Military Supercomputer Sets Record
Frank,A LANL story that isn't all doom and gloom.
- anonymous
By JOHN MARKOFF, The NY Times
SAN FRANCISCO — An American military supercomputer, assembled from components originally designed for video game machines, has reached a long-sought-after computing milestone by processing more than 1.026 quadrillion calculations per second.
The new machine is more than twice as fast as the previous fastest supercomputer, the I.B.M. BlueGene/L, which is based at Lawrence Livermore National Laboratory in California.
The new $133 million supercomputer, called Roadrunner in a reference to the state bird of New Mexico, was devised and built by engineers and scientists at I.B.M. and Los Alamos National Laboratory, based in Los Alamos, N.M. It will be used principally to solve classified military problems to ensure that the nation’s stockpile of nuclear weapons will continue to work correctly as they age. The Roadrunner will simulate the behavior of the weapons in the first fraction of a second during an explosion.
Before it is placed in a classified environment, it will also be used to explore scientific problems like climate change. The greater speed of the Roadrunner will make it possible for scientists to test global climate models with higher accuracy.
To put the performance of the machine in perspective, Thomas P. D’Agostino, the administrator of the National Nuclear Security Administration, said that if all six billion people on earth used hand calculators and performed calculations 24 hours a day and seven days a week, it would take them 46 years to do what the Roadrunner can in one day.
The machine is an unusual blend of chips used in consumer products and advanced parallel computing technologies. The lessons that computer scientists learn by making it calculate even faster are seen as essential to the future of both personal and mobile consumer computing.
The high-performance computing goal, known as a petaflop — one thousand trillion calculations per second — has long been viewed as a crucial milestone by military, technical and scientific organizations in the United States, as well as a growing group including Japan, China and the European Union. All view supercomputing technology as a symbol of national economic competitiveness.
By running programs that find a solution in hours or even less time — compared with as long as three months on older generations of computers — petaflop machines like Roadrunner have the potential to fundamentally alter science and engineering, supercomputer experts say. Researchers can ask questions and receive answers virtually interactively and can perform experiments that would previously have been impractical.
“This is equivalent to the four-minute mile of supercomputing,” said Jack Dongarra, a computer scientist at the University of Tennessee who for several decades has tracked the performance of the fastest computers.
Each new supercomputing generation has brought scientists a step closer to faithfully simulating physical reality. It has also produced software and hardware technologies that have rapidly spilled out into the rest of the computer industry for consumer and business products.
Technology is flowing in the opposite direction as well. Consumer-oriented computing began dominating research and development spending on technology shortly after the cold war ended in the late 1980s, and that trend is evident in the design of the world’s fastest computers.
The Roadrunner is based on a radical design that includes 12,960 chips that are an improved version of an I.B.M. Cell microprocessor, a parallel processing chip originally created for Sony’s PlayStation 3 video-game machine. The Sony chips are used as accelerators, or turbochargers, for portions of calculations.
The Roadrunner also includes a smaller number of more conventional Opteron processors, made by Advanced Micro Devices, which are already widely used in corporate servers.
“Roadrunner tells us about what will happen in the next decade,” said Horst Simon, associate laboratory director for computer science at the Lawrence Berkeley National Laboratory. “Technology is coming from the consumer electronics market and the innovation is happening first in terms of cellphones and embedded electronics.”
The innovations flowing from this generation of high-speed computers will most likely result from the way computer scientists manage the complexity of the system’s hardware.
Roadrunner, which consumes roughly three megawatts of power, or about the power required by a large suburban shopping center, requires three separate programming tools because it has three types of processors. Programmers have to figure out how to keep all of the 116,640 processor cores in the machine occupied simultaneously in order for it to run effectively.
“We’ve proved some skeptics wrong,” said Michael R. Anastasio, a physicist who is director of the Los Alamos National Laboratory. “This gives us a window into a whole new way of computing. We can look at phenomena we have never seen before.”
Solving that programming problem is important because in just a few years personal computers will have microprocessor chips with dozens or even hundreds of processor cores. The industry is now hunting for new techniques for making use of the new computing power. Some experts, however, are skeptical that the most powerful supercomputers will provide useful examples.
“If Chevy wins the Daytona 500, they try to convince you the Chevy Malibu you’re driving will benefit from this,” said Steve Wallach, a supercomputer designer who is chief scientist of Convey Computer, a start-up firm based in Richardson, Tex.
Those who work with weapons might not have much to offer the video gamers of the world, he suggested.
Many executives and scientists see Roadrunner as an example of the resurgence of the United States in supercomputing.
Although American companies had dominated the field since its inception in the 1960s, in 2002 the Japanese Earth Simulator briefly claimed the title of the world’s fastest by executing more than 35 trillion mathematical calculations per second. Two years later, a supercomputer created by I.B.M. reclaimed the speed record for the United States. The Japanese challenge, however, led Congress and the Bush administration to reinvest in high-performance computing.
“It’s a sign that we are maintaining our position,“ said Peter J. Ungaro, chief executive of Cray, a maker of supercomputers. He noted, however, that “the real competitiveness is based on the discoveries that are based on the machines.”
Having surpassed the petaflop barrier, I.B.M. is already looking toward the next generation of supercomputing. “You do these record-setting things because you know that in the end we will push on to the next generation and the one who is there first will be the leader,” said Nicholas M. Donofrio, an I.B.M. executive vice president.
By breaking the petaflop barrier sooner than had been generally expected, the United States’ supercomputer industry has been able to sustain a pace of continuous performance increases, improving a thousandfold in processing power in 11 years. The next thousandfold goal is the exaflop, which is a quintillion calculations per second, followed by the zettaflop, the yottaflop and the xeraflop.
52 comments:
Unfortunate that this computer is going classified. If it were kept in the open, it would soon be evident that LANL doesn't have enough scientists left to keep it busy.
Maybe Bechtel can use the computer to keep track of the money they steal from the LANL budget.
"A LANL story that isn't all doom and gloom"
LOL...based on first comment, I think not :)
OK, let's cut to the quick. What games can it play? Video frame rates? And most important of all, does it support Microsoft Office?
This article was fun.
I seem to know how to get this computer to go faster and run more cheaply.
I also propose that beyond yottaflop we create the term 'lottaflop.'
Later
It's my understanding that most of the intended users in the nuclear weapons community see Roadrunner as an impractical machine for getting real work done due to its extremely funky architecture.
Is this true, or have the programming libraries solved most of the programming impediments?
http://bits.blogs.nytimes.com
/2008/06/10/apple-in-parallel
-turning-the-pc-world-upside-down
/?ref=technology
Further information of the Roadrunner can be found here:
1) www.lanl.gov/news/index.php/fuseaction/nb.story/story_id/13552/nb_date/2008-06-09
2) www.lanl.gov/news/newsbulletin/QuickTimes/Roadrunner.larger.mov
(Brian Albright, Charlie McMillan, Lin Yin, John Turner, John Hopson, and Mike Anastasio are in the video clip.)
(Associated Press, ABC News, CBS News, CNN, FOX News, MSNBC, Washington Post, Reuters, The Huffington Post, Yahoo! News, Google News, et cetera also covered the Roadrunner story, June 9, 2008.)
Why does the article refer to it as a "military computer"?
"Unfortunate that this computer is going classified. If it were kept in the open, it would soon be evident that LANL doesn't have enough scientists left to keep it busy." (6:39 AM)
LANL's chief scientist for Roadrunner software development recently announced that he is bailing out of LANL and heading over to ORNL. This is not a good sign, IMHO.
The loss of an extremely bright computer scientist like this should be sending up red flags for upper management. Unfortunately, I don't think LANS is paying any attention to the steady brain drain that is occurring at LANL. The current environment at LANL has little to entice good scientists to stay and LANS will do nothing to alter this situation.
6:29 pm: "Why does the article refer to it as a "military computer"?"
Because it's going to be used to design and develop military weapons??
