaching to an optical printed circuit board employing densely spaced polymer waveguide channels using mass assembly processes.

“The explosion in the amount of data being transferred, when downloading movies, TV shows, music or photos, is creating demand for greater bandwidth and higher speeds in connectivity,” said Dr. T.C. Chen, vice president, Science & Technology, IBM Research. “Greater use of optical communications is needed to address this issue. We believe our optical transceiver technology may provide the answer.”

This work was partially funded by Defense Advanced Research Project Agency (DARPA) through the Chip to Chip Optical Interconnect (C2OI) program.



Comments     Threshold


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

By peternelson on 3/27/2007 2:19:33 AM , Rating: 2
This is claimed a HD movie in 1 second at 160 gigabit/s
which from that means their movie is 20GByte file size.

Sure that's fast.

Let's consider commodity off the shelf tech (even without optical) is 10 Gigabit/s (expensive yes but readily available). That is 16 times slower.

So my move will take 16 seconds and I can buy that today.

Let's even go with the 1 Gigabit tech found on nearly every motherboard and many home switches, and costs peanuts.

That is about 16x10=160 secs = 2 minutes 40 seconds.

These are very different from the supposed comparison with HALF AN HOUR to download that movie that the article claim is based on.

This demonstrates it is not the technical bandwidth but other factors like cost of fiber routes, transatlantic cable laying, ISP profits, telco profits, government legislation and taxes, that are preventing us enjoying high download speeds, except within our own home or campus from a local server.

As has been pointed out, the computer needs to be able to keep up with that much data. And even if it can, it needs delivering off some storage medium with the speed required.

It's cool for certain applications like high-end core switching, and chip to chip interconnect of scientific sensor arrays, but this won't be changing your movie downloads anytime soon.




By peternelson on 3/27/2007 2:23:24 AM , Rating: 2
So, the question is on what slow line speed is the massive speedup claim based?

For bonus points, can you use the conventional metric of "how long would it take to transfer the entire encyclopedia britannica or library of congress" through this link?


By ahkey on 3/27/2007 4:25:17 AM , Rating: 3
I think that you're confusing theoretical speeds with actual results. A gigabit Ethernet port has a theoretical speed of up to 1gbps. Try copying a 20gb file and see how long it really takes. Actual transfer speed will in no way approach the theoretical limit.

The article states that the chipset has transferred at 160gbps. So unless factually wrong, we may understand that real-world performance is on this level.


By napsi on 3/27/2007 6:15:40 AM , Rating: 2
Gbit ethernet can do speeds very close to the theoretical limit.

A quick test with my OpenBSD firewall (P3 1GHz, Intel Gbit dual port PCI-X NIC) gave me about 850Mbit/s UDP passing through the machine (not sure how much protocol overhead iperf includes in those numbers). :)

Tests with machines not acting as firewall: http://www.eslea.uklight.ac.uk/cookbook/Performanc...

They claim 995Mbit/s for Gbit Intel NICs. :)


By theapparition on 3/27/2007 8:15:35 AM , Rating: 2
GigE does perform very close to theoretical limits. However, there is some overhead that occurs during transfers, such as send/receive confirmations, that lowers effective bandwidth. I doubt that this new technology will be any different with lower effective bandwidth.

This is technology that is ~20years away from the CONSUMER market. As another poster mentioned, I can buy, right now, a 10Gb/s Fiber Channel director, with an IBM T-TEX mainframe attached to a FC storage array that can transfer to storage at ~10Gb/s. Cost is way prohibitive. Still, that is not the bottleneck, it is the internet pipes themselves.

Cracks me up when everyone wants faster wireless routers, GigE, 10GigE, and then hook them up to Cable or DSL with max 6Mb/s.


By Lakku on 3/27/2007 11:07:10 AM , Rating: 3
Cracks you up, eh? What cracks me up is people who can't think outside their own little world.

As a relatively not so well to do college student, I have found quite the benefit from having a GigE LAN attatched to my wireless router, which in turn is attatched to my 10Mb/s maximum internet connection. Do I care about the GigE for the internet? Of course not, because I'm not stupid, but I have 2 laptops, a couple PCs, a PDA, some NAS storage and game consoles. I don't use all of it at once, but I do do a lot of streaming of video and audio content. Some of it HD video. This is why some people desire GigE and I guess, 10GigE. Ever tried to back up gigs of data over a 10Mb or 100Mb network? Ever tried to send gigs of data to a laptop over 10 or 100Mb ethernet? It takes awhile to say the least. So if I have found a use for it, others have as well, so don't just assume its funny that all us "n00bz hook up our super fast networking stuff to our lame internet connection" as if we have no need for it.


By peternelson on 3/27/2007 11:55:49 AM , Rating: 2
You could do as I do and not use a lame embedded switch in your ADSL router to do the switching at all. Instead get a standalone switch which will give you more ports, maybe manageability etc.

