Atomic City Underground
OAK RIDGE - The next golly-gee-whiz supercomputer at Oak Ridge National Laboratory will be called "Titan," and it will reportedly be capable of performing 20,000 trillion mathematical calculations per second - a capability also known as 20 petaflops.
That's about 10 times faster than Jaguar, ORNL's Cray XT5 supercomputer that is currently ranked the second-fastest machine in the world and arguably the best anywhere at producing science.
The first cabinets of ORNL's new Cray system are expected to arrive before the end of the year, according to Jeff Nichols, associate lab director for scientific computing. The initial units will be operated as a test-bed before the full system is delivered sometime in 2012, Nichols said.
Titan is a follow-up to ORNL's highly successful Jaguar system and an outgrowth of the big Department of Energy competition the lab won in 2004 for leadership computing.
Following a series of reviews, DOE gave the lab approval to negotiate a contract with Cray for a supercomputer that will include NVIDIA's accelerating GPU technology.
Nichols said the cost of the computer will be in the range of $100 million.
The timing of the deliveries will be of interest, too, because there's not much room left at the National Center for Computational Sciences - where Jaguar and the lab's other supercomputers reside. Nichols acknowledged the existing facility is pretty much maxed-out in terms of space, power and chilled water capacity.
A new 160,000-square-foot computer facility is planned for the east end of the ORNL campus, but it's not yet built. In fact, it's not yet under construction.
Nichols confirmed that a proposal to privately finance the new facility is still under review at the federal Office of Management and Budget.
"We really need the facility - not only for our DOE aspirations, but our 'work for others' aspirations," the ORNL official said. The Oak Ridge lab houses not only DOE's top supercomputer, but also the top machines of the National Science Foundation and the National Oceanic and Atmospheric Administration.
Meanwhile, ORNL is looking ahead to the next big thing, which is known as exascale computing - a thousandfold increase in capability over today's Jaguars and such.
ORNL has joined with Los Alamos and Sandia National Laboratories as partners in the high-priority supercomputing effort over the next few years.
The three labs signed a memorandum of understanding to form the Scientific Partnership for Extreme Scale Computing, and they will compete against another team - of Lawrence Livermore, Lawrence Berkeley and Argonne National Laboratories - to help the U.S. retain or regain world leadership in computing.
Senior writer Frank Munger may be reached at 865-342-6329.
PrepXtra Girls' Track First Team
BET Awards
Reader's Eye 2012
Scripps Interactive Newspapers Group
Comments » 36
cjensen writes:
Oh, my, just reading the headlines gives me a sinking feeling.
Debt writes:
two drums and a cymbal fall off a cliff...
sphynxgirl writes:
Another supercomputer in this economy???? What exactly do these supercomputers do to make society better?? Oak RIdge is looking at the loss of hundreds of jobs with the upcoming budget cuts yet there is money available to build a new facility and house yet another one of these monsters? Shouldn't we be investing in alternate fuel sources or ways to increase food production for the ever-growing population of the world? What 'science does the Jaguar computer produce and what does it do for mankind? I'm an avid news junkie and I have yet to read one article about how these machines have made the world a better place. Please enlighten me and I'm sure there are others that would love to know what we are getting for our tax dollars here.
madmaxvol writes:
I'm surprised that a self proclaimed "avid news junkie" didn't see this article in last week's News Sentinel.
http://www.knoxnews.com/news/2011/mar...
aeronicapilot writes:
http://www.msnbc.msn.com/id/41498300/...
jsternbe#298183 writes:
Here is an example of something done on Jaguar (the current fastest computer at ORNL)from last month that isn't so esoteric that you need a Ph.D. to understand it.
http://blogs.knoxnews.com/munger/2011...
$5 billion annually on saving our country's fuel for a project on the order of $100 million sounds like a good investment to me! There is a lot of good stuff coming out of the ORNL supercomputer program but it is easy to miss when it gets overshadowed by all of the other stuff happening around the world.
rayt435 writes:
Really? Just to play solitaire.
jsternbe#298183 writes:
You beat me to it!
calvinbama (Inactive) writes:
so cynical. There is outside funding to install one of the most powerful machines in the world right here in East Tennessee. It does research for the smartest people in the country, and provides high quality knoxledge economy jobs and spinoffs in Knoxville and Oak Ridge. Yet the cynical posters on here moan and complain just like they do with everything else. Give me a break and please do something else with your life one day.
rayt435 writes:
Yeh, that is one of those "duh" articles. Everyone knows that if you streamline a vehicle it moves more easily through the air and uses less fuel.
A few days in a wind tunnel would have provided all the information that was needed to optimize the airflow around a truck. Any additional fractions of a thousandth of a percent gained by the supercomputer would be insignificant. And most of those trivial savings would be offset by damaged wind deflectors.
But instead ORNL has to use their newest toy, to justify the expense of their toy, doing something could be accomplished using other cheaper tools. NASCAR does it all the time.
Yeh, it is nothing by bragging rights. My truck is louder than your truck, my dog is smarter than your dog, a waste of money.
cjensen writes:
sphynxgirl:
Another supercomputer in this economy???? What exactly do these supercomputers do to make society better?? Oak RIdge is looking at the loss of hundreds of jobs with the upcoming budget cuts yet there is money available to build a new facility and house yet another one of these monsters? Shouldn't we be investing in alternate fuel sources or ways to increase food production for the ever-growing population of the world? What 'science does the Jaguar computer produce and what does it do for mankind? I'm an avid news junkie and I have yet to read one article about how these machines have made the world a better place. Please enlighten me and I'm sure there are others that would love to know what we are getting for our tax dollars here.
----------------------
You're right, stinxgirl, we don't need no silly new computers. We need guns, more guns, because guns are what made this country great. The Floundering Fathers didn't mention nuttin' about computers in our floundering documents, but they mentioned guns, so if we want to be FREE, we need guns, and more guns.
rayt435 writes:
Post a link to any REAL WORLD applications that were developed by these supercomputers that make any of our lives better. I don't want "pie in the sky" stuff, theoretical musings, but real world examples. I don't need to know what they can do, but what they have done in the real working world.
WillyP writes:
Golly, I sure hope they budgeted for another McAfee anti-virus software license...
profut writes:
Some comments on why supercomputing is of interest:
First off, it's noteworthy that ORNL was home to the fastest machine in the world. Now China is home to the fastest computer. Titan will be about 10x faster than Jaguar - it remains to be seen if it will be fastest in the world when it is fielded. Also note that by the end of the decade machines approaching 1000x faster are planned to be deployed.
I started working on parallel computers 20+ years ago and have watched as speeds have increased by about a million times faster.
So back to the question of why anyone should care. Some have mentioned climate modeling, and that's true. We also can do much more accurate weather modeling to predict the paths of hurricanes and other storms.
For the power grid, we'd like to be able to predict the behavior of the system and its response to dynamic changes in load as well as in production (e.g., accounting for changing wind speeds for windmills/turbines).
There's lots of fundamental research to understand materials, chemistry, and physics. This is important for gaining the knowledge to enable the nanotechnology developments that will help drive development in the future.
Computing is transforming biology and medicine, enabling explorations of sequences and the behavior of molecules and genetic circuits. This leads to potential drug exploration with computers instead of in the lab - a much more cost effective and affordable approach. Check out DE Shaw and the custom supercomputers they are building for molecular dynamics to use with drug design.
Supercomputing has been used to understand where oil deposits can be found underground - the oil companies have been big supercomputer users for years.
Scientists and engineers exploring next generation reactors and other energy generation approaches use supercomputers to explore designs, potential dangers, as well as opportunities.
Engineers use supercomputing to understand stresses and strains for various structures, fluid dynamics for engines and wings, and a host of other applications.
Finally, one major reason that the Dept of Energy uses supercomputing is that it has allowed us to stop setting off nuclear weapons for testing - it's done using simulations on huge, fast machines now. I'd like to think that's a good thing...
For all of the above, you can't give computational scientists and engineers enough cycles of supercomputing! They just will seek out even more challenging problems to tackle.
One more thing to point out - Titan will use graphics processors that are similar to what you might use to play games. Instead of rendering monsters to shoot though, one can perform linear algebra to model electrical circuits or quantum chemistry.
madmaxvol writes:
Coal Gassification Research
http://www.greencarcongress.com/2011/...
Anti-Coagulant Research
http://www.knoxnews.com/news/2010/nov...
Genome Research:
http://www.scientificcomputing.com/ne...
So if you are concerned about energy consumption, heart/circulatory disease, or cancer then you might see a benefit in having a supercomputer that can help advance research in these areas.
madmaxvol writes:
Oops...forgot about helping develop the next generation of nuclear reactors
http://www.nccs.gov/2010/01/04/nccs-s...
profut writes:
It's probably a waste of time responding, but how about this practical idea:
Check out how many companies have supercomputers. If you go to the Top 500 site that lists the 500 fastest supercomputers (at doing some linear algebra) and do a search for which are in industry (instead of gov't, academia, and research), you'll find that 282 of the top 500 are used by companies for their work.
<sarcasm> Now, I'm sure that all of these companies are wasting their money. After all, there's no good practical reason to use these big machines. </sarcasm>
Or, you could note that while the gov't and academic researchers are pushing the frontiers with the largest machines, they are also helping us to develop what will be typical high-end machines for a couple years from now or the technologies on desktops for a decade from now (if not sooner).
