backtop


Print 29 comment(s) - last by AlexTRoopeR.. on Feb 7 at 4:17 AM

IBM is currently developing a supercomputer it hopes will be able to deliver 20 petaflops per second

IBM announced ambitious plans to create a new supercomputer that will be 20 times faster than its current Roadrunner supercomputer.  The new supercomputer, dubbed "Sequoia," will operate at a whopping 20 petaflops, and is significantly faster than IBM's previous supercomputers.

The new supercomputer will be used at the Lawrence Livermore National Laboratory in Livermore, California, and will allow researchers to use the powerful computer for simulations of U.S. nuclear weapons.  Lawrence Livermore is using the IBM BlueGene/L system until Sequoia is ready.

The system will be stored and used in a 3,422 sq. ft. building in Livermore -- it will be energy efficient, with IBM expecting it to use 6 megawatts per year, which is equivalent to 500 American homes. 

Sequoia may be able to provide a 40- to 50-fold improvement in the country's ability to provide data, including severe storm forecasting, earthquake predictions and evacuation routes due to national emergency, IBM said in a statement.

The system will use 45nm processors that have up to 16 cores per chip, and will have 1.6 petabytes of memory shared by 1.6 million cores.  It will be 15 times faster than BlueGene/P and have the same footprint with only a "modest" increase in power consumption.

IBM's latest announcement comes just seven months after IBM delivered the fastest supercomputer, Roadrunner, to the U.S. Department of Energy's Los Alamos National Laboratory.  The supercomputer was the first system to break the 1 petaflop barrier, clocking in at 1.026 petaflops.

IBM also is working on other supercomputers that will be used by the Defense Advanced Research Projects Agency (DARPA), and should be available before 2011.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Power usage
By mcnabney on 2/3/2009 4:08:55 PM , Rating: 2
An average home does not use 12KW. That is about $900 per month electric bill as an average household. I think their decimal point floated over a bit. Six megawatts is enough for thousands of homes.




RE: Power usage
By TennesseeTony on 2/3/2009 4:34:38 PM , Rating: 2
No, the article is correct. The figures given were for an entire year, which, depending on how much you pay per kilowatt of juice, is only about $70-$90 per month, for each house, for a year.


RE: Power usage
By TennesseeTony on 2/3/2009 4:36:25 PM , Rating: 2
Must be really small houses, by the way. Or maybe they have wood burners, or rely on natural gas for heating, or...


RE: Power usage
By SiliconJon on 2/3/2009 4:37:17 PM , Rating: 2
My house averages about 1.3MW's per month, so that's about right considering my house is all electric.


RE: Power usage
By SiliconJon on 2/3/2009 4:45:08 PM , Rating: 3
Me thinks we have a lot of errors from the posts and the blog when added together...egads.


RE: Power usage
By masher2 (blog) on 2/3/2009 4:39:34 PM , Rating: 5
> "The figures given were for an entire year"

The statement "Megawatts per year" is nonsensical. You pay for a unit of energy, which is power times a unit of time.

The OP was correct. Homes do not average 12 kW of power consumption. The figure is off by a factor of 10.


RE: Power usage
By jbartabas on 2/3/2009 4:54:45 PM , Rating: 2
The sad thing is that I've have already seen these exact 2 very obvious mistakes on a few web sites today ...


RE: Power usage
By jimpaka on 2/4/2009 3:38:39 PM , Rating: 3
Finally - thank you. It's shocking that so few people understand the difference between a Watt and a Watt hour.

"1 Watt" is a RATE of energy consumption (1 Joule per second), not an AMOUNT of energy. An AMOUNT of energy can be stated in Joules (Watt seconds) or as is more common, Watt hours (Wh), KiloWatt hours (KWh), MegaWatt hours (MWh) and so on.

As masher pointed out, MegaWatts per year does not make sense in this context. MegaWatt hours per year does make sense when describing how much energy a particular system is using. You can describe the same thing by simply stating that the system uses X MegaWatts.

Also, shame on the author of the article for getting this wrong.


RE: Power usage
By diego10arg on 2/3/2009 5:08:13 PM , Rating: 2
They just feel the need to make it sound greener.

Oh, wait..!


RE: Power usage
By winterspan on 2/4/2009 1:46:59 AM , Rating: 5
I can't believe how many of you guys (in addition to the writer) aren't seeing the glaringly obvious ERROR!

Energy is NOT measured in Watts! The Watt is a unit of power! It's Watt-hours! Who knows what the actual correct figure is..


RE: Power usage
By teldar on 2/4/2009 5:46:23 AM , Rating: 2
What's a lightbulb use?