Any minor delay in two propagation types when going to the wan will be masked by the much lower speed anyway. On the upside, the switching in a dedicated switch may be slightly faster than the cheapo switching chip used in the router box.


By peternelson on 3/27/2007 11:49:01 AM , Rating: 2
Gigabit ethernet on PCI bus is usually constrained by either computer cpu OR the single direction at a time PCI, and the fact that pci bus is usually shared with other devices.

650+ megabits is achieveable even with lame cards.

If you attach your gigabit using pci express, you have full duplex, separate links for each device.

If you use a good card it may well have some onboard acceleration.

Further you can enable jumbo frames so that more traffic is transferred without excessive interrupts on the cpu.

Such cards are full gigabit wire rate.

At 10 gigabits per second, wire rate is also achieveable.

A $799 Myrinet 10gigabit card (including acceleration) has been benchmarked at 9.9+ out of 10 and you have to remember that protocol overhead like headers, interframe gap will eat a small percentage.

Therefore I maintain that full rate 1Gbps is achieveable and also 10Gbps if you have the money.

10Gbps switches are still quite pricey, but you get what you pay for.

Some internet backbone between datacentres (eg 3 in manchester) use 10Gbps ethernet in the WAN (optically). It's also possible to use DWDM to achieve say 40Gbps or more down a fibre using conventional techniques.

Most peering at exchange points still occurs at 1gbps.

One bottleneck is between those centre and your house. Another is the ISP doesn't want to give you that much data because they pay transit fees for their upstream feeds. That's why ISPs often rate limit or ban things like torrents. Therefore the network tech itself is not the limiting factor but commercial issues.

Of course nothing to stop you deploying this fast tech yourself.


By GTMan on 3/28/2007 5:12:06 AM , Rating: 2
Why are you people discussing NICs and bandwidth etc. The point is chip to chip interconnects. CPU to memory, CPU to GPU, etc. They only mentioned how long it would take to download a movie so pseudo techies could understand how fast this is.


By obeseotron on 3/27/2007 11:54:43 PM , Rating: 2
Your hard drive is what isn't capable of anything near 1Gbps, not your NIC. Fast SATA drives average no better than 50-60MB/sec (not even half a Gbps) and that's only if the file is totally sequential physically on the disk.


By GotDiesel on 3/27/2007 11:44:38 AM , Rating: 2
They cost peanuts ??.. well they cost LESS than
peanuts these days... !!!

Peanuts at least in north America are more expensive.. LOL


By peternelson on 3/27/2007 12:03:13 PM , Rating: 2
Here in the UK we don't get that much sun for growing.

So I believe we have to import our peanuts, adding to the costs ;-)

I prefer cashew nuts but they cost even more.


By FITCamaro on 3/27/2007 12:52:44 PM , Rating: 2
You assume that hard drives are capable of those read and write speeds. That in the end is the true bottleneck. Not the network.


By PrezWeezy on 3/27/2007 2:07:33 PM , Rating: 2
You are correct in saying that in a LAN it doesn't seem to make a huge difference. Let's forget overhead for a second and assume everything works like it's suposed to. However, what about the fact that when phone companies run fiber lines to residences they can run 1/4 of the lines (I believe MCF is still 40gbps), thereby reducing the cost of the overal job. Or even increasing our speed to the home 4 times. And if you've ever tried to download a 20GB movie online you are talking about a half hour to an hour. And just remember, people at one point thought they would never need more storage than their 1.44 MB floppy. Things get bigger, and when they do we need more bandwidth to push them.


Great!
By Lifted on 3/26/2007 8:04:58 PM , Rating: 4
Now we just need those 160Gbps routers and switches and we're good to go.




RE: Great!
By AnotherGuy on 3/26/07, Rating: 0
RE: Great!
By AnotherGuy on 3/26/07, Rating: 0
RE: Great!
By osalcido on 3/26/2007 10:21:39 PM , Rating: 2

this technology (as I interpreted the article) has nothing to do with information storage


RE: Great!
By Macazian on 3/27/07, Rating: 0
RE: Great!
By spartan014 on 3/26/2007 11:18:04 PM , Rating: 2
But which storage medium will be able to singlehandedly provide that much data rate? I think you will still take more than just a second to transfer a full length HD movie..

What can be safely said is that the term " Broadband " will be redefined.. You'll be able to access any file over internet as though it resided on your machine.


RE: Great!
By alienbibin on 3/27/2007 12:00:55 AM , Rating: 2
Holographic storage is the way to go. When laser is used for communication, it can be routed to optical amplifiers by this transceiver and inturn used to write to holographic media directly. Here i'm speaking about storage in the bulk material and not onto the holographic discs. This type of storage has the potential to store huge amounts of data in a small space and at high rates.
The technology is still under development, i guess! But heard about a holographic disc being developed that can store 300GB, anyway they still can't handle such data rates.In the future they will...