You might not realize it, but the computer industry has been one of the strongest and most vibrant parts of our economy over the past few decades. I don't consider it a complete coincidence that research (including gov't funds) has gone into this area for decades. I also think the investment has more than paid off.
That said, I don't see why continued investment is a bad idea. Probably one of the better investments to be made, in my opinion.
WilliamDawes writes:
As an engineer, I cannot comment on this news in an unbiased manner or respond to the usual gaggle of naysayers in the reasoned, rational fashion they deserve.
That said, Yippee!!! Yahoo!!!! Woweeee!!!!! 20 petaflops? Sweet mama, what a machine!!!
cjensen writes:
profut writes:
in response to rayt435:
Post a link to any REAL WORLD applications that were developed by these supercomputers that make any of our lives better. I don't want "pie in the sky" stuff, theoretical musings, but real world examples. I don't need to know what they can do, but what they have done in the real working world.
It's probably a waste of time responding, but how about this practical idea:
Check out how many companies have supercomputers. If you go to the Top 500 site that lists the 500 fastest supercomputers (at doing some linear algebra) and do a search for which are in industry (instead of gov't, academia, and research), you'll find that 282 of the top 500 are used by companies for their work.
<sarcasm> Now, I'm sure that all of these companies are wasting their money. After all, there's no good practical reason to use these big machines. </sarcasm>
Or, you could note that while the gov't and academic researchers are pushing the frontiers with the largest machines, they are also helping us to develop what will be typical high-end machines for a couple years from now or the technologies on desktops for a decade from now (if not sooner).
You might not realize it, but the computer industry has been one of the strongest and most vibrant parts of our economy over the past few decades. I don't consider it a complete coincidence that research (including gov't funds) has gone into this area for decades. I also think the investment has more than paid off.
That said, I don't see why continued investment is a bad idea. Probably one of the better investments to be made, in my opinion.
-----------------------
Sure, you would think that way, because obviously you are one of those Educated Elitists. See, if you were a real American and Great Patriot, you'd want more guns and real solutions, that work.
Arounds here, our motto is 'if you can't drink it, eat it or chamber it, then it's a waste of money'....besides, that's our money you're spending and we know how to spend it better than you do, so just give it back. If we wanted one of them thar super-duper-computers, we'd buy it ourselves; we don't need to depend on government to do things for us.
We simply have too many people in Oak Ridge on the government dole, when they should be doing something productive with their lives, like being a greeter at Walmart or shift manager at McDonalds.
whizkidtn writes:
Excellent post! +1
BTW: Cuda running on these GPU's is pretty awesome.
whizkidtn writes:
CJ, you've got your sarcasm mode on full I see. The problem with you CJ is that you then go overboard and take what good things the government does and want to turn it into a license to go all "socialist" on us. Particularly, when it come to big government liberal spending programs and tax policy. But hey, that's your right even if you look very foolish to the rest of us sane people.
studioxps writes:
Thank you for your comment
jsternbe#298183 writes:
Definitely, but it will take some work to get used to programming this new type of machine. My experience is that efficiently programming GPU computers is as different to the standard MPI programming that is done on modern supercomputers as MPI programming is to standard serial programming. Reducing the overhead of transferring data across nodes is important but somewhat ameliorated by using InfiniBand connections, but so far all of the GPU computers I know of still connect to the main CPU by the PCI-E bus, which is comparatively *very* slow. Hopefully the new supercomputer will have a better way of connecting these GPUs. The other difference is that each core on a GPU has very little memory so it must be used extremely efficiently.
WilliamDawes writes:
I'm just an IE, more accustomed to programming on PLC's than supercomputers. But I believe the NVIDIA Tesla GPU is designed to transfer data to the pinned memory system utilized by the Mellanox Infiniband.
LongShot writes:
I got one question. Can the new computer beat "Watson" on Jeopardy? If so the $ 100 million dollars is a good investment.
dweezil13 writes:
No,but these cats will devastate you in World Of Warcraft.
SmokiesWoman writes:
Thanks to all of you that know about the super computing world and have shared it today. My own knowledge wouldn't even be equal to a three year old in a pre-school nursery compared to one with a PhD in computer science. I do know that anytime China gets ahead of us in anything is not good and it's time for our side to get busy regaining our position.
