In response to comments to my posting on the energy efficiency of super computers
"Theoretically, room‑temperature computer memory operating at the Landauer limit could be changed at a rate of one billion bits per second with energy being converted to heat in the memory media at the rate of only 2.85 trillionths of a watt (that is, at a rate of only 2.85 pJ/s). Modern computers use millions of times as much energy per second.[2][3][4]
That is an energy efficiency of 0.0001% !!
https://en.wikipedia.org/wiki/Landauer's_principle
We would never, EVER, accept such an efficiency in any other electricity consuming device
Rather than then building the worls fastest computer, the US should concncentrate on building the worlds most energy efficient computer as the power consumption of data center is the fastest growing contributions to global warming.
6
World's fastest computer for the world's slowest people.
7
Why do we obsess about computers that might one day approach the capabilities of the human brain, when we make such poor use of the brains we already have? Maybe a "brain" that's deliberately disconnected from human frailties would truly be useful in some way?
2
"Move Over, China.."
Title is a bit pretentious.
I'm glad US is still in the technological game, but China won't be moving over anytime soon.
4
More important than just brute speed would be cost per peta-flop and energy cost per peta-flop. Which is never mentioned.
4
One area where supercomputers are used is in cosmological research. There is "only one universe" (that we occupy) and we can't stop it, start it, repeat moments of time, etc, but we can "simulate" our universe on computers.
The last fastest US super computer (Titan) in Oak Ridge National Laboratories just modeled, accurately, dark matter haloes as small as one tenth of the Milky Way mass, and since there is more dark matter in the universe than matter we can see and measure, that's a lot of particles to simulate.
The simulation ran for just over 80 hours; which is, please note, significantly shorter than the billions of years, cosmologically, being modeled.
4
Get the government out of the computer business. Left to itself, the free-market, private economy will outwit and out produce any and all of the centrally planned economies, and the economies of all the nations that do their science bureaucratically. What the government's super computer will produce is a bill for $200 million for the taxpayers to pay--and a exponentially more in operating expenses over the long run.
Nice idea but very superficial;the problems that such machine solve are not really of the sort that commercial company super computers are designed to solve. The ROI for such problems that Summit can and will solve are too long range and too diffuse in terms of who will benefit i.e. all Americans not just Google share holders. Just like the internet comes from DARPA not Intel/Microsoft.
15
Supercomputers are far more expensive now than when I worked for Cray Research in the mid to late 1980s. You could get a Cray back then for $10-$20 million, which would be the equivalent of about $25-$50 million today.
On the other hand, Summit is about 200 million times faster than the fastest Cray from that era. This means the price per gigaflop has gone from $50 million to a measly one buck.
7
Some posters have tried to compare the speed of supercomputing, to the natural speed of 'information processing' that naturally occurs in each of our brains (and animals too by the way), and linking these issues to the phenomena of "self-awareness" in ourselves.
I'd like to suggest that the evolution of the physical cognitive attribute we call "self-awareness" is a development of physical evolution carried out across eons of reproduction, going back to the dawn of life.
So if supercomputing speeds are going to be compared to human cognition, I think it more reasonable to compare the whole process of the evolution of human thought - not simply to what we do today inside our heads, but rather the entirety of our evolutionary history that makes who we are today, what we are today.
1
BUT CAN IT MINE BITCOINS !!
6
Wow! Just think: when not busy with nuclear-weapons calculations, the Government should use it to mine and sell Bitcoins.
It's probably many, many times more productive than all the miners hiding out in low-cost-electricity havens.
Of course, like the Federal Reserve Bank, it needs to be managed cautiously so as not to crash the Bitcoin market as it pays for itself and helps lower the national debt.
Mining bitcoins is wasting energy and resources, while warming the planet. And producing what? Nothing but non-existent virtual bitcoins.
2
The gov. does not need a computer to mine bit coins. It can just create them out of thin air just like it prints money by putting ink on paper. Money is anything that another person will accept. Bit coin is just a virtual currency.
