The Seven Most Powerful Supercomputers.

1. Christophe
   June 25th, 2007 at 10:25 am

   A teraflop measures a computer’s ability to do Floating Point Operations Per Second.
   So 360 Teraflops = 360,000 gigaflops = 360,000,000 megaflops = 360,000,000,000 flops.
   360,000,000,000 Floating Point Operations Per Second. That’s 360 Billion.

2. Miss Cellania
   June 25th, 2007 at 10:32 am

   So a megaflop = 1,000 flops. Kinda like Ishtar.

3. kevin
   June 25th, 2007 at 10:48 am

   A megaflop is a 1,000,000 Floating Point Operations Per Second - not a 1000
   http://whatis.techtarget.com/definition/0,,sid9_gci212543,00.html
   so it’s 360 TRILLION! - now, that’s a lot!

4. Adam
   June 25th, 2007 at 10:56 am

   Actually, a megaflop is 1,000,000 flops, so 360 teraflops is 360,000,000,000,000 flops. (360 trillion)

5. Anita
   June 25th, 2007 at 11:11 am

   Depending on how you look at it, the largest supercomputer could be the Folding@Home project, which has peaked at over 1,000 teraflops.

   In comparison, if every single neuron in the human brain fires at its max rate (~200 times per second), you’d have an internal 100 teraflop super computer. Unfortunately you’d probably also die from what is essentially a massive full-brain seizure … your max sustained teraflops are probably closer to 5-10 teraflops (still VERY fast).

6. Simon
   June 25th, 2007 at 11:15 am

   Kevin and Adam are absolutely correct.

   there is a missing “kilo” in the conversion of the units:

   1 Megaflop = 1,000 kiloflops = 1,000,000 flops

   kilo means “times 1 thousand”
   Mega means “times 1 million”

7. Sid Morrison
   June 25th, 2007 at 12:51 pm

   Update 3: Christophe and Kevin/Adam are all right!

   The term “billion” means different things depending on which English you speak. It comes down to something called the “short and long scale numerical systems”, which vary their terminolgy in multiplying units of 1,000 (short system) or in 1,000,000 (long system). You can read about it here (http://en.wikipedia.org/wiki/Long_and_short_scales), but here’s the gist:

   10^9: Americans (and Canadians) call this a “billion”, but Britons (at least old ones) call it a thousand million. Sometimes the term “milliard” is used, but that is mostly in non-English speaking countries.

   10^12: Americans can this a “trillion”, whilst Brits call it a “billion”.

   10^15: Americans call it a quadrillion; Brits call it a thousand billion or sometimes a “billiard”, but that is pretty archaic.

   10^18: Americans call it quintillion;
   Brits call this a trillion.

   Reading the Wiki article, it seems like the Brits themselves have been changing since 1974 to the “short scale” system used by the US, Canada, Australia, and other English speaking countries. Most of continetal Europe continues to use the long scale (like Christophe).

   Straight talk from Sid.

8. Samui
   June 25th, 2007 at 3:36 pm

   I don’t know whether to be disturbed that 4/7 of them or used to design nukes or relived that the other 3 aren’t

9. Zeke
   June 26th, 2007 at 9:15 am

   Soon we’ll be measuring in petaflops (1,000,000,000,000,000 flops)!

   To put this into a greater perspective, a Core 2 Duo processor puts out a *theoretical maximum* 4 flops per clock (per core, or 8 flops per clock for both) and runs somewhere between 1.8GHz and 2.66GHz.

   That means that an average (2GHz) processor operating at theoretical peak efficiency (and for various reasons this never actually happens in real life, and the Top500 scores are actually measured with a tool called Linpack that calculates closer to real performance), your average Core 2 Duo processor would put out about 16 gigaflops. Real performance probably peaks at about 70-80% of that, which means about 12-13 gigaflops.

   So…
   1,900,000,000 flops — Cray-2 supercomputer (1985)
   12,000,000,000 flops — average Core 2 duo performance
   94,210,000,000,000 flops — MareNostrum (#5)
   360,000,000,000,000 flops — Blue Gene/L (#1)
   1,000,000,000,000,000 flops — one petaflop.

   As you can see the top cluster systems are thousands of times more powerful than your average desktop system, and your desktop system has the advantage of not having to actually talk to the other systems to achieve that performance! These clusters take a performance hit from the individual nodes having to talk to each other.

   Anyway, supercomputing has come a long way but still has plenty of room to grow.

10. raytard
    June 26th, 2007 at 7:31 pm

    blue gene/p is a petaflop (it was unveiled today)
    aparantly it can be scaled to 3 petaflops

    so yes, Zeke, we are measuring in petaflops now

    it is the new number one, officially it is 3 times faster than its predecessor

    just because you may not believe me, heres a link to it
    http://www.gizmag.com/go/7511/

    “The Blue Gene/P system can be scaled to an 884,736-processor, 216-rack cluster to achieve three-petaflop performance”

    few more weeks and they’ll cure cancer

11. George
    August 11th, 2007 at 3:24 pm

    I’ve been having trouble understanding the whole concept of “flops” (no collage education…) so I decided to stick with “GHz”. I was doing some conversion flops to Ghz and I can’t believe the results. By my calculations Blue Gene/L (360,000,000,000,000 flops) works roughly at 60,000,000 GHz (??!!). I’d appreciate some oppinion on that. Please. I can’t sleep since then.
    Pax et bonum.

PLEASE LEAVE A COMMENT