8:25pm: Another one, eh? The computing divisions have been losing the good people left and right. I was in the burst that left around the LANS transition, and every time I talk to my buddies who remain, they're either about to jump ship or know someone who has. Roadrunner looked like a quagmire before the brain drain - I'd hate to be one of the poor suckers who has the responsibility of porting codes to it.
"The loss of an extremely bright computer scientist like this should be sending up red flags for upper management. Unfortunately, I don't think LANS is paying any attention to the steady brain drain that is occurring at LANL. The current environment at LANL has little to entice good scientists to stay and LANS will do nothing to alter this situation."
The loss of one scientist isn't going to send up red flags. Many good computer scientists sent up red flags stating that this machine was a bad idea long before now, yet they were overruled by layers of people who have floated up to middle or upper management who saw nothing but the prospect of winning the latest supercomputer pissing contest, regardless of how useful the machine would be to users. Everyone who has touched this machine or the cell-based systems like it knows that porting realistic, relevant multiphysics models to the platform is an exercise in complete pain and futility. The software tools for it are at the level of those that existed for the old Control Data Corporation machines in the 1960s, yet the codes of interest are many orders of magnitude more complex than those that LASL folks struggled with 30 years ago. Bottom line -- this machine is Pete Dominici's earmark baby that is his last act of giving a huge pile of money to the lab in the computing area.
If you want to know how relevant the machine actually is, ask what type of code actually achieved the petaflop. It is both completely unclassified, and extremely irrelevant to the real workload lab supercomputers are expected to support. You'll never see that in the press releases. The machine is currently running in an open, unclassified environment with open, unclassified codes -- so it is fair game to ask -- how relevant is the code that hit the petaflop to the codes that the machine was purchased for. I assure you - this benchmark is completely and totally irrelevant when it comes to answering the needs and demands of the production lab users.
If LANL really did hit the petaflop/s mark using the standard Top500 benchmark code (suitably ported) -- even if it's only a peak number rather than sustained -- it's big news and a coup for the Lab. And the fact that the hardware configuration is something of a hybrid, and therefore hard to program, hasn't stopped anyone before. It's just challenged them. (Who is around to be challenged is, of course, a separate issue.)
A bigger question is "why"? Aside from the rather silly pissing contest for operating the fastest machine, there used to be a reasonable argument for this sort of investment -- the old, cold-war (and post-cold-war) "intellectual deterrent" rationale.
But I expect that the AQ boys aren't really going to be impressed by such an argument, even a little bit. And whether the other nuke-enabled countries of questionable stability are impressed by this rationale is uncertain as well.
Maybe all this processing power can be put to use somehow on the analysis of the contents of all those containers coming into the US on ships every day. Ya think?
Yes, yes, yes, but take a look at my enormous RoadRunner BONUS CHECK!
Why should a lab that will be devoted to pit construction even need this computer?
Roadrunner is mostly flash with little substance for handling real, everyday problems. It's a nice PR machine for LANS. In fact, it even makes the old LANL ASCI-Q machine look extremely practical by comparison! That's quite a feat.
Poster 4:47 AM is right on the mark. Roadrunner was St. Pete's last hunk of big pork for LANL. The money could have been used to buy a slightly slower, but far more practical machine, perhaps one built around quad-cord AMD or Intel processors. However, that's just not the LANL way. Flash over function. That's the LANL way.
"so it is fair game to ask -- how relevant is the code that hit the petaflop to the codes that the machine was purchased for. I assure you - this benchmark is completely and totally irrelevant when it comes to answering the needs and demands of the production lab users."
Probably true for all the NWC supercomputer scoring, since the scoring is on unclassified codes. The performance relationship to classified codes varies depending upon the problem.
As someone else said, the flop foot race is a P.R. campaign. We need to focus on solving problems, not how many flops we can string together.
"We need to focus on solving problems, not how many flops we can string together."
My goodness! Would somebody please make this person the next director!
"I thought the mantra so many of you keep chanting is science..."
Exactly. The point is some have become enamored with the tool rather than the outcome. The fact that a tool is unique (or cost a bazillion $) is irrelavent if it cannot solve a problem, or solves it for more money and effort than alternatives.
