backtop

Hardware ORNL to Get New Supercomputer
Tim Thorpe (Blog) - June 20, 2006 2:51 AM
Print E-mail del.icio.us 31 comment(s) - last by rushfan2006.. on Jun 28 at 2:44 PM
Oak Ridge National Labs has signed a $200 million USD contract with Cray to upgrade its computing capacity

Supercomputer maker Cray Inc. has signed a multi-year $200 million USD contract to upgrade Oak Ridge National Laboratories (ORNL) current Cray XT3 Jaguar supercomputer. The first phase of the project will be complete by the end of this year and will upgrade Jaguar to dual-core AMD Opteron processors increasing current performance from its current 25 teraflops (25 million million calculations per second) to 50 teraflops. Phase two will bring Jaguar to 100 teraflops by the end of 2006. Phase three will further increase Jaguar's performance to a final 250 teraflops by the end of 2007.

In 2008 ORNL is expected to install a next generation Cray supercomputer code named Baker capable of running up to one petaflops, 4 times faster than Jaguar after its upgrade.  Baker will utilize a future version of the AMD Opteron processor (likely K8L based). Researchers at ORNL use Cray supercomputers to explore the frontiers of neutron science, biological systems, energy production and advanced materials.

Last month DailyTech reported that Los Alamos National Labs (LANL) had begun taking bids to increase its computing capacity.  The new LANL Supercomputer, dubbed ‘Roadrunner’, will be able to run at a sustained two petaflops by the time it is completed, twice that of ORNLs future Cray cluster.

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
I want One!
By a1trips on 6/20/2006 9:00:39 AM , Rating: 2
on the serious side.. several commonplace workstations these days are packing Supercomputing Horsepower.. judging by stabdards of yesterday.. so maybe one day we all will have 2 petaflops insanity in each house!

Lookin forward to it:)




RE: I want One!
By Master Kenobi (blog) on 6/20/2006 9:17:32 AM , Rating: 1
Well if they were smart they would simply use a distributed computing system across all the computers on campus. I bet there are plenty of underused or unused computers with some decent processing power. On a side note though, if they have the money, why not. Never hurts to think ahead.


RE: I want One!
By LazLong on 6/20/2006 2:07:58 PM , Rating: 2
Obviously you have neither a computer science nor any other computational science background. It's true that there are calculations that could work using your suggestion, but most of what ORNL, LANL, LLNL, etc. do require faster interconnects than gig-e.

"If [you] were smart [you] would simply" engage your brain before you engage whatever it is you use to type with.


RE: I want One!
By TomZ on 6/20/2006 3:05:23 PM , Rating: 2
Your point would be more concisely made if you didn't throw in the unnecessary insult at the end.


RE: I want One!
By bob661 on 6/20/2006 4:04:01 PM , Rating: 2
quote:
Your point would be more concisely made if you didn't throw in the unnecessary insult at the end.
Yeah. I'm the only one that can insult people around here, damn it. :)


RE: I want One!
By Shadowself on 6/20/2006 9:25:44 AM , Rating: 2
The original Cray-1 back in the 70s was a 12 megaFLOP machine. I don't believe anyone (Apple, Dell, Gateway, HP, or Lenovo, etc.) is shipping even laptops which are that slow today. The definition of "super computer" keeps moving as much as the capabilities keep moving.

Will we have 2 petaFLOPS on the desktop? Yes. Will it happen within 20-30 years? No one knows.


RE: I want One!
By masher2 (blog) on 6/20/2006 9:50:01 AM , Rating: 3
Assuming Moore's Law remains roughly constant, we'll have 25 teraflop desktops in 30 years. A 2-petaflop desktop would be more like 45 years out...too far out to make any accurate sort of prediction at all.


RE: I want One!
By Hypernova on 6/20/2006 10:10:56 AM , Rating: 2
Floating poing capabilities aren't good measure of power anymore. I prefer MIPS since not every instruction need FP calculations.


