Slashdot Mirror

← Back to Stories (view on slashdot.org)

Intel To Ship 48-Core Test Systems To Researchers

Posted by timothy on from the but-one-link's-enough-really dept.

MojoKid writes "Just when you thought your 6-core chip was the fastest processor on the planet, Intel announces plans to ship systems equipped with an experimental 48-core CPU to a handful of lucky researchers sometime by the end of the second quarter. The 48 cores are arranged with multiple connect points in a serial mesh network to transfer data between cores. Each core also has on-chip buffers to instantly exchange data in parallel across all cores. According to Sean Koehl, technology evangelist with Intel Labs, the chip only draws between 25 and 125 watts."

45 of 135 comments (clear)

  1. I just have to ask by toygeek · · Score: 4, Funny

    Can you imagine a *Beowulf cluster* of these things!? Think about the possibilities!

    --
    Nobodies Prefect
    Tidbits for Techs Technology Blog
    1. Re:I just have to ask by Anonymous Coward · · Score: 5, Interesting

      Yes, I can think of the possibilities...

      A Jaguar (or Roadrunner) of these processors would still be too slow to numerically solve the geomechanics problems I grapple with daily though. A Jaguar equipped with these processors would be approximately 20 petaflops peak. To simulate 1 sec of fracture of a 10mm cube of rock on the atomic scale would require of order 10^36 floating point operations. To do that would take 10^20 sec at 10 petaflops. Not bad really...that's only 10^12 years. Oh wait, the universe hasn't even been around that long...

      Having said that I'm a researcher who writes and uses high-performance parallel software daily. How might I become one of Intel's select few to trial these chips? I can certainly think of ways to keep them warm!

      Please Intel please! ;)

    2. Re:I just have to ask by MrMr · · Score: 4, Funny

      fracture of a 10mm cube of rock on the atomic scale
      Ha, I can do that in less than a second, with my serial mallet.

    3. Re:I just have to ask by urusan · · Score: 3, Funny

      It certainly beats the 10^13 years it would take with a Jaguar!

    4. Re:I just have to ask by Jurily · · Score: 5, Insightful

      Now 48 CPUs can wait for the disk!

    5. Re:I just have to ask by WrongSizeGlass · · Score: 4, Funny

      How might I become one of Intel's select few to trial these chips? I can certainly think of ways to keep them warm!

      Please Intel please! ;)

      Well, posting as AC certainly will help your chances.

    6. Re:I just have to ask by anarche · · Score: 2, Insightful

      Having said that I'm a researcher who writes and uses high-performance parallel software daily. How might I become one of Intel's select few to trial these chips? I can certainly think of ways to keep them warm!

      ummm, lets start by not explaining why one of these things won't help your research?

      --
      Wait! Whats a sig?
    7. Re:I just have to ask by leromarinvit · · Score: 3, Funny

      fracture of a 10mm cube of rock on the atomic scale
      Ha, I can do that in less than a second, with my serial mallet.

      You left out an important detail: 1 sec of fracture of a 10mm cube of rock on the atomic scale

      Whoa! Chuck Norris has a Slashdot account?

      --
      Proud member of the Ferengi Socialist Party.
    8. Re:I just have to ask by Jurily · · Score: 2, Insightful

      That doesn't really help me as long as people are doing disk I/O in the GUI thread.

    9. Re:I just have to ask by White+Flame · · Score: 2, Insightful

      Now 48 CPUs can block on cache misses!

      fixed. Memory is the new disk.

    10. Re:I just have to ask by haruchai · · Score: 2, Insightful

      Let me save you some time, trouble and hair - your problems are too difficult to solve numerically. Go find something else to do
      unless you plan to live for a very, very long time.

      --
      Pain is merely failure leaving the body
  2. Larrabee by TheKidWho · · Score: 5, Interesting

    I believe this is the remnants of Intel's failed Larrabee chipset which was supposed to compete with Nvidia and ATI.

    A nice article on the story behind Larrabee and it's failure:
    http://www.brightsideofnews.com/news/2009/10/12/an-inconvenient-truth-intel-larrabee-story-revealed.aspx

    1. Re:Larrabee by Anonymous Coward · · Score: 2, Informative

      Indeed. Or, to be more precise, it's their "Bangalore" chip which is basically the same thing as Larrabee without the graphics-specific subunits (texture unit) and perhaps the fancy-pants cache coherency / ring bus architecture.