Department of Redundancy Department
By Mclendo06 on 2/3/2009 3:42:01 PM , Rating: 5
quote:
IBM is currently developing a supercomputer it hopes will be able to deliver 20 petaflops per second

Sorry, just a pet peeve of mine, but FLOPS stands for "Floating Point Operations Per Second". Unless this computer computationally accelerates, you need to change this line in the article.

/soapbox




By jnn4v on 2/3/2009 4:22:49 PM , Rating: 4
And is it really 20 petaflops? If so, at least one of the following statements from the article describing its relative performance doesn't make sense:

quote:
IBM announced ambitious plans to create a new supercomputer that will be 20 times faster than its current Roadrunner supercomputer.

quote:
It will be 15 times faster than BlueGene/P

quote:
IBM's latest announcement comes just seven months after IBM delivered the fastest supercomputer, Roadrunner, to the U.S. Department of Energy's Los Alamos National Laboratory. The supercomputer was the first system to break the 1 petaflop barrier, clocking in at 1.026 petaflops.


So following that logic, BlueGene/P would be faster than RoadRunner. And as of the November 2008 list, BlueGene/P was #5 on the Top500 whereas RoadRunner was #1.


RE: Department of Redundancy Department
By LRonaldHubbs on 2/3/2009 4:24:55 PM , Rating: 5
It depends upon the capitalization. If someone says FLOPS, all caps, then it is assumed to mean FLoating point Operations Per Second. However, FLOP without the capital 'S' is singular - FLoating point OPeration - and putting a lower-case 's' on the end makes it plural, FLOPs = FLoating point OPerations. This is commonly substituted with all lower case, flop(s) to avoid confusion with the rate acronym. The usage of this lower-case form is acknowledged in various dictionaries and on Wikipedia.

In the case of this article, the lower-case noun form was used, followed by a verbal rate quantifier. It's more work to write than the traditional acronym form, but it is stil correct.


RE: Department of Redundancy Department
By jbartabas on 2/3/2009 5:10:53 PM , Rating: 2
quote:
However, FLOP without the capital 'S' is singular - FLoating point OPeration - and putting a lower-case 's' on the end makes it plural, FLOPs = FLoating point OPerations. This is commonly substituted with all lower case, flop(s) to avoid confusion with the rate acronym.


So what does this mean:

quote:
The new supercomputer, dubbed "Sequoia," will operate at a whopping 20 petaflops , and is significantly faster than IBM's previous supercomputers.


with all lower case?

quote:
The usage of this lower-case form is acknowledged in various dictionaries and on Wikipedia.


Can we have references? 'flop' without a 's' does not bring anything in Wikipedia, section Acronyms.


By jbartabas on 2/3/2009 5:34:41 PM , Rating: 2
Ok, I've found it (that wasn't that hard after all :_( ). There does not seem to be a distinction between usage with upper/lower case. It sounds like a bad idea to use both in the same article, and/or without proper definition.


By marvdmartian on 2/4/2009 9:58:30 AM , Rating: 2
What pickup line do IBM computer geeks use when trying to pick up girls in the bar?

"Hey baby.....want to come over to my place and see my petaflop???"


2012??
By icanhascpu on 2/3/2009 3:39:50 PM , Rating: 2
Hi, Dave.




RE: 2012??
By dj LiTh on 2/3/2009 5:03:47 PM , Rating: 2
Dave's too busy playing Crysis.


RE: 2012??
By Dianoda on 2/3/2009 6:27:07 PM , Rating: 3
I'm mildly surprised and slightly impressed with the maturity of DT readers. This same news was reported on Gizmodo, and I could swear about 2 out of 3 posts were some variation of "but will it blend/run Crysis?" Damn near made me vomit (well, that might've been my flu, but I digress). It was truly as if I had stumbled into the eighth circle of hell....

Anywho, from what I understand, it was decided commission the creation of what will be the world's most powerful supercomputer, for the purpose of simulating nuclear weapons? Truly, surely, there could be no cause nobler than thou, oh simulation of nuclear weapons.

Because there's nothing better than using a computer to further our understanding of a class of weapons which the world has determined taboo. This sounds like continued development of nuclear weapons, I thought we were trying to prevent this. Say it ain't so.


RE: 2012??
By icanhascpu on 2/3/2009 8:33:51 PM , Rating: 2
Depends on who buys time for computations. I would hope it will be used for more social science based rather than military science. Figuring out cures for whatever ect.


RE: 2012??
By grath on 2/4/2009 1:21:14 AM , Rating: 5
As long as potential bad guys have them, we need to have them. As the stockpile ages, warheads need to be decomissioned, so we need to build new ones. We need to know if the new ones will work. We have signed the Comprehensive Test Ban Treaty, not ratified it yet, but we adhere to it and have not tested since 1992. Given the option between resuming underground nuclear testing and building a supercomputer to simulate it, would you not build the supercomputer? At least it can do other things too.


RE: 2012??
By Dianoda on 2/4/2009 12:56:49 PM , Rating: 1
The US is so sneaky. 'Sure, we won't actually test the nuclear weapons, but that doesn't mean we're not going to test the weapons.' Building a computer to test it for you kinda violates the spirit of the treaty, but whatevers.

Now I just want to know, how the heck does it work? Are they simulating atom to atom interactions? On a grand scale? I guess they just build as complex and realistic a model for an atom as possible and replicate it a few billion times into the shape of the blueprint, and tell the computer to make it happen. And while the computer's working on the output, they probably just dick around.

'Suppose the bomb was under assault by a legion of fuzzy bunnies...We should really know exactly what would happen, just in case.' Ah, taxpayer dollars at their finest.

And I'm all for decommissioning old warheads, but as far as making new ones, how many do we really need, tactical or otherwise? It's not like were in a cold war with anyone. I say, just keep a few good eggs and say you got lots, I mean, come on, who would ever be able to call your bluff? Makes security a lot easier. And would be oh so much cheaper, spend that money you saved on some of those sweet looking rail-guns or a really big orbital laser. Or an orbital laser rail-gun. Impossible, completely illogical? Maybe. Awesome? definitely. Wave of the future!


By the way
By Pirks on 2/3/2009 5:47:56 PM , Rating: 2
Windows HPC 2008 is now #10 in Top500.

Yes, number TEN.

I'm not kidding: http://www.top500.org/lists/2008/11

Holy f#ck...




RE: By the way
By surt on 2/3/2009 6:59:56 PM , Rating: 2
What's the surprise? You throw enough hardware at a problem and even windows can't get in the way enough to stop it.


RE: By the way
By Pirks on 2/3/2009 7:10:26 PM , Rating: 2
Supercomputers were the domain of Unix since the dawn of time, it's really shocking to see MS Windows encroaching in there, and especially _this_ fast, in less than a year!

It's a desecration! Blasphemy!!! :)) *cue idiotic BSOD jokes here*


throwin' code
By codeThug on 2/3/2009 6:15:07 PM , Rating: 2
I find it strange that software is never mentioned in this type of article. Yeah, yeah, I know it's more fun to brag about silicon feeds and speeds than auto-parallelizing compilers.

But, what is 20 petaflops if the inner code loops are sloppy and waste most of the crunch power.




RE: throwin' code
By Fritzr on 2/3/2009 9:18:43 PM , Rating: 3
More computing power is meaningful ... the faster the computer the less optimizing needs to be done by the program designer ... just look at desktop computers.

The bare operating system today cannot run at an acceptable speed (in many cases not at all) on early Intel & Motorola cpus.

There is so much bloat using processing power that it is ridiculous. A 64 bit version of one of the multitasking MS-DOS clones with GEM or GEOS for a GUI, would be a speed demon on a modern entry level computer barely able to run windows. More useful would be an early Linux recoded to use 64 bit addressing, but maintaining the tight fast coding of the designed for i286 Linux. Even in the Linux world bloat is stealing cpu cycles.

It's much easier to use the compiler libraries instead of custom high speed code. When you are doing this for small simple functions you are trading ease of coding for speed. For complex functions where the compiler is hiding a lot of machine specific code variations then you need to look at the library source to make sure that the customized binary isn't loading all the unused code that your version will never use. This is the weakness of generic .dll files.

The science mainframes have always used computing power to overcome slow executables. FORTRAN was developed to make it easy to write a science program, but it was not designed to generate the fastest executables. It is still used today due to the number of legacy applications that are still useful, as well as allowing researchers who have learned FORTRAN to continue writing code without taking time off to learn a new language.

A really good programmer can take a few extra weeks or months to go through the flowcharts and final code and find places where a rewrite will accelerate the program. Researchers will instead allow the program to run a bit slower while they get other work done. Instead of spending the time they would save, saving the time, they buy time on a faster machine and speed up their code the lazy way :P


FLOPS
By AlexTRoopeR on 2/7/2009 4:17:29 AM , Rating: 2
FLOPS stands for FL oating point O perations P er S econd. So the subtitle "IBM is currently developing a supercomputer it hopes will be able to deliver 20 petaflops per second" says per second twice...




"You can bet that Sony built a long-term business plan about being successful in Japan and that business plan is crumbling." -- Peter Moore, 24 hours before his Microsoft resignation











botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki