Slashdot Mirror
Blue Gene/L Tops Its Own Supercomputer Record
DIY News writes "Lawrence Livermore National Laboratory and IBM unveiled the Blue Gene/L supercomputer Thursday and announced it's broken its own record again for the world's fastest supercomputer. The 65,536-processor machine can sustain 280.6 teraflops. That's the top end of the range IBM forecast and more than twice the previous Blue Gene/L record of 136.8 teraflops, set when only half the machine was installed."
Imagine a Beowulf cluster of...oh, nevermind.
Click here or here.
lets put folding@home (http://folding.stanford.edu/) on that mother!
They say it can launch Adobe Acrobat Reader in ELEVEN SECONDS!!!
..figure out what the hell we are going to be doing for energy in 15 years??
"Look to the future and the present will be safe"
An IBM engineer was caught remarking "And boy can it hold a lot of porn."
The damn thing's smarter than I am. Well, that's taking an estimate of 100 teraflops for the human brain, which seems to be popular.
Real_men_don't_need_spacebars.
and the answer had better not be 42.
The legitimate thing that I can imagine is if it was a cost based contract that was given out before the cost of the hardware was known.
Was it?
Back when It was only half installed I got to take a tour of it while it was in Rochester, MN... Got to walk through it and touch it. Turns out the computer that controls blue gene takes up about half as much space as blue gene itself.
Comment removed based on user account deletion
WOULD YOU LIKE TO PLAY A GAME? aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
But it was paid for by the US government.
Read more of this story at Slashdot.Read more of this story at Slashdot.Read more of this story at Slashdot.
I have some very limited experience with this kind of computing, and I don't think the compiler is anywhere near the limiting factor.
I strongly suspect the limiting factor is algorithms. That is, the problem is designing code that can efficiently use a massively parallel machine. It's enormously difficult to even imagine how a problem could be solved by breaking it up into 65,000 mini-problems that can be solved simultaneously, and therefore mostly but not entirely independently. People just don't think that way. (Or rather, they do, but only at such a basic level close to the neurons that they are utterly unaware of how it's done.)
This is one reason "parallel computing" has been the Wave Of The Future(TM) for decades, and exhibits the same kind of "promise" as fusion power -- namely, we are told that ten years from now it will change everything -- and we hear it again every ten years.
Easy - you'd run a huge federal deficit, and let future generations sort it out.
Read more of this story at Slashdot.Read more of this story at Slashdot.Read more of this story at Slashdot.
.. since Quake 4 just hit the shelves.
What useful science has the "Earth Simulator" produced?
You might try reading The Journal of the Earth Simulator.
Or perhaps this summary of 2003 research
The 2005 projects are listed here
Here's a picture of the momma: http://en.wikipedia.org/wiki/Image:BlueGeneL-600x4 50.jpg
What useful science has the "Earth Simulator" produced?
Yes, better climate models and weather forecasts are obviously not needed. A little rain never hurt anyone, as this years hurricane season clearly shows.
Being bitter is drinking poison and hoping someone else will die
What useful utility has top sports athletes contributed to society? We probably pay the top athletes the cost of several super computers and all they do with all this money is throw orange plastic balls pumped with air into little string baskets hoisted on a pole. Sometimes I wonder what is the purpose of paying some man to hit a tiny ball with dimples into a hole far far away on a grassy playing field. And every time someone breaks someone else's record of hitting the tiny white ball, they get large sums of cash. But when these super computer people break records, all they get is a little pat on the back. BlueGene/L should be signing contracts with Nike and Pepsi.
On something like this, they would probably be programming in High Performance Fortran or Fortran w/ OpenMP -- or some similar dialect that supports massively parallel execution. I'm sure IBM develop an in-house compiler for the language.
Tubal-Cain smokes the white owl.
the article states 10 Mega watts (although I think that was the combined power usage for this and another supercomputer)
watch "the money masters" on google video
WOULD YOU LIKE TO PLAY A GAME? aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
If you're going to be an 80s geek, don't half-ass it like most people. The correct line from WarGames is "SHALL WE PLAY A GAME?"
Actually, the system is provided with the IBM XL family of compilers.
Just in time for... Windows Vista(tm)
Notice that the performance has actually increased PER proccessor as you add more proccessors... This is very remarkable in computer technology.
Normally when you add cpus to a computer you get a increase in performance, but it doesn't increase linearly with each cpu. You have one cpu you have 100% performance, add one more and you may have 180% the performance and add 2 more you may have 300% of the performance etc etc.
Notice that with half the machine there it got 138 GFlops.
So if you doubled the size of the machine you'd expect to get something like 260 Gflops per second.
But you have 280 Gflops per second.