RE: I want One!
By masher2 (blog) on 6/20/2006 1:03:57 PM , Rating: 2
Considering any problem hard enough to require teraflop-level performance (much less petaflop) is going to involve floating point numbers-- its still the best measure of supercomputer performance there is. That's why they use it.


RE: I want One!
By TomZ on 6/20/2006 3:08:55 PM , Rating: 2
quote:
Considering any problem hard enough to require teraflop-level performance (much less petaflop) is going to involve floating point numbers-- its still the best measure of supercomputer performance there is. That's why they use it.

I'm curious why this post has been rated down, even though nobody has posted any refutation of masher2's statement.


RE: I want One!
By Obadiah on 6/21/2006 6:21:21 PM , Rating: 2
quote:
I'm curious why this post has been rated down, even though nobody has posted any refutation of masher2's statement.


Well, it was kind of inane and circular - no better than any of these:

quote:
Considering any problem hard enough to require teraiop-level performance (much less petaiop) is going to involve i/o rates--


quote:
Considering any problem hard enough to require teraiop-level performance (much less petaiops) is going to involve integer numbers--


quote:
Considering any news article about teraflop-level performance (much less petaflop) is going to involve floating point numbers--



RE: I want One!
By rushfan2006 on 6/28/2006 2:44:16 PM , Rating: 2
quote:
I'm curious why this post has been rated down, even though nobody has posted any refutation of masher2's statement.


I never understood that whole bit about points for the posts and what "modded" means. I never bothered to "research it" or really ask -- until now....lol. What is that stuff all about, and why we are at it..how come by default when I login I won't see all the posts unless I change the threshold?

Oh well..beyond that like the one guy said a topic like Supercomputers what is much to post about except to say "damn -- wish I had one to play games on?"..lol


Numbers?
By 1n73rl0p3r on 6/20/2006 10:19:40 AM , Rating: 2
Ummm, last time I checked, GIGA meant "billion", tera meant "trillion" and peta was 1*10^15, whatever that is. Dunno where you went to school. :)




RE: Numbers?
By Hare on 6/20/2006 10:48:42 AM , Rating: 1
American Billion is European Milliard
American Trillion is European Billion

It's not that simple. Maybe you should follow the rest of the world and open arms to the ISO-standards...


RE: Numbers?
By 1n73rl0p3r on 6/20/2006 10:56:01 AM , Rating: 2
Million is always million, even in Europe. So STFU. Kthx.


RE: Numbers?
By Howard on 6/20/2006 12:11:25 PM , Rating: 2
SI is:

10^6 = million
10^9 = billion

According to http://en.wikipedia.org/wiki/SI_prefix


RE: Numbers?
By 1n73rl0p3r on 6/20/2006 12:33:25 PM , Rating: 2
Yeah, so "petaFLOPS" is definitely not "millions of Flops". Somebody fix the dang article.


RE: Numbers?
By gerf on 6/20/2006 10:56:33 AM , Rating: 2
Maybe Europe is on the non-intuitive side for once. Ok, not maybe.

/American who uses metric enough at work/school/travel that it's fine by him


RE: Numbers?
By LazLong on 6/20/2006 2:15:44 PM , Rating: 1
You are an idiot. We are using modern English here. True, at one time 'milliard' was used in addition to 'billion.' However, never has 'trillion' meant 10^9.

Quit your childish America-bashing, at least until you have a better grasp of English, Queen's or otherwise.


RE: Numbers?
By TomZ on 6/20/2006 2:42:27 PM , Rating: 2
Are you naturally offensive, or do you have to work at it?

Seriously, is it necessary to insult someone each time you make a post? Do you think it helps make your point?


RE: Numbers?
By eomhS on 6/22/2006 1:03:17 AM , Rating: 2
and to Geaorge W. its a Brazillion


RE: Numbers?
By eomhS on 6/22/2006 1:04:35 AM , Rating: 2
and to George W. Bush, its a Brazillion


ORNL to Get New Supercomputer
By lifeguard1999 on 6/21/2006 7:45:52 AM , Rating: 2
The Cray XT3, while it looks great on paper (5,200 Opterons anyone?), does have some serious shortcomings. Yeah, sure, it can "do" 25 Tflops. But that is just the number that Cray (or any vendor) guarantees you will never meet. To find out what it can really do, go to top500.org where the ORNL Cray is listed at number 10 in the world based on LINPACK performance. While not the best measure of the speed of a supercomputer, it is the industry standard. What we (and others) actually do in the industry is run benchmarks of our codes on smaller test systems to determine which one is faster.

The Cray XT3 consists of two primary components: compute nodes and service nodes. A microkernel runs on the compute nodes while SUSE Linux runs on the service nodes. The XT3 Catamount microkernel interacts with an application in a limited way by managing memory and performing scheduling. This microkernel architecture ensures high-performance, low-latency MPI communication.

Finally, think about all the unused components on your motherboard. Do you have a second gig-e port that is unused? That is 5 Watts of power. Multiply that by 5,200 nodes and it starts to chew up some serious power. These supercomputers are stripped down of all unnecessary components to save on that power. (Now think of how many unused components are on all computers; multiply that by how many computers there are in the world; that is wasted power.)




RE: ORNL to Get New Supercomputer
By Clauzii on 6/21/2006 7:19:23 PM , Rating: 2
Imagine all that exessive energy - must be golden ages for the energy producing industries.

And nobody cares :(


RE: ORNL to Get New Supercomputer
By lifeguard1999 on 6/21/2006 8:23:27 PM , Rating: 2
quote:
Imagine all that exessive energy - must be golden ages for the energy producing industries.

And nobody cares :(


Trust me. When you have to have the power company build a new power substation just to power your HPC machines, you care. The HPC companies (Cray, IBM, SGI, etc.) are all racing to be the fastest while using the least amount of power and the least amount of cooling.


RE: ORNL to Get New Supercomputer
By Clauzii on 6/22/2006 12:33:28 AM , Rating: 2
Yes, big companies does so. But do we? I mean, 99% of the people I know that use computers, don't put a single thought to power consumption, or the pollution created thereby. THAT's what I think is sad... :)

It is us (the users) that don't cooperate to save energy. As individuals, we also think in money, but the 'other way around': "Nah, I only save a few bucks, don't matter."

BUT a few watts times millions of users?? And if we include all those products on standby (more 'few watts').....

WE should do that :)


Correct it please
By murray13 on 6/20/2006 2:59:40 PM , Rating: 2
quote:
...its current 25 teraflops (25 million calculations per second)...


Please correct the article to read (25 million million calculations per second) or
(25 trillion calculations per second).




By KristopherKubicki (blog) on 6/20/2006 3:36:09 PM , Rating: 2
Thanks. I updated the text.

Just FYI though, trillion can actually mean two different numbers, so I try to avoid it.
http://en.wikipedia.org/wiki/Trillion


Peataflops on desktop....
By Clauzii on 6/20/2006 6:27:16 PM , Rating: 2
For what do a normal human living being need 1 Petaflop?

For Word? Photoshop? Internet? Music? Movies? Chatting?
LOL!!

MAYBE for games... :S

It's like a car. We'll reach a point wh




RE: Peataflops on desktop....
By Clauzii on 6/20/2006 6:29:10 PM , Rating: 2
*CONT*

We'll reach a point where enough is enough, speed/energy and spec-wise.

I think we are getting at that point :)


For a second...
By AbelIAN on 6/20/2006 8:39:34 AM , Rating: 1
For a second, I thought that headline said "ORLY"!

Seriously, does anyone have any meaningful comments about freakin' supercomputers? Yay, scientific research and insane specs?




"A politician stumbles over himself... Then they pick it out. They edit it. He runs the clip, and then he makes a funny face, and the whole audience has a Pavlovian response." -- Joe Scarborough on John Stewart over Jim Cramer











botimage
Copyright 2010 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki