Intel, Corning show off 1.6Tbps interconnect for servers


Intel's (INTC) new MXC server interconnect cables leverage the chip giant's silicon photonics tech and optical fibers developed by Corning (GLW) to enable 1.6Tbps transfer speeds (800Gbps each way) via 64 fibers running at 25Gbps apiece. Traditional solutions work at 10Gbps, and max out at 12-24 fibers.

Corning plans to begin selling MXC cable assemblies, which have a range of 300 meters, in Q3. Intel asserts the technology's per-fiber throughput could eventually double to 50Gbps. The companies think supercomputer (HPC) vendors and cloud computing firms will be among the first buyers.

Over time, MXC's blistering speeds stand to further Intel's goal of creating server architectures in which CPU, memory, storage, networking, and other resources can be separately pooled and upgraded as needed. "The ability to take my memory and stash it a rack away, optical can enable that," says Intel exec Mario Paniccia.

Intel's efforts pose a challenge to optical component vendor Finisar (FNSR), whose shares sold off a year ago on worries about Intel and Cisco's silicon photonics efforts. They're also a challenge to high-speed interconnect leader Mellanox (MLNX), which last year acquired silicon photonics startup Kotura and 100G interconnect developer IPtronics. Since then, Mellanox has said it will ship 100G silicon photonics products in late 2014 or early 2015.

From other sites
Comments (9)
  • Arnold Frisch
    , contributor
    Comments (821) | Send Message
     
    So, I have been advising clients that I thought that Intel would quickly move into optical interconnect, based upon published information about their developments.

     

    Looks like it was good advice!

     

    Also, I advised about moves to 3D in order to preserve Moore's Law. Let's see where that goes.
    11 Mar 2014, 07:06 PM Reply Like
  • King Rat
    , contributor
    Comments (1741) | Send Message
     
    If the advice corresponds with bullish for Intel... sorry, there is not enough revenue in this market to make a material difference in INTC's EPS.

     

    If the advice is to avoid being long in their smaller competitors, yes, that is good advice. I personally believe INTC is undervalued (to the market), but for a host of unrelated reasons.
    11 Mar 2014, 11:48 PM Reply Like
  • sbrncra
    , contributor
    Comments (258) | Send Message
     
    great research funding creates great products even though the stock
    stalls, moving up from here as $intc has an improving portfolio high tech outlook
    11 Mar 2014, 07:21 PM Reply Like
  • Just Some Guy
    , contributor
    Comments (2451) | Send Message
     
    Transfer is not latency, and this zippy technology can't be cheap.

     

    Fast always finds a market, but I think it's evolutionary not revolutionary.
    11 Mar 2014, 07:26 PM Reply Like
  • Arnold Frisch
    , contributor
    Comments (821) | Send Message
     
    @Just Some Guy
    "Transfer is not latency, and this zippy technology can't be cheap.
    Fast always finds a market, but I think it's evolutionary not revolutionary."

     

    Cheap is a relative thing. Getting the bandwidth with wires is an expensive proposition. It means increasing the chip size to create more I/O's. Making lots of high speed channels that need equalization. More careful layout. More power, and a bunch of other things, all of which cost money. optical I/O can be a bargain.

     

    And,...... everything that gets into mass production is, by definition, evolutionary. It's the research that's revolutionary!
    12 Mar 2014, 11:21 AM Reply Like
  • gulkushniba
    , contributor
    Comments (2) | Send Message
     
    Does using Silicon photonics and 1300nm wavelength imply an onchip source or is the laser off chip ? Also, does each channel have its own laser or is there one master laser that is split off into the 64 channels ? It would interesting to see how the cost figures turn out for this technology. IBM by contrast has been using more " conventional " VCSEL based optical links for interconnects on their Supercomputers for sometime now, even though IBM research has been working on Silicon photonics for awhile ! Does this say something about IBM's problems or INTEL's superior scheme ? Just asking !
    11 Mar 2014, 09:21 PM Reply Like
  • MattZN2
    , contributor
    Comments (1260) | Send Message
     
    I don't know but likely on-chip. And remember that IBM and Intel have been collaborating on photonics stuff and IBM has been using Intel's FABs.

     

    My guess this is intended to become an extension of Intel's PCIe. The idea is to finally be able to divorce storage subsystems from cpu subsystems in data centers, which would lead to significant improvements in density. At the moment storage can't be moved far enough away from the cpus, 10GigE and 100GigE aren't fast enough and the latency is too high.

     

    -Matt
    12 Mar 2014, 01:15 AM Reply Like
  • hschaier
    , contributor
    Comment (1) | Send Message
     
    IBM has been sluggish on doing anything with their technical ability these past 15 years. The reality is they let Dr. Alan Gara of IBM. IBM has nothing to show, nothing. They are not doing any research in hardware / servers. The IBM CEO wants to do paper stuff. Say so long to IBM soon they will be out priced by the competition.

     

    Inside Track: Blue Gene Architect Alan Gara Leaving IBM for Intel
    April 25, 2011 by Rich Brueckner

     

    insideHPC has confirmed that IBM Fellow Dr. Alan Gara is leaving the company to join Intel. As the chief system architect of the three generations of Blue Gene supercomputers, Gara was an IBM Fellow at T.J. Watson Research Center and was leading exascale system research at IBM.

     

    As the 2010 recipient of the Seymour Cray Award, Gara has had a storied history in HPC. His Blue Gene/L system was the #1 system on the top500 list for 5 consecutive dates. The Blue Gene/P and the latest Blue Gene/Q systems both debuted as #1 in terms of energy efficiency on the green500 list. Alan Gara received his PhD in physics from the University of Wisconsin, Madison in 1987. Dr. Gara has received two Gordon Bell Awards in each of 1998 and 2006 for his scientific work in supercomputing.
    12 Mar 2014, 01:04 PM Reply Like
  • Pwuff
    , contributor
    Comments (37) | Send Message
     
    Dayam, GLW! That is fast!
    12 Mar 2014, 08:09 AM Reply Like
DJIA (DIA) S&P 500 (SPY)
ETF Hub
ETF Screener: Search and filter by asset class, strategy, theme, performance, yield, and much more
ETF Performance: View ETF performance across key asset classes and investing themes
ETF Investing Guide: Learn how to build and manage a well-diversified, low cost ETF portfolio
ETF Selector: An explanation of how to select and use ETFs