For Christsake, the damn thing (phase 2) hasn't even been delivered here and the malcontents, accurately portrayed by 1:13 PM) are saying it's a bunch of crap. Sounds like the whiner/complainers are taking a Mechels approach of denigrating anything and everything LANL.
You want this place to get better...take some frackin' pride wherever you can get it. Otherwise shut the hell up and go someplace else;
assuming you're actually an employee.
Tech Note:
Please stop referring to Roadrunner as a "pissing contest."
The proper technical term is "E-Penis contest".
Got it? Ok, continue on :)
"Everyone who has touched this machine or the cell-based systems like it knows that porting realistic, relevant multiphysics models to the platform is an exercise in complete pain and futility." - 4:47 AM
Good point, but I suggest we all just keep drinking the Kool-aid that 6:25 PM is dispensing and pretend that everything is fine with Roadrunner. To critique its many shortcomings would be disloyal to LANL. Besides, the scientific method of proving things is hard to do, so why even bother?
It has been pointed out that the machine while being very fast may not be the correct tool for the job. But bragging rights will always trump practicality. Try explaining to a congressman why a lesser peak performance machine would work better when all he wants to see is "we're number 1". Show a senator a Ferrari and pickup truck, and say the truck will haul that load of manure better but is not as fast as the sports car, they'll foot the bill for the sports car and not the truck.
So you end up getting the Ferrari then spend a lot of time and effort trying to weld in the pickup bed. It might or might not work, but if you didn't play the penis envy game, you would not have gotten any machine at all.
This is America, we love bright lights and loud noises. Especially if you can get someone else to foot the bill.
"This is America, we love bright lights and loud noises. Especially if you can get someone else to foot the bill." 6/11/08 11:23 PM
This may be true, but it is part of the attitude problem that is currently sending America into the global dumpster. It's part of the PR machine and celebrity craze that dominates American culture over things of solid substance.
We're living off candy bars. The temporary sugar buzz sure feels great, but coming down off that sugar buzz is going to be a real bitch.
"Everyone who has touched this machine or the cell-based systems like it knows that porting realistic, relevant multiphysics models to the platform is an exercise in complete pain and futility." - 4:47 AM
Good point, but I suggest we all just keep drinking the Kool-aid that 6:25 PM is dispensing and pretend that everything is fine with Roadrunner. To critique its many shortcomings would be disloyal to LANL. Besides, the scientific method of proving things is hard to do, so why even bother?
6/11/08 8:43 PM
Exactly. What does the scientific method have to do with weapons designers, anyway? Sadly, absolutely nothing. But if it makes the Labs sound good, then keep that Koolaid flowing. After all, fast machines are cool, and appearances are everything.
anonymoron 6/11/08 4:47 AM drooled:
"Everyone who has touched this machine or the cell-based systems like it knows that porting realistic, relevant multiphysics models to the platform is an exercise in complete pain and futility. The software tools for it are at the level of those that existed for the old Control Data Corporation machines in the 1960s, yet the codes of interest are many orders of magnitude more complex than those that LASL folks struggled with 30 years ago."
These statements reveal a complete and utter lack of knowledge about both Roadrunner and the future.
The same misinformed throwback also asserted:
"I assure you - this benchmark is completely and totally irrelevant when it comes to answering the needs and demands of the production lab users."
Two points:
1. No one would claim that the LINPACK benchmark is relevant. That's why it's not the only application running on the machine. Get a clue.
2. Your precious "production lab users" have resisted every single change in computing, no matter how inevitable or beneficial. This is a crowd that still includes people who use line editors and resisted Unix, make, etc. If decisions were left to them, LANL would be using a giant slide rule (or at best a bunch of Crays running CTSS).
Erm, this whole discussion makes me happy that I'm a chemist. :-)
5:02pm: Correct about LINPACK. It's a shame that is still the primary benchmark used to measure performance for lists like top500. Unfortunately, the other benchmark suites developed over the years like parkbench and others never got accepted to replace the old standard.
Do you remember the uproar that occurred when the last Cray was decommissioned at the lab? It was amazing. Your reference to line editors I can only assume refers to fred, the precious Fortran-based editor. (I thought fred was a joke when someone told me about it the first time a few years ago - I was speechless when I was told it was still a serious topic after the year 2000.)