This pretty much means that as you add cpus the performance of each cpu actually increases slightly. That's a exponentional growth rate, at the beginning of the curve.
Of course there has to be a technical limit to the system and the amount of space, heat, and electricity it can handle.. but technically if you double the size of the cluster again I wouldn't be suprised if you'd get close to 750 GFlops per second performance.
This is some seriously hardcore stuff, the future of computing hardware. Todays supercomputer, tomorrow's desktop.. I can't wait.
Time for Earth Simulator to make a Walmart run and get some more Athlons to regain the top of the "supercomputer" chart.
Don't forget to compile with:
make -j 65536
The faster they make these things, the slower they sing that damn Daisy song.
When this was announced, world chess champion Gary Kasparov said "ok, no way am I playing this fricken thing"
"Leo Fender was in a 'state of grace' when he designed the Stratocaster." -- Paul Reed Smith
What's so hard about releasing these things under an open source license.
Basic scientific method, really - control the environment as tightly as you can, and then document everything as thoroughly as you can. The first precludes open source while the experiment is ongoing, while the second requires opening up the source once the experiment's done.
Aren't universities supposed to encourage the spread of information?
Accurate information, yes. How would you propose that accuracy could be guaranteed with an open client that anyone could alter?
Oh, and don't bother starting in on how binary-only nodes could be hacked, wires can be tapped, etc. - I know that. It's irrelevant. The goal of an experiment like this is to eliminate any variables other than the ones you're testing. Not every variable can be eliminated, but that's not a legitimate reason to abandon the effort entirely. Besides which, only someone with malicious intent would bother going to that kind of trouble; an open-source client could be comprimised by a well-meaning hacker who tried to "optimize" his copy of the client by taking short-cuts.
Lost: Sig, white with black letters. No collar. Reward if found!
65536 processors = 64K processors.
damn that IBM, they take geekiness to just a whole different place.
Funtage Factor: Purple
It still takes 15 seconds to start up OpennOffice.org
"The Chinese use two brush strokes to write the word 'crisis.' One stands for danger; the other for opportunity
What do you think this machine is doing? You use this machine to calculate how far you can push your reactors to provide the energy you're talking about.
"There is nothing nice about Steve Jobs and nothing evil about Bill Gates." - Chuck Peddle
The main drawback to forecasting models is that it takes soo long to run all the data, so we have to cut back on the data so that we can actually see what's forecast before it happens. With this this thing running an expanded version of the GFS with 10KM resolution, we might be able to actually get it right for once. ;)
Someone save me from this sanity.
That probably only means that they have optimised the architecture over time as would be expected. Things like improved resource management, a slimmer kernel for each CPU, a better compiler, etc. can easily make up for that small performance gain.
Actually, most machines are partitioned into front end and back end. The back end is for running large production runs (1000's of PEs) and is usually on accessible as a batch queue. The front end is for compiling and debugging and is interactive (perhaps even running serially). The front end might even be another machine.
Contrary to popular /. opinion, compiling is not a big task. Especially when compared to the real calculations done.
Big machines like this usually have another queue on the front end for long compilation jobs.
So make -j 4 might be more appropriate. Unless you wanted to piss other users off.
No, this is IBM. Wang went out of business years ago.
Censorship is telling a man he can't have a steak just because a baby can't chew it. --Mark Twain
Linux is substantially more scaleable now than it was even just 6 months ago (not the vanilla, but quite well tested scaleability patches). This could account for the improvement. I suspect if they ran just half of it now, they'd get a little bit over half the performance (but not much over half - that is how good Linux is these days).
Sorry, but I have to disagree with your conclusion that this represents exponential growth.
The effect you speak of (doubling the number of processors giving less than double the final "power") is due to additional overhead - various processors coordinating their work with each other, deciding things like "Should I split this 2 ways or 4?" and so on - and that sort of stuff inevitably increases with the number of processors.
You can use improved algorithms, special-purpose hardware, etc, etc, to minimize this "friction", but it will always exist, and the percentage of processing that is "overhead" will inevitably climb as you increase the number of processors.
It's far more likely that either the earlier number resulted from some inefficiencies that existed then (due to it not being built as designed yet, perhaps), or there have been improvements in the algorithms or infrastructure which give greater efficiencies.
If it's the latter case, if you unplugged the 2nd half of the CPUs and made the measurement again, you'd probably get 150 GFlops or so.
Basically, you could write the equation for total power something like:
X - O - i**x, where X is the number of processors, O is the basic overhead (for doing things like I/O, for example), and c is the incremental cost of adding each processor.
To have what you describe would require that i**x be a negative number, which is like saying that you can have 10 individual conversations in less time than you can have five. Ain't gonna happen.