With that said, there will be enough money wasted to possibly pay an hour's interest on the national debt. Maybe part or all of that stimulus money (it's for the roads and the children, you know. Maybe it can be redirected before the roads and the children figure that we got the shaft. Again.)that our illustrious Congressmen got sent to ORNL could be put in a piggy bank and pulled out to help fund the Titan. Methinks that an oversight department selected from we the people might be a huge asset to the entire project. We do have to live within our means and we are the only ones who get that one fact.
sphynxgirl writes:
I apologize for missing the articles on the benefits of these super computers in last week's news...my mother passed away and I was doing everything but reading the news. I think I had a good question but the responses were so unkind and cruel that I promise not to seek out answers on this forum again in the future. Thankfully I work with some awesome engineers who would have gladly answered this question for me if I had been at work...I will stick to the 'real world' where I know I will be treated with dignity and respect. Profut was the only exception here today.....thank you so much for your answers with judging me too harshly for my question.
profut writes:
Excellent point. One could argue that programming these systems is the biggest challenge we face. You're correct that using CUDA or OpenCL is different than MPI programming. More than likely we'll have to find a hybrid approach to exploit node-level resources as well as scaling to tens or hundreds of thousands of nodes. This shift impacts programming across the board though.
For large systems, don't forget to toss fault tolerance and power/energy usage as big issues to address too!
The good news is that my students have plenty of challenging problems to investigate!
profut writes:
My condolences about your mother - that's a terrible blow.
Speaking for myself, I welcome any questions that are respectful. It's far too easy to post a harsh response (even if not intended). That's one of the social aspects of computing we still haven't managed to get right...
CelticsGirl writes:
When you give attitude, expect to get attitude back.
madmaxvol writes:
I agree that your question was a good one, but whenever you ask a question followed by 3 or 4 question marks, it comes across as being sarcastic and condescending. That is probably why you recieved curt responses to your questions. Apparently, netmanners agrees with me (see the attached link).
So, it is my belief that those who were responding felt they were matching the condescention in your original post.
http://www.netmanners.com/email-etiqu...
P.S. My condolences for your loss.
cjensen writes:
Pearl_drummer_37667 writes:
Pretty soon ORNL will need a dedicated nuclear power plant just to run their "super-computers" Why isn't good good enough? I understand that every once in a while you need to upgrade but this kind of upgrade is pretty expensive. Does ORNL bill it out so the taxpayer can recover the investment? Or is it just another jewel around the Queen's neck?
------------------
Dang glad you weren't incharge of our national technology policy, or we'd all still be pounding on Trash 80s.
profut writes:
Your point about the nuclear power plant is apt: power/energy usage for next generation machines is extremely important. The computing field is trying to figure out how to gain orders of magnitude of power efficiency improvements to make exascale machines possible and practical.
Anyone trying to push the limits of technology will need to invest time and money. One good GPU card can provide more performance than a supercomputer from 20 years ago. That doesn't just happen - researchers trailblaze, with industry following behind.
With respect to costs, it's worth pointing out that the NSF supercomputer at ORNL (actually UT's machine Kraken) is the most cost effective computing resources on the teragrid (basically it's the best machine from a cost effectiveness perspective for the open science that NSF does). I don't think the costs for Jaguar are as well published (I suspect some customers wouldn't want their usage published), but I don't see any reason to expect that Jaguar isn't in the same neighborhood for efficiency.
I haven't run the numbers comparing the investment for the biggest machines from 20 years ago to today's machines, but I suspect they are similar with adjusted dollars.
I'd say that the folks at ORNL are as proud of all these machines as any jewels, but they are also quite a bit more useful and productive.
There are certainly some programs funded by the government that have been wasteful. I'm sure there are some aspects of gov't funded computing research that are wasteful. Nonetheless, I'd argue that our long-term investments in supercomputing have been very successful and have paid big dividends.
whizkidtn writes:
But what CJ really wants to know is can he play "Space Invaders" on the darn thing! Otherwise, to him it's just a waste of money.
cjensen writes:
whizkidtn writes:
in response to cjensen#613296:
Pearl_drummer_37667 writes:
Pretty soon ORNL will need a dedicated nuclear power plant just to run their "super-computers" Why isn't good good enough? I understand that every once in a while you need to upgrade but this kind of upgrade is pretty expensive. Does ORNL bill it out so the taxpayer can recover the investment? Or is it just another jewel around the Queen's neck?
------------------
Dang glad you weren't incharge of our national technology policy, or we'd all still be pounding on Trash 80s.
But what CJ really wants to know is can he play "Space Invaders" on the darn thing! Otherwise, to him it's just a waste of money.
-----------------
I modestly enjoyed Pong on the very first computers and later, pacman on bar computers, but since then, I'll have to admit to a complete lack of interest in computer games.
While games are mildly interesting or amusing at first, soon enough, one realizes there is no redeeming value, point or purpose to them over the long haul....much like watching Fox.
Share your thoughts
Comments are the sole responsibility of the person posting them. You agree not to post comments that are off topic, defamatory, obscene, abusive, threatening or an invasion of privacy. Violators may be banned. Click here for our full user agreement.