    2. Re:Larrabee by jmknsd · · Score: 5, Interesting

      No, actually It is basically a bunch of Pentium 3s with cache coherency removed for a small chunk of on chip RAM, and a message passing interface for inter core communication. It has alot of interesting features, and is more usable than the 80 core chip they came out with a few years ago.

    3. Re:Larrabee by Anonymous Coward · · Score: 2, Insightful

      Yeah. that's what I was thinking too. It suddenly becomes a whole lot less exciting when you consider that it's just a 48 core first generation Pentium rather than a 48 core i7.

    4. Re:Larrabee by Anonymous Coward · · Score: 5, Funny

      That's not a particularly auspicious name for a chip. I'd assume that a "Bangalore" CPU would promise that it could get the work done twice as fast for half as much money due to "parallel architecture" - but you'd launch a program, only to discover that it actually took 10x as long, every instruction needed to be told *exactly* what to do, and the results were so full of errors that it took an additional non-Bangalore CPU working full time just to get things right.

    5. Re:Larrabee by Spy+Hunter · · Score: 2, Informative

      No, actually this is a separate effort entirely. This is a product of the same group which produced the "Polaris" 80-core chip, and is meant for research into communication models and memory architectures for massively parallel systems.

      Larrabee is still ongoing as a separate project with a different focus. Larrabee is all about getting maximum throughput by adding a wide vector unit with a whole new instruction set to each x86 core. As far as anyone outside Intel knows, the plan is still to eventually release some Larrabee prototypes as-is (with the texture units and everything), and to develop a Larrabee 2 with the lessons learned that can actually compete directly with GeForce and Radeon in the graphics card market.

      --
      main(c,r){for(r=32;r;) printf(++c>31?c=!r--,"\n":c<r?" ":~c&r?" `":" #");}
  3. bullshitter by Anonymous Coward · · Score: 3, Insightful

    >>> Sean Koehl, technology evangelist
    Oh... a bullshitter

  4. And it runs Linux by Macka · · Score: 2, Informative

    According to the video they're running Linux on this thing with a custom kernel. No specific details on the changes they had to make to get it running yet.

    1. Re:And it runs Linux by klingens · · Score: 2, Insightful

      If they distribute it to the researchers they must release it to the researchers or commit a GPL violation.

      Of course the researchers don't want to demand source since then they won't get freebies like this the next time Intel does such a Santa Claus imitation of distributing presents.

      There's an interesting thought: what happens if you are a beta tester who has to sign a NDA to get something which includes GPL code. What takes precedence? Your NDA, or your right to demand source to the GPL stuff and redistribute it publically?

    2. Re:And it runs Linux by solevita · · Score: 2, Informative

      or your right to demand source to the GPL stuff and redistribute it publically?

      But this is all just wrong. There's no requirement to make GPL code public, you only need to make it available to the people that receive the binaries. So the researchers will likely be given some source code, but nobody has to release that to the rest of us.

    3. Re:And it runs Linux by mdwh2 · · Score: 2, Interesting

      http://www.gnu.org/licenses/gpl.html , section 10:

      Each time you convey a covered work, the recipient automatically receives a license from the original licensors, to run, modify and propagate that work, subject to this License. You are not responsible for enforcing compliance by third parties with this License.

      An entity transaction is a transaction transferring control of an organization, or substantially all assets of one, or subdividing an organization, or merging organizations. If propagation of a covered work results from an entity transaction, each party to that transaction who receives a copy of the work also receives whatever licenses to the work the party's predecessor in interest had or could give under the previous paragraph, plus a right to possession of the Corresponding Source of the work from the predecessor in interest, if the predecessor has it or can get it with reasonable efforts.

      You may not impose any further restrictions on the exercise of the rights granted or affirmed under this License. For example, you may not impose a license fee, royalty, or other charge for exercise of rights granted under this License, and you may not initiate litigation (including a cross-claim or counterclaim in a lawsuit) alleging that any patent claim is infringed by making, using, selling, offering for sale, or importing the Program or any portion of it.

      So the way I see it, either they have the right to redistribute, or Intel would be committing copyright infringement.