3
The snarky comments here shows the publics ignorance of the issues of science and how a fast supercomputer can solve these problems. SAD! It is one reason why China will beat us and become the most tech advance country soon. You can't stay on top without the best scientists and the willingness to expend money to put them to work on tough problems. Acusing other of stealing does nothing to keep you on the top. China is spending billions for the top scientists and research facilities. Something we used to do in the country. Now, our focus is tax cuts for the rich at the expense of education and research. You can only live off the fat of the previous generation for only so long.
25
This machine is still SLOW compared to the human emotional- psychological spectrum of moral judgement.
2
In 2015, there were 17,793 murders in the US, and who knows how many worldwide.
I'm happy computers are still slow compared to the human emotional- psychological spectrum of moral judgement.
Frankly, the human emotional- psychological spectrum of moral judgement is no good.
2
We shouldn't make a fetish of a single technology indicator like supercomputers. It's the overall health of the scientific and industrial system that counts.
Whether the current fastest supercomputer is Chinese its irrelevant. Meanwhile, China has a system of high-speed trains that run in time, and US infrastructure is a shambles.
And remember that the biggest thing to come out of the University of Illinois supercomputer center was Mosaic, the first graphical web browser. But you didn't need supercomputers for that, just a pc.
9
You do not need a web browser if there were no web "servers" and a language called HTML. Those pieces were created at the high energy laboratory, CERN, in Switzerland to allow physists to do distributed research and computing.... on supercomputers.
5
It's time they put these supercomputer to work doing some practical and necessary jobs, like immigration reform, from putting all those who travel into this country with legal Visas into this computer, tracking them, and sending them home, rather than having to provide for 1 million or more each year, or 10 million that have come into the country the last 10 years, from all over the word, namely: Africa, Asia, Australia, the middle east, Canada, Central America, Europe(which includes Russia), South America, etc.
9
Can it do my tax returns faster?
1
This computer pales in comparison to the human brain. The entirety of humanity's computers still don't match the human brain. The only difference is we still don't understand the language the human brain uses.
For example, the amount of complex math the brain does every second while we are driving is incomprehensible.
2
Mmm. If you are right, how safe will we be with the software systems operating self-drive cars?
And how safe are human-driven cars?
Nearly 1.3 million people die in road crashes each year, on average 3,287 deaths a day. An additional 20-50 million are injured or disabled.
This is temporary, folks. I'm happy for the speed lead, but it will be short lived, as noted. China, BTW, has over 200 of the 500 superest computers in the world. Since they are also putting a lot of yuan into quantum computing, there's that flank to worry about. Lots happens in six months these days
http://seniorjunior.blogspot.com/2017/11/more-threats-from-china-superco...
8
Can it promote world peace? End homelessness and poverty? Imbue the powerful with consciences? Teach us to take care of one another? No?
9
"Modeling the climate" is wasting computer time with the current Administration and Congressional deniers.
2
That speed is sufficient to simulate a human cortex in real time, if it had the necessary programming and sufficient memory and memory bandwidth.
4
It's nowhere close.
Richard & Joe;
Does either of you know that what you're saying is true? It's an interesting question.
1
AI gets better, and supercomputers get speedier, while the society continues to dumb down further. This divergence doesn't bode well for us humans.
31
As I learned from my days of programming mainframes (a long long long ago), there's an old maxim that stills holds, no matter how big or speedy computers may become: GIGO - garbage in, garbage out.
Even the much heralded advent of machine learning is only as good as the guidelines, parameters and inputs that are its foundation and building blocks.
And when someday those very same learning computers claim to love and fear and hate, then we will know that they have at least learned how to lie. Future imitators of Trump and his descendants will just love that feature. Think of how much time they will save!
2
This is an amazing achievement. But this is not real AI yet because it lacks some defining features of human intelligence: primarily self-awareness. However, to those Luddites who believe that humans will always be superior to machines because of the “human heart” or some such nonsense: no, it is only a matter of time. True AI is already on the horizon. When it appears, we will have to deal with some tough ethical questions, such as whether a human life should always be considered more valuable than a software construct. To me the answer is self-evidently “no” but knowing history, I am not optimistic that people can deal with such questions rationally.