Roadrunner has it's issues though. The cell is not a mature processor, and the development tools are quite primitive. Even the gaming community has had big problems moving to it for the PS3. Given the sheer volume of money that the gaming world has to throw at the problem, it's telling that they're even finding it to be a hard processor to digest and make good use of. Hopefully the situation on the cell will turn around by the time the machine gets installed at LANL, because if the deep-pocketed gaming community can't make it as programmable as established architectures, the less-than-ideally funded LANL community is not likely to either. Especially if, as one of the previous commenters stated, key computer scientists on the project have left for places like ORNL.
Erm, this whole discussion makes me happy that I'm a chemist. :-)
Apparently you have no clue about
chemical structure prediction and large scale molecular modeling. Ok, the ribosome is a biological unit but the structure prediction was done at the atomistic/chemical level (calculated on the ASCI-Q machine):
http://preview.tinyurl.com/657tu7
Here is the heart of the situation in regards to Roadrunner. The 'bread-and-butter' computations performed at Los Alamos involve hydro-code simulations. If Roadrunner can't perform the current and future hydro-code calculations better than a more conventional architecture for the $100 M cost, then it was a unwise procurement for the lab.
It's my understanding that the recent 1 Pentaflop computation was made doing an image processing code called Pentavision that simulates the neurons in the human visual cortex. The 1 PFlop rate was not made doing hydro-code problems.
Roadrunner's unusual architecture looks well equipped to tackle signal and image processing type problems, but as I mentioned at the top of this post, hydro-codes are what LANL is all about. In addition to this, many signal and image processing problems can be done very economically at very high rates of speed using low costs FPGAs and DSP based solutions. Using a custom designed $100 M super to perform these tasks is costly overkill.
Let's see what the benchmarks are when Roadrunner does weapons style hydro-code problems for everyday research at LANL. If it can do this effectively and faster than more traditional supercomputers at a $100 M cost mark, then it will be a true success. I hope it will be, but the jury is still out on this one.
6/10/08 8:43 AM said...
"It's my understanding that most of the intended users in the nuclear weapons community see Roadrunner as an impractical machine for getting real work done due to its extremely funky architecture."
As 6/12/08 5:02 PM pointed out, most "intended users in the nuclear weapons community" are extremely resistant to change of any kind. Many dismiss use of C, much less C++ or scripting languages, as radical crazy talk.
The ones who continue to stick their heads in the sand and think they can ignore Roadrunner and hope that compilers will solve all their problems on future machines are sadly mistaken.
The smart ones will realize that Roadrunner provides an opportunity to prepare for future architectures now.
6/13/08 11:45 AM said:
"Let's see what the benchmarks are when Roadrunner does weapons style hydro-code problems for everyday research at LANL."
Please take a look at the presentations at the LANL Roadrunner site, particularly the Roadrunner Techincal Seminar Series. If you know as much about the 'bread and butter' computations at LANL as you claim to, you'll understand why your post badly mis-represents the relevance of the work that has already been done.
6/10/08 8:25 PM wrote:
"LANL's chief scientist for Roadrunner software development recently announced that he is bailing out of LANL and heading over to ORNL. This is not a good sign, IMHO."
I'm assuming that the above refers to me. If so, I think I should explain a few things. I'm actually Group Leader of the Computational Physics Group (CCS-2), a part of the Computer, Computational, and Statistical Sciences Division (CCS). CCS-2 participates in a wide range of research in computational science such as hydrodynamics and radiation transport methods and ocean modeling for the Climate, Ocean, and Sea Ice Modeling (COSIM) project, a critical part of the larger Community Climate System Model (CCSM), in addition to research into algorithms for advanced computing architectures such as Roadrunner.
My role in Roadrunner was actually quite small. Officially, I led the algorithms/applications team, which was tasked with determining whether or not Roadrunner would be useful for applications of relevance to core LANL missions. We did not enter into this assuming the answer would be yes - if you ask anyone on the team they will confirm that I said throughout the year that they would still be successful even if the verdict at the end of the year was that LANL should not move forward on Roadrunner.
