Holy jeeeeesus!!!!! Damn, it's good to see IBM has some competition on top500. I remember the Earth Simulator reigning king for a while, I didn't think it would get beat as soon as it did by the Blue Gene. I wonder if their projected 10 petaflop speed is still going to be impressive (at least in a top500.org sense) in 6 years? SHould be an interesting battle to watch.
its scary how powerful these computers are. In a 100 years or so will this be the power the desktop PC will be?? Imagine how real the games will be on a machine like that
an average spec desktop pc today would actually vastly outperform it in gaming these arent built for games obviously TSO on z/OS (uber ibm mainframes, the bit you actually interact with) is considerably less responsive than using windows, as a comparison
I wonder if their projected 10 petaflop speed is still going to be impressive (at least in a top500.org sense) in 6 years? I don't really see how its big news. I mean, yeah a 10 petaflop machine is kick ass and all, but 6 years ago wasn't 136 teraflops just as impressive? Its to be expected that it'd grow that much I think. I mean, the Blue Gene is only using 700 MHz power PC chips. I'm sure that what ever chips the new Japan machine uses wont be 'state of the art', but it'll get the job done just like the Blue Gene.
Strangely, the article linked to from top500.org says that the Japanese NEC/Hatachi machine planned for 2010/11 will aim to be around 3/4 petaflops not 10. Here's the article: http://cio-asia.com/ShowPage.aspx?pagetype=2&articleid=1400&pubid=5&issueid=48 I've just made some predictions by taking the last 20 top ten machines from the past 10 years on top500.org and projected their speeds using an exponentional function (with Excel doing the maths). It predicts that by June 2010 we'll have just reached the 3 petaflop mark which is more in line with what they say they are going for in that article. However by 2015 I predict we'll get to around 80 petaflops and by 2100? Well how does 250000 yottaflops sound?
It seems they are just amassing huge amounts of chips and calling them supercomputers. Why don't they build one superchip. Then I'd be impressed.
Isn't Blue Gene top without being finished yet? I seem to remember seeing that it pwned the World Simulator with only being half-finished. What's the end power supposed to be? As for the new masheen,
Moore's law doesn't dictate super computer power, only individual processor power. It says the that number of transistors on a chip doubles around every 18 months. This trend is exponential which means the power of our processors only increases faster and faster. The main limitation on a super computer is actually memory bandwidth but looking at the last few top supercomputer speeds the trend in their growing power also seems to be exponential. That's how I got such massive figures when extrapolating to 2100. Of course people say that Moore's law will break down due to physical limitations but they've been saying that for years and it doesn't appear to be slowing down any time soon. We'll just have to wait and see I guess.
In the end we'll just be moving over to parallel processing anyway. A la Supercomputer architecture at the moment.
yeh but kirkkit there is a still the problem that you can't get more than n junctions in x*y space. The problem is that plain, regardless of how many chips you have, you still need more silicon. (more silicon, tends to mean more money needed).
What I ment was there will be a cap on individual processor power as the maximum transistor density is reached for silicon. After which, multiple prcessors will be the only option with the computers physical size growing exponentially. I'm waiting for the first bio transistor
The sad thing is if they are using this to do weather forcasting with each succesivly more powerfull computer there will be less and less of an improvement in acuracy over the previous generation.