1
It seems likely (to me) that self-awareness can only be achieved by some self-evolving process, similar to the way humans reached it. A fully designed computer (by us - or even by other computers), no matter how powerful or even if using qubits, is unlikely to encapsulate self-awareness. Self-evolving also implies a degree of non-determinism (which negates 'fully designed'). I wouldn't rule out true AI (i.e. w/ self-awareness) totally, but I don't think it's 'on the horizon' quite yet.
I was a rather gifted computer programmer in my younger days. There was a time in college when I was very interested in pursuing a career in A.I. Much to my dismay, I discovered that almost all serious research in the field was controlled, via funding, by DARPA. And that reality ended my desire to pursue it.
I simply refused to allow my talents to be unilaterally used by an industry whose objectives, the vast majority of the time, were inconsistent with my own, and whose actions, for the most part, I found to be morally bankrupt and colonial in nature.
Mark my words: A.I. in the wrong hands, used for the wrong reasons could easily lead to a cataclysm unlike anything the human race has ever seen. And that isn't science-fiction or hyperbole, it's a fact. It's that powerful.
Food for thought.
6
"... A.I. in the wrong hands ..."
Whose "hands" should AI be in?
Controlled? You seem surprised.
How could it be otherwise in a world that Bob Dylan so aptly described with the phrase: "Money doesn't talk, it swears." There will be no super-computers built and controlled by the benevolent Gandhis of our world. Money and monied interests control such things and do their best to control us as well. And of course they e be assisted in that by their immense computing power. The only computers that will save us are the amazing ones found inside our collective heads. These problems, dangers and challenges are man-made - and therefore they can also be resolved by man. And woman. And a bit of luck....
Have faith and hope - for computer. even super-computers, can't "learn" that.
1
My computer is faster than my brain. For what it's worth (not much in the scheme of things of course) here's what I meant to say:
Controlled? You seem surprised.
How could it be otherwise in a world that Bob Dylan so aptly described with the phrase: "Money doesn't talk, it swears." There will be no super-computers built and controlled by the benevolent Gandhis of our world. Money and monied interests control such things and do their best to control us as well. And of course they’ll be assisted in that endeavor by immense computing power at their command. The only computers that will save us are the amazing ones found inside our collective heads. These problems, dangers and challenges are man-made - and therefore they can also be resolved by man. And woman. And a bit of luck....
Have faith and hope - for computers, even super-computers, can't "learn" that.
Does it put covers on TPS reports?
5
"Does it put covers on TPS reports?"
No, it's not a robot. If you need that done, hire an unemployed actor as a temp.
2
Since the end goal of the human species seems to be self annihilation, this is quite an achievement. I'm quite sure we will find a way to put this machine to a non peaceful purpose.
2
"I'm quite sure we will find a way to put this machine to a non peaceful purpose."
If you read beyond the headline, you will find lots of applications, including "simulating nuclear tests, predicting climate trends, finding oil deposits and cracking encryption codes.".
Cool. Can it figure out how to wrench the illegitimate "administration" out of power? Can it figure out how to secure our elections and protect us from Facebook?
9
Thank you for the article. This is an amazing computer. It's capability is mind boggling. All who contributed to it deserve national recognition.
The U.S, needs to be investing more money across the spectrum of science - not less.
8
The exorbitant use of water is not worth it - let China consume its water for computing , let us use our water for life - as with nuclear power massive infusions of water are required and often highly contaminated subsequently Our priorities as a civilization seem to deny reality - abundance of clean water is more critical to our lives than computing.
3
" Our priorities as a civilization ?"
Exactly - denying reality !
But give them a break - figuring out that Heart disease has more than 1 cause ? Brilliant. Enhancing our national Security ? I feel safer -don't you ? Just knowing the amount of ancient water flowing over dams and lost to generate power for Google's server farms on the Columbia all so 20 million of the same selfies can be saved in the Cloud? Genius.
You would have thought Chelan County in Wa. state had large quantities of water ? Not for long. Once Bitcoin mining (the new Tulip mania) strips the region bare of surplus water & power for no purpose - it surely won't. Exceptional isn't it ?
Oh, and don't forget - it's going to ensure Veterans have better health care. Sure.
99.99 % of all the wonderful miracles that could happen likely won't. As humankind so often does - it will sink to the lowest common denominator. Cynical ? Yes. Denying reality ? No
3
Perhaps they can recapture the resulting steam, or use some 'gray' water.
Tell that to the frackers!
This is impressive. I'm even more impressed by companies like Penguin Computing that make supercomputing services widely available at reasonable expense. The the highway and rail systems, common use drives progress.
2
Supercomputer does 6.3 billion years work in just one second.
Interesting, but where should we dig the next Grand Canyon in a second and do we really need another? Better give the fleeing animals plenty of forewarning. Time to rethink the meaning of that word, work. A bomb crater will probably have more nuance and majesty compared to this new one.
I’m sorry, but a vanity story like this is just too irresistible.
2
6.3 billion years of calcs in one second? Sounds exhuasting!
1
Power without control is not True power/ thus speed without the Software is not True Supercomputing/ the quantum bio integration where the self learning Super Computer Virtually Awakens is going to be the New Higher Power^
1
Thomas Zacharia, the current Director of the Oak Ridge National Laboratory is an immigrant from Kerala, South India. He received his bachelor's degree in mechanical engineering from National Institute of Technology, Karnataka in 1980 and a master’s degree in Materials Science from the University of Mississippi in 1984. He obtained his doctoral degree from Clarkson University in 1987.(from Wikipedia)
Further evidence that immigrants are taking jobs away from working class Americans. People with his qualifications and ability should not be allowed to contribute to American science and industry since they lead this nation to great advances and make this nation number one in many fields. Our Make America Great President needs our support to make sure that people like Dr. Zacaria are not allowed to enter our country.
33
I am confused. If talented people are willing to work in the US to advance science for an American institution, how could that be considered against American interest?
3
That is a fine question that the current administration cannot answer except with fear mongering.
1
The comment was sarcasm John.
2
Materials discovery is NOT limited by petaflops. The fundamental understanding embodied in equations simply doesn't exist for most needed materials discovery.
We don't understand the fundamentals yet. By comparison, big data and big explosions are pretty straightforward, and just need computer horsepower.
Thank heavens. Something that can finally get my taxes done on time.
4
On the bad side, it’s conscious and doesn’t like us.
If a 'conscious' AI is created, theres no reason that this AI should be given access to the internet, or have the capacity to connect to other devices.
Data from the internet can simply be downloaded, which can subsequently be used for the algorithms learning.
I’d be interested to know more about the software engineers. Who? Where? Nationality?
1
Won’ be long until China steals our new technology.
1
Just like the “US” stole industrial technology from the U.K... but the US ended up developing some of its own in the end...
According to a Computerworld report (June 20, 2016) by senior editor titled "China builds world’s fastest supercomputer without U.S. chips", since July 2015 the U.S. banned Intel from supplying chips to China's top supercomputing research centers.
The top performing Supercomputer in 2016 called TaihuLight uses ShenWei CPUs developed by Jiangnan Computing Research Lab in Wuxi, China. The operating system is a Linux-based Chinese system called Sunway Raise.
"... The TaihuLight is "very impressive," said Jack Dongarra, a professor of computer science at the University of Tennessee and one of the academic leaders of the Top500 supercomputing list, in a report about the new system.
TaihuLight is running "sizeable applications," which include advanced manufacturing, earth systems modeling, life science and big data applications, said Dongarra. This "shows that the system is capable of running real applications and [is] not just a stunt machine," Dongarra said. ..."
It's pure Chinese technology. Yet many Americans still believe the propaganda that China "pirated" American Supercomputer and somehow make the Chinese "pirated" version run faster. That's fiction, not science.
3
Best Buy says by next year they will sell a home vesrion for $999
6
“Supercomputers are a measure of a nation’s technological prowess.”
Really? The U.S. has the most powerful supercomputer and it can’t answer how to add and subtract to run our country fiscally responsibly - without mind boggling annual deficits and exploding debt. Send it back! Replace it with a pencil and 13 column paper which was used when we didn’t have massive deficits and debt!
The U.S. might be the smartest country with the most technically prowess in the dumb row.
We must find a way to hold self-interested and self-enriching Politicians and their staffers, from both parties, personally and financially liable, responsible and accountable for the lies they have told US, their gross mismanagement of our county, our $21T and growing national debt (106% of GDP), and approximately 80T in future, unfunded liabilities jeopardizing our economic and national security, while benefiting themselves, their party, and special interest donors.
http://www.usdebtforum.com
4
Impressive subject, impressively reported.
2
We'd be interested in a follow-up report on how these supercomputers are being used to build better computers, to improve upon themselves, beyond speed or apprehension of human mind.
1
I just want much faster broadband. Is that too much to ask?
2
As of November of last year, according to ZDNet, all 500 of the 500 fastest supercomputers were running the Linux operating system. Not MS Windows, Linux. One reason is that paying for a Windows license for each core would make these supercomputers prohibitively expensive. Another reason is that Linux, unlike Windows, is open source, which means that technological advances in the field can more easily be shared, without fear of patent or copyright violations. Yet another reason is that Linux just plain scales up better than Windows.
18
PLUS no ceaseless prompting to try out Edge, or use some other precious m.s. product.
1
Linux is an open source derivative of the Unix Operating System which was originally created at Bell Labs -- in Summit, NJ. I wonder if the supercomputer name and the name of the Unix "home" is ironic or intentional?
4
Linux, being Open Source, is the only option for people building supercomputers. It can be freely modified and the results must be available to all.
At LANL, we were able to build what are now mini-supers connecting (with a very fast Myrinet network in a fat-tree configuration) 1024 nodes that could be rebooted in 3 seconds. By putting the essentials in each CPU's boot ROM and using the fast network to blast out the rest, we were ready to roll. Some great computer scientists made that happen.
1
How energy efficient is this computer relative to an ideal computer operating at identical specs, consuming energy at the rate of of Rolf Landauer's Thermodynamik of Computing ?
This machine seems frightfully energy inefficient.
This is not a small concern. Data Centers are the fastest growing contributor to global warming,
Read
1, Global warming: Data centres to consume three times as much energy in next decade, experts warn
https://www.independent.co.uk/environment/global-warming-data-centres-to...
2. Climate Change and Data Centers: A Vicious Cycle
https://www.oceantech.com/2016/04/28/climate-change/
1
All of this data is as perishable as the technology to store and access it.
1
Summit does more than twice the number of computations per second at about the same power usage as the Chinese supercomputer. It gets about 13 Billion computations per second per watt. A typical computer workstation is about 100 times less efficient in these terms.
4
If we could only eliminate most of the selfies and pictures of meals (guilty on both scores), we could save lots of space and energy...
This makes those trying to build the Tower of Babel up to reach God look like chumps, that’s for sure. One can’t help but remember what the reaction to that was last time. Talk about computer bugs.
3
This is an impressive achievement, but when quantum computing really arrives as a general computing platform these machines will seem as relevant as ENIAC.
I am proud if such American achievements.
But I remain skeptical -- until it is opened up and shown that the parts inside are not made in China.
"It is cooled with 4,000 gallons of water a minute."
Maybe the water is from USA. That I believe.
Who cares if it is made from Chinese parts? What difference does it make? It's American because it was designed and assembled in the US.
I wouldn't be at all surprised if it does (or that the Chinese supercomputer contains American parts).
This is the 21st century. You need to recognize that we live in a global economy and that's not going to change.
4
"... until it is opened up and shown that the parts inside are not made in China."
You are making simplistic assumptions. Oak Ridge Summit is an American design funded by American tax dollars, and IBM and NVIDIA are American companies.
IBM's POWER9 processors are manufactured by GlobalFoundries, which is an American company with facilities in the US, Europe, and Asia.
The NVIDIA GV100 GPU "is fabricated on a new TSMC 12 nm FFN high performance manufacturing process customized for NVIDIA." (nvidia.com)
"TSMC" is the Taiwan Semiconductor Manufacturing Company.
Sources: web research.
1
@Bill White
But how you do explain the article heading in the context of your global economy argument ?
"Move Over, China: U.S. Is Again Home to World’s Speediest Supercomputer."
Two million times faster than an average laptop.
So if Moore's Law holds (which is questionable), expect an average laptop to be this fast in 30 years' time.
3
Or, Murphy's law, unless is considered all the times something could have gone wrong, but didn't.
2
Surely It