RE: Great!
By therealnickdanger on 3/27/2007 11:03:57 AM , Rating: 2
Holo-discs will never be fast enough (including latency) to take advantage of these interconnects. A much more advanced form of SSD or cellular (biological cells) storage would probably be quick enough. But we're still talking several years in the future.


RE: Great!
By FITCamaro on 3/27/2007 12:53:59 PM , Rating: 2
Also need hard drives capable of about 200x their current average read/write rates.


Wat abt ISPs ?
By shady3005 on 3/27/2007 1:35:10 AM , Rating: 3
No matter how fast the technology progresses or how cheap it turns out for them but these fu&%ng ISPs will neither increase the speeds nor cut the prices on existing speeds.

Here in INDIA i pay arnd 25$/month for a 256kbps(32KB/s) unlimited line. They put caps on cheaper subscriptions like 200-400 MB download limit per month. Even with dozens of players no one is willing to lower charges. It's like a big cartel here :(




RE: Wat abt ISPs ?
By videoclone on 3/27/2007 1:53:57 AM , Rating: 2
Me and two other flatmates pay $99AU per month in Australia for ADSL2+ 24000/1000kbps, thats 2.4Mbs download a second and 100kbps upload with 80GB per month limit, after that we get caped! to 74/1000 lol :)

We DO get the full 2.4mbps speeds from sites that can dish it out But most cap users at 300kbps :)

The net costs allot of money anywhere you go, it just works out cheaper when you share the cost with others.


RE: Wat abt ISPs ?
By Bladen on 3/27/2007 6:18:47 AM , Rating: 2
AFIAK most if not all Aussie ISP' cap at 64/64, 128/128, or if you are lucky, 256/256.

Sounds like your ISP is cheap, but doesn't give the best shaping speed.

Do you often d/l more than 80gb a month?


Ht or FSB applications?
By verndewd on 3/27/2007 1:53:14 AM , Rating: 2
What could this do for a mobo on the fsb or AMD's HT links.




RE: Ht or FSB applications?
By otispunkmeyer on 3/27/2007 3:45:20 AM , Rating: 2
i could see it being good for the HT link perhaps, though last time i looked HT speed didnt have too much affect on system performance...(might be different now)

i dont think it could be used for memory interfaces just yet, the memory wouldnt be able to keep up and it'd be pointless.

but for moving data round CPU's/GPU's internally, it might work


so how long....
By otispunkmeyer on 3/27/2007 3:40:40 AM , Rating: 2
until we see this type of tech in a CPU or a northbridge?

i understand this is like a CMOS-Optical hybrid chip, some parts in standard copper wire, other parts in optical.

could help in GPUs? ATi could make a ring bus from this hybrid to shunt data around the chip. that would be stellar




IBM press release
By crystal clear on 3/27/2007 9:12:28 AM , Rating: 2
IBM Researchers Demonstrate World's Fastest Optical Chipset
Prototype Technology Points to New Era of Data Sharing, Instant Movie Downloads

http://www-03.ibm.com/press/us/en/pressrelease/212...




Thats great and all, but...
By clemedia on 3/27/2007 10:00:48 PM , Rating: 2
What about the interconnect between this chip and the rest of the computer? Can you say HUGE bottleneck?




Leave it to our tax dollars...
By Noya on 3/26/07, Rating: -1
RE: Leave it to our tax dollars...
By cplusplus on 3/26/2007 8:23:05 PM , Rating: 2
But to me, this is a perfect example of the type of DARPA funded project that could easily have great consumer benefits in the next 10-20 years. Cutting edge research tends to be very expensive.


RE: Leave it to our tax dollars...
By Red and black on 3/26/2007 9:41:43 PM , Rating: 4
Yeah, but by that time the killer robots will have finished off everyone opposed to military rule.


RE: Leave it to our tax dollars...
By Tsuwamono on 3/26/2007 11:33:42 PM , Rating: 1
agreed... cut all funding to this and we should start building killer robots to defend us... we should have 4 killer robots per human for defense against any other killer robots who try to hurt us...


RE: Leave it to our tax dollars...
By vorgusa on 3/27/2007 10:56:00 AM , Rating: 2
Yeah but we could have these chips in the killer robots so they can download music faster and we can pimp our robots out with some speakers to listen to while they are defending us. This of course will lead to reality TV shows called pimp my robot


RE: Leave it to our tax dollars...
By glennpratt on 3/27/2007 1:51:12 AM , Rating: 2
RE: Leave it to our tax dollars...
By Noya on 3/27/2007 1:37:36 PM , Rating: 1
Thanks for the budget poster. Talk about a militaristic country. 65%+ of the national budget, and we're surrounded by water, Canada, and Mexico.


"You can bet that Sony built a long-term business plan about being successful in Japan and that business plan is crumbling." -- Peter Moore, 24 hours before his Microsoft resignation











botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki