Slashdot Mirror


Cell Broadband Engine Docs: VIP Access

I writes "The revolutionary Cell Broadband Engine Architecture (CBEA) is the result of collaboration among Sony, Toshiba, and IBM. The following papers define the Cell specification and will be posted to the IBM Semiconductor Solutions Technical Library in September. You can access them early as long as you have a current IBM ID."

17 comments

  1. dupe by Anonymous Coward · · Score: 3, Informative
  2. I like the pattern, but I want more. by shapr · · Score: 1

    I like the idea of many simple CPUs on a die, but I want to extend that.
    Why not go for a quilt architecture?
    For example, what about using tiled hexagons and squares. The hexagons could be local memory, the squares could be the CPUs, the edges would be links to the next tile.
    The price of a single die would be set according to the number of flaws on the die.
    Smaller CPUs are less likely to be dead because of a single flaw, so you'd get a higher yield.
    You can fit more complete hexagons into a circular die than complete squares, so you'd get more CPUs.
    I wonder why the CPU makers don't do this already?

    --

    Shae Erisson - ScannedInAvian.com
    1. Re:I like the pattern, but I want more. by rjw57 · · Score: 1

      Actually they already do. The cell in fact has one more processor on it than they claim and there is some logic to choose the n-1 best chips at any one point so that the entire thing is fault tolerant both for manufacturing and, if you happen to get a die with all of the processors working, for use.

      --
      Rich
    2. Re:I like the pattern, but I want more. by ciroknight · · Score: 1

      They do this because they don't have a choice; all of that silicon, errors are happening all over the place, and the best way to deal with the errors is to allow the 7 SPEs that work, to do so, and the lone SPE that probably didn't survive the manufacturing process is then cast as irrelevant.

      Personally, I've wondered if this is what has kept today's higher end chips out of my desktop. One has to realize how much it would affect pricing with error rates being higher than ever (smaller etching, and so many more transistors than before).

      I'm just waiting for my eight-die ARM machine, with maybe a Pentium 3 thrown in for some occasional light gaming.

      --
      "Victory means exit strategy, and it's important for the President to explain to us what the exit strategy is." G.W.Bush
    3. Re:I like the pattern, but I want more. by Soul-Burn666 · · Score: 1

      It certainly is. But probably more on the graphic card side.
      Modern cards have something like 16 rendering pipelines, usually grouped in quartets. Cards that have all 16 working sell as the "XT PRO SUPER GT". Those who only get 3 of the quartets working sell as "LESS XT LESS PRO LESS SUPER LESS GT".
      They have the full functionality, but have less pipelines.
      Price obviously changes accordingly.

      Same for things like LCD screens. One of the reasons the PSP costs more than a DS, is because making a good big LCD screen is much harder than making 2 much smaller screens. So the yields are higher and the prices drop.

      --
      ^_^
    4. Re:I like the pattern, but I want more. by sirrobot · · Score: 1
  3. IBM ID by wowbagger · · Score: 3, Insightful

    Getting and IBM ID is only slightly harder than getting a /. ID - fill in a form, click submit.

    I registered last week at work to get the documentation, though I have not yet had a chance to even take a quick look at it.

    But for what I do (communications and signal processing) the Cell looks really good - I think I could replace several DSPs and protocol processors with one Cell.

    1. Re:IBM ID by ciroknight · · Score: 1

      Hey thanks. I thought it'd be harder to get one of those IDs.

      I've been wondering when I could get my hands on one; I'd like to see how it could do some artificial intelligence research that I've been working on. Here's for hoping!

      --
      "Victory means exit strategy, and it's important for the President to explain to us what the exit strategy is." G.W.Bush
    2. Re:IBM ID by Decker-Mage · · Score: 1

      IBM does some nice things. Even a developer ID, with all those nice packages, DVD's, alpha and betas is dead easy to get. I've had one for years now. And the documentation available for free is simply awesome. "Need INPUT!" ;-) Not like Microsoft at all where you have to pay for the privilege of documentation that is useful (and I have paid and paid and...).

      --
      "[I]t is a wise man who admits the limits of his knowledge or skill, and that pretending either causes harm." --Terry Go
  4. Bug me not by Saiyine · · Score: 2, Informative


    Bug me not to the rescue! Who knows, maybe there's someone interested in RTFA...

    --
    Dreamhost superb hosting.
    Kunowalls!!! Random sexy wallpapers (NSFW!).

    --
    Hosting 20G hd, 1Tb bw! ssh $7.95
  5. Wealth of information but... by ciroknight · · Score: 2, Insightful

    I love IBM for all of the documentation they've provided at this point in time; one could actually start writing code to run on the SPUs right now, and probably only have to do a minimal amount of debugging once they get the hardware.

    But.. that being said, where's the hardware :(? I don't want to buy a PS3-dev kit (even if I could manage to get my hands on one); I just want to make my code fly on those seven SPUs. I wonder if there's an emulator currently available for the Cell processor, does anyone know?

    --
    "Victory means exit strategy, and it's important for the President to explain to us what the exit strategy is." G.W.Bush
    1. Re:Wealth of information but... by ja · · Score: 1

      ... where's the hardware?

      According to the forums at the IBM site, Toshiba is having a hw/sw dev kit in the pipe.

      mvh // Jens M Andreasen

      --

      send + more == money? ...
  6. Mutli Proc In General by ezweave · · Score: 2, Insightful

    It seems from TFA that this is really the place where the Cell proc would excell.

    But for more centralized processing, it seems like the world at large is not quite ready for this sort of multi-processing. I have only written assembly for single CPU multi threaded code, but locking problems are enough of a headache in that environ.

    A little O/T perhaps, but with the hope to use multicore procs as CPUs, does Vista/Longhorn seem even more... I don't know, maybe under-developed? From my limited experience, even with a CPU mutli-proc system coded properly you are really just speeding up locking problems (i.e. the same problems exist for synchronus bits of code, it just does it alot faster), like for I/O or HD access. Again, my experience with working at that low a level in a multi-proc environ are limited (basically non-existent).

    I guess my question is, wouldn't the SW world need to do some rethinking to develop consumer level apps that use multi-proc capabilities? That of course assumes that you could compile "multi-proc" code for the Cell or multi-core systems (perhaps this is me misunderstanding the cell proc).

    Then, aside from Linux or Darwin (I assume Darwin would work as multi-proc G5s are shipping now), what OS could hack it at a consumer level?

    (This may get modded as O/T, but I think it's related to the Cell as a signal proc discussion.)

    1. Re:Mutli Proc In General by qdhe · · Score: 1

      nice

  7. nice. by Anonymous Coward · · Score: 0

    this is nice.


    Al Qaeda Piracy

  8. Same stuff sans registration from Sony by sirrobot · · Score: 2, Informative
    You can access all the same pdfs (but not the IBM forum) from Sony without registering.

    http://cell.scei.co.jp/index_e.html

    Now that's VIP access.
    -SR

    1. Re:Same stuff sans registration from Sony by qdhe · · Score: 1

      ok,i should access the same pdfs from Sony without registering.