      And it would be an obvious loophole if a contract, NDA, EULA or anything else could trump the GPL, because then people would just distribute a GPL application with an additional bit saying "Actually you have to agree not to redistribute this application".

    4. Re:And it runs Linux by TheRaven64 · · Score: 2, Insightful

      What takes precedence? Your NDA, or your right to demand source to the GPL stuff and redistribute it publically?

      It's complicated. The GPL takes priority because the NDA is in violation of it, however the problem is that, by making you sign the NDA, Intel[1] is in violation of the GPL, but you do not have standing to sue them for it. The kernel developers could sue Intel for copyright infringement, by distributing their code without a valid license to do so. Intel could sue you for breach of the NDA. You, however, would have no recourse against Intel if you chose to distribute the code.

      In fact, Intel could distribute the kernel under some terms other than the GPL, preventing you from distributing it legally. They would also be committing copyright infringement, but only the original copyright owners would have standing to sue, it wouldn't help you at all if they decided not to bother.

      [1] Names of organisations in this post are hypothetical placeholders. I am not accusing Intel of GPL violation.

      --
      I am TheRaven on Soylent News
  5. 640 C (cores) should be enough for everybody by youn · · Score: 5, Funny

    maybe that's what bill gates meant when he said 640K should be enough... K as in Core .. it was a spelling mistake;)

    --
    Never antropomorphize computers, they do not like that :p
    1. Re: 640 C (cores) should be enough for everybody by selven · · Score: 3, Funny

      With the kinds of things Bill Gates did, I don't know if even 640' C would be enough.

    2. Re: 640 C (cores) should be enough for everybody by TeknoHog · · Score: 2, Funny

      Too bad multiprocessing did not exist back then, as Intel had yet to invent the Core.

      --
      Escher was the first MC and Giger invented the HR department.
  6. Can you imagine... by EnsilZah · · Score: 3, Insightful

    ...a Beowulf cluster of engineers awkwardly reading marketing information from a teleprompter?

  7. Is the 48-core chip the same as the one ... by Taco+Cowboy · · Score: 2, Informative

    ... reported in this Slashdot entry - http://hardware.slashdot.org/article.pl?sid=09/12/02/215207 ??

    --
    Muchas Gracias, Señor Edward Snowden !
  8. Tilera by loufoque · · Score: 3, Informative

    Might as well buy a Tilera if it's for research...
    The only good thing about x86 is that it runs legacy Windows programs, but who cares about that in research?

    1. Re:Tilera by oakgrove · · Score: 2, Insightful

      Just because source code is available... Even ironing out x86-64 (which millions of people can use) has taken years for the linux distros.

      Interesting. See, I was running 64 bit distros when Vista was still called Longhorn. I'm also quite sure it was a year or so before XP x64 edition was released. And everything ran great. The only problem I had was flash. Of course, I had the source for everything except for, you guessed it, flash. Imagine that.

      --
      The soylentnews experiment has been a dismal failure.
  9. Correction by Anonymous Coward · · Score: 5, Informative

    "Just when you thought your 6-core chip was the fastest processor on the planet, Intel announces plans to ship systems equipped with an experimental 48-core CPU to a handful of lucky researchers sometime by the end of the second quarter.

    Actually, the 8-core (Nehalem EX) and 12-core (Opteron "Magny-Cours") CPUs are already faster than your 6-core CPU. And oddly enough, this 48-core CPU is actually slower than your 6-core, 8-core, or 12-core CPUs. Intel didn't design the 48-core CPU to sell it. They did it as a research project/experiment to develop new ways of interconnecting so many processing cores. While there are technically 48 cores they are far less complex and slower performing than anything that Intel is shipping retail. If you go back a year or two you can find articles where Intel unveiled the CPU and talked about performance. This is simply an exercise in massively parallel CPU design, not an effort to make a faster CPU. That's why they are shipping them to researchers, so they can study and learn how to develop uses for such massively parallel systems.

    1. Re:Correction by mdwh2 · · Score: 3, Insightful

      The fastest commercially available x86 chips have been a little under 4 GHz for about five years now.

      Megahertz Myth. As far as I can tell, over the last 5 years individual cores have still been getting faster, just not with higher clock speeds.

    2. Re:Correction by TheRaven64 · · Score: 2, Interesting

      Intel didn't design the 48-core CPU to sell it

      Actually, they did. Unfortunately, it was delayed and didn't work as well as they'd hoped and it would have been a complete flop in its original target market so they're shipping it as a research toy to try to recoup some of their investment.

      --
      I am TheRaven on Soylent News
    3. Re:Correction by Spatial · · Score: 2, Interesting

      Correct.

      Clock frequency is worthless as a measure of CPU performance. Cores have never stopped getting faster.

      For example: Each individual core in a 2.66Ghz i5-750 is more than twice as fast as a 3.8Ghz P4. Often many times faster than that, depending on the workload.

    4. Re:Correction by symbolset · · Score: 2, Interesting

      More likely they've got this widget sitting around with all the requisite engineers raring to go. But it's a wrench that fits no bolt - they need research scientists with the type of problems that this solves to put a load on it, define the scope of its use and put it to work so they can refine the toolchain and broaden the scope.

      --
      Help stamp out iliturcy.
    5. Re:Correction by RAMMS+EIN · · Score: 2

      ``This is simply an exercise in massively parallel CPU design, not an effort to make a faster CPU. That's why they are shipping them to researchers, so they can study and learn how to develop uses for such massively parallel systems.''

      Perhaps it would be interesting to mention Azul Systems at this point. They sell systems with 108 to 864 cores, so they may know a thing or two about "massively parallel".

      --
      Please correct me if I got my facts wrong.
  10. Video reminds me of ... by LoudMusic · · Score: 3, Funny

    3DFX, so powerful it's kind of ridiculous.

    http://www.youtube.com/watch?v=DmaYH1F6kho

    http://www.youtube.com/watch?v=ldiYYJNnQUk

    http://www.youtube.com/watch?v=o72T8qQr7GE

    Great ad campaign.

    --
    No sig for you. YOU GET NO SIG!
  11. Ghost of the Transputer by Mr+Pippin · · Score: 2, Interesting

    Sounds quite a bit like the INMOS Transputer http://en.wikipedia.org/wiki/Transputer
    Wonder what version of Occam (the programming language) will ship with it?

  12. Chip trial by Anonymous Coward · · Score: 5, Funny

    Hi,

    I'm an engineer at Intel and we are looking for a few more candidates to test our 48-core chips. Your scientific computing project sounds like a perfect fit for our trial. Please contact me (see my account info for my email address) and we'll get you in the program.

    Cheers!

  13. AMD already has this... by Skaven04 · · Score: 3, Informative

    AMD's new 12-core "Magny-Cours" Opteron parts will be available in 4P configurations with 48 cores and up to 512GB RAM, so...::yawn::

    --
    ---- Breakbeats are not just music...they're the soundtrack for my life.
    1. Re:AMD already has this... by Skaven04 · · Score: 3, Informative

      Actually 768GB RAM...12 dimms per socket (if an OEM chooses to max out the config) with 16GB DDR3 dimms == 768GB.

      --
      ---- Breakbeats are not just music...they're the soundtrack for my life.
  14. 48 cores by nurb432 · · Score: 4, Insightful

    And still one external memory bus.

    --
    ---- Booth was a patriot ----
    1. Re:48 cores by Anonymous Coward · · Score: 3, Informative

      Actually, 4 DDR3 memory controllers, each of which can independently talk to a bank of DDR3 memory.

  15. Re:Multiprocessing by TeknoHog · · Score: 2, Insightful

    *whoosh* I wrote my first SMP code in 2001, and it was the typical thing to do in scientific computing, had been for decades. Thus I occasionally like to comment on the recent years' "multicore" marketing phenomenon, where even some developers seem to think they have a completely new problem and they need completely new tools.

    --
    Escher was the first MC and Giger invented the HR department.
  16. Re:Multiprocessing by Anonymous Coward · · Score: 2, Funny

    Mainframe, VAX, Supercomputer had Multiprocessing at the time?

    No

    Intel actually developed the first multi-core CPU and multi-processor systems at the behest of Steve Jobs as a condition for migrating OSX to the x86 platform. Further, it is speculated on good authority that Jobs personally headed up a crack engineering team sent to Intel expressly for the purpose of transitioning their fabs from the netburst to the core architecture. Seriously, study and learn.

    Posted anonymously from my iPad at the Starbucks in Cupertino. You know the one.