In the end, they demonstrated incontrovertably that Roadrunner absolutely will be used for relevant applications. As 6/13/08 3:13 PM states above, you can look through the Technical Seminar Series slides for more on this.
Back to my role and my departure for ORNL. As I noted above, my role as algs/apps team lead meant that in effect I went to a lot of meetings, did a little project management, did a little technical leadership (very little, since a team of this caliber needs very little leadership), and had the distinct honor of presenting the outstanding results of an incredible collection of extremely talented staff in a number of venues.
That team will remain at LANL, and they will continue to demonstrate the success that Roadrunner truly is. Watch for further announcements over the next weeks and months, because results are already being obtained even while the machine is still in Poughkeepsie - not just on performance of the machine, but actual science. I have no doubt that the criticism will continue, but with each announcement the naysayers will sound more out of touch with reality.
The bottom line is that the team and Roadrunner will be perfectly fine without me.
Finally, it's important to point out that my degrees are in Nuclear Engineering, and I'm going to ORNL to build a group to develop applications for energy simulation. Obviously, a primary thrust will be nuclear power, but it's really an attempt to bring together two of my areas of interest - high-performance computing and energy security. So if we can bring HPC to bear on scientific questions of importance to solar, wind, bio-fuels, or energy storage, we'll do so.
If anyone would like to talk more about Roadrunner (results, implications, future architectures, or anything else) or energy, please feel free to contact me. I hope this helps to clarify things.
-John Turner
Apparently you have no clue about
chemical structure prediction and large scale molecular modeling. Ok, the ribosome is a biological unit but the structure prediction was done at the atomistic/chemical level (calculated on the ASCI-Q machine):
http://preview.tinyurl.com/657tu7
Right you are! You missed my point. All the computer talk is greek to me.
Didn't say the computers weren't worthwhile and useful. Gheesh.
"If you know as much about the 'bread and butter' computations at LANL as you claim to, you'll understand why your post badly mis-represents the relevance of the work that has already been done." (3:13 PM)
I don't claim to be an expert at weapons hydro-code computations. However, I do know they are a major part of the computer efforts at LANL.
From your tone, I suspect you may have taken my post as some type of insult. No insult was intended. Regardless of how well Roadrunner is suited for performing LANL tasks, the Roadrunner team can be proud of their accomplishments to date.
I'm am asking question about the long-term suitability of this highly unusual architecture to doing hydro-code simulations. Yes, I know some of the initial figures look impressive, but I still have my doubts. Time will tell. The Roadrunner naysayers (of which there are plenty at LANL) aren't ready to say it's a success just yet.
2. Your precious "production lab users" have resisted every single change in computing, no matter how inevitable or beneficial. This is a crowd that still includes people who use line editors and resisted Unix, make, etc. - 5:02 PM
What's wrong with tools like FRED? Some of the brightest scientist in the world used the FRED editor to help design the "Ferrari's" of the US nuclear weapons arsenal. What have you helped design of vital consequence during your short LANL career, 5:02 PM? Precious little, I suppose.
It's not the tools, but how good you are at using then that counts, and some of us are very good at it by the looks of the weapons in the US stockpile.
6/14/08 11:27 PM puffed:
"What's wrong with tools like FRED?"
FRED is a nice line editor, no question. But guess what? Sometimes the new thing is actually an improvement. I know it's tough to actually learn something new, but sometimes it's worth it.
"Some of the brightest scientist in the world used the FRED editor to blah blah blah"
When FRED was the best there was, great. When HISTORIAN was the best there was, great. But refusing to learn new tools just because what you know is "good enough" is a path to mediocrity.
"It's not the tools, but how good you are at using then that counts,..."
And are you still telling that to your wife?
Good luck John - sad to see you moving on.
"My role in Roadrunner was actually quite small. Officially, I led the algorithms/applications team, which was tasked with determining whether or not Roadrunner would be useful for applications of relevance to core LANL missions" (John Turner)
I don't understand, John. How could you "officially" lead the Roadrunner algorithms team, and yet have little to actually do with Roadrunner? I'm not trying to be critical of your post or you, but I find this comment perplexing.
