Dropsafe

by Alec Muffett

grids, and gigaflops-per-cubic-foot

I saw my friend Chris [www.csamuel.org] and his wife Donna [www.donnawilliams.net] weekend before last; stopping Chez-Muffett for breakfast off the flight from Chicago, I served them a breakfast of grilled lamb chump steaks, bacon, mushrooms, and well… an enormous mixed-grill basically. Chris looks fantastic, having dropped 2 Stone / 12Kg and dumped a whole pile of stress to boot, trading it for an idyllic-sounding life working in Melbourne for a Grid-computing research quango.

I was most enticed by one of the tales Chris had to tell, of what he’d heard at the IEEE CCGrid 2004 conference; apparently some bunch of researchers are measuring the per-instruction electrical current drain of various processors, and settling on the Transmeta chipset had managed to squeeze 19 of them into a 1U rack-mountable unit.

19 Processors in 1U. In a full-height rack, that’s (erm) rather a lot.

Damn, I’d like to play with a few of such things. Running a decent SMP operating system they’d make an excellent crypto-crunching platform.

⊞

Comments

6 responses to “grids, and gigaflops-per-cubic-foot”

  1. Chris
    re: grids, and gigaflops-per-cubic-foot

    I think it was actually 17 per 1U, which includes an integrated gigabit ethernet switch. I think they’ve had two of these units custom made for them so far. Basically they’re looking at technologies to get a terraflop in a rack and the main issue there is gigaflops per kilowatt and that was the metric that the Crusoe won on.

    This came from the keynote by Satoshi Matsuoka, Tokyo Institute of Technology, called “Towards Petascale Grid Infrastructures”, which isn’t (yet) available from the CCGrid website (none of the talks are up yet).

    Chris

    Reply
  2. alecm
    re: grids, and gigaflops-per-cubic-foot

    time to google! thanks mush!

    Reply
  3. Chris
    re: grids, and gigaflops-per-cubic-foot

    Oh, the boxes were running Linux by implication from what was said.. 🙂

    Reply
  4. Chris
    re: grids, and gigaflops-per-cubic-foot

    Looking at the notes I took the name of the system was “MegaProto v2” (although that may refer to a followup system, and seems to contradict some other things that refer to “MegaScale”).

    Here’s an abstract for a paper on Megaproto

    http http://www.ipsj.or.jp/members/SIGNotes/Eng/12/2003/096/article015.html

    and

    http http://www.arl.hpc.mil/Clusters2004/info_matsuoka.html

    There are also web pages that refer to this at:

    http http://www.para.tutics.tut.ac.jp/project/

    and one that only says “under construction” at:

    http http://www.para.tutics.tut.ac.jp/project/megascale/

    Aha both are correct – the following PDF document lists “Poster 13” as “MegaProto: A Prototype of Ultra Low-Power Mega-Scale System”

    http http://www.coolchips.org/cool7/pdfDocuments/VII-FinalProg-2004-04-021.pdf

    The coolchips.org domain belongs to the COOL Chips Symposium – “COOL Chips is an International Symposium initiated in 1998 to present advancement of low-power and high-speed chips.”.

    Chris

    Reply
  5. Chris
    re: grids, and gigaflops-per-cubic-foot

    Pah – those links didn’t come out as hyperlinks!

    Reply
  6. alecm
    re: grids, and gigaflops-per-cubic-foot

    sorry about that – it’s a CSS/Pr0n deterrent.

    Reply

Leave a Reply

Your email address will not be published. Required fields are marked *

More posts