Does this indicate that LANL puts people into "official" leadership positions, but they then have little to do with what they are tasks with leading? I'm confused here.
Regardless of this, LANL is losing a great scientist with the loss of John Turner. He knows a great deal about computers and he will be missed. Good luck and best wishes on your new position at ORNL!
6/15/08 3:43 AM said:
"Good luck John - sad to see you moving on."
thanks very much - I appreciate it
-JT
6/15/08 11:38 AM said:
"I don't understand, John. How could you "officially" lead the Roadrunner algorithms team, and yet have little to actually do with Roadrunner? I'm not trying to be critical of your post or you, but I find this comment perplexing."
What I mean is that although I tried very hard to do everything I could to empower the team, shield them from whatever crisis or political stuff was happening, and certainly felt plenty of stress, the folks that really deserve the credit are those who did the actual algorithm and application development.
"Does this indicate that LANL puts people into 'official' leadership positions, but they then have little to do with what they are tasks with leading? I'm confused here."
I don't think so, at least in this case.
I'm sure if you asked them you'll find that I made plenty of mistakes along the way this year (passing on too much information or too little, missing too many meetings with them, etc.), but I tried to stay as close to the work as possible and encourage without interfering.
If anything, it was my other duties as CCS-2 Group Leader that suffered due to my attention to Roadrunner. In that regard I owe a great deal to my fantastic deputies and admins.
"Regardless of this, LANL is losing a great scientist with the loss of John Turner. He knows a great deal about computers and he will be missed. Good luck and best wishes on your new position at ORNL!"
I appreciate that a great deal. Thank you very much.
-JT
How's this for wild spin?...
www.channelregister.co.uk/
2008/06/13/
roadrunner_drives_car_outmodes_humans/
******************
Boffins: Roadrunner hypercomputer could drive a car - Channel Register
*****************
By Lewis Page
13 Jun 2008 12:43
A bloody big car. Also, human obsolescence imminent
American nuke boffins who have just fired up the world's first petaflop hypercomputer* are extremely excited, and contend that the machine may enable them to accurately simulate important segments of the human brain. Conceivably, the mighty "Roadrunner" - as the computer is known - may exhibit capabilities verging on human cognition.
“Roadrunner ushers in a new era for science,” said Terry Wallace of the Los Alamos National Lab - birthplace of the atomic bomb.
“Just a week after formal introduction of the machine to the world, we are already doing computational tasks that existed only in the realm of imagination a year ago.”
Specifically, it seems that Roadrunner's unprecedented computational puissance has allowed the American boffins to "model more than a billion visual neurons ... to reach a new computing performance record of 1.144 petaflop/s. The achievement throws open the door to eventually achieving human-like cognitive performance in electronic computers".
The excited brainboxes reckon that this type of computing power would be able to achieve visual tasks which thus far only the human brain can accomplish, such as handling a car in dense rush-hour traffic.
Terry has recently been making some pretty outrageous claims to the press about Roadrunner, but this latest one really takes the cake:
www.computerworld.com/
action/article.do?
command=viewArticleBasic&
articleId=9098439
*** Los Alamos: Roadrunner as important as first computer ***
June 16, 2008 (Computerworld)
A research director at Los Alamos National Laboratory said the addition of a petascale supercomputer is as big a leap forward as when scientists got their hands on their first computer ever.
IBM's Roadrunner supercomputer, which was built for the federal Department of Energy's Los Alamos National Lab, smashed the high-tech equivalent of the four-minute mile last month by breaking the lofty petaflop barrier.
"I think it allows us to explore problems we couldn't think of before, but my real hope is that this is a new generation of how we look at science," said Terry Wallace, principal associate director for science, technology and engineering at Los Alamos. "This is as big as when scientists were first given a computer. It's like driving to work in a real car instead of on a tricycle."
What kind of car does Terry drive to work?
Frank Young said, "What kind of car does Terry drive to work?"
Probably this one...
http://tinyurl.com/5nmxj7
X*X*X*X*X*X*X*X*X*X*X*X*X*X*X*X*X*X*X*X*X*X
Roadrunner to Cure Cancer!!! (Odd News, June 18, 2008)
X*X*X*X*X*X*X*X*X*X*X*X*X*X*X*X*X*X*X*X*X*X
Dr. Terry Wallace, Director of research at Los Alamos National Lab, has announced that the new Roadrunner supercomputer being installed at the New Mexico lab will soon be able to cure cancer. "We don't have the necessary algorithms for making this breakthrough," said Dr. Wallace, "but we should be able to type in a query, Google style, and expect to see the correct answer within about an hour."
Dr. Wallace said that some of the researchers who work in the presence of this machine have already noticed the sudden appearance of whiter teeth, higher IQs, and significant growth in certain male parts of the human anatomy. "What can I say?", said a cheerful Dr. Wallace. "It's a fantastic machine!"
It is expected that Roadrunner will be hooked up to the internet sometime later this fall. At that point, Dr. Wallace expects it will only be a short period of time until Roadrunner decides to start talking to other computers throughout the world and takes over the Earth, ushering in a glorious 1000 year period of world peace.
It's perhaps worth adding a footnote to this discussion to note that Roadrunner did indeed make it onto the June, 2008 Top500 list as the first petaflop/s machine -- so the press release and so on wasn't just LANL hype.
This doesn't change the value of the LINPACK benchmark or anything; but it does put LANL back on the big-time computing map, for the time being, at least.
TOP500 List - June 2008 (1-100)
(www.top500.org/list/2008/06/100)
Although I have consistently insisted that the LINPACK benchmark isn't very representative of LANL applications, there are a couple of reasons it should not be dismissed completely.
First and foremost, the IBM team that did the hybrid LINPACK implementation for Roadrunner deserves a lot of credit. They did an outstanding job, as you'll be able to see for yourself when details are published and the code is released. As you may or may not know, they were able to exceed 1 PF/s with 17 CUs rather than the 18 in the original design (see the technical presentations pointed to above for details of the machine and an explanation of "CU"). I don't know of anyone who thought that was possible, and many expected them to need to add a 19th CU.
Second, a little-reported fact is that IBM was able to complete the full-system petaflop/s LINPACK run only 3 days after the machine was assembled and turned over to the LINPACK team. This is an astounding achievement, and indicates far greater system stability than is typical of a machine of this scale. This includes not just the hardware but also the system software.
And relevant or not, LINPACK is an excellent stress test for a machine. The fact that it was able to run for several hours so soon after assembly is impressive.
I'm guessing many here don't fully comprehend this achievement. That's fine - just don't be so quick to brush it off as insignificant and irrelevant, because it's not.
Again, there's no question that the science and delivery to DOE/NNSA/LANL programs is of paramount importance. Just give credit where credit is due - IBM built a solid machine and demonstrated impressive expertise in hybrid software development with their LINPACK implementation.
John, just curious, but is ORNL having thoughts about going with an IBM Cell-based super, also?
"John, just curious, but is ORNL having thoughts about going with an IBM Cell-based super, also?"
not that I know of... but remember, one of the things that we have emphasized is that this is not just about Roadrunner or Cell - RR embodies many of the architectural features that are coming, whether from Cray or Intel or any of the vendors (multicore processors with short-vector instructions, local stores instead of cache, threading, heterogeneous instruction sets, relatively remote accelerators, etc.)
existing codes may be able to "just compile and run" and get reasonable performance for maybe another generation of machines, but after that, they are going to have to address the issues
(you may be interested in this discussion with Kathy Yelick of UCB and LBL on this topic)
LANL has an opportunity to get ahead of the curve here, and some applications will realize this and take advantage of it
finally, although it might not seem like it, I'm actually pretty agnostic about hardware - it's interesting and all, but I care more about the apps and whether the hardware can deliver
-JT
ORNL announced plans for a petascale Cray a couple years ago:
http://www.hpcwire.com/offthewire/17885309.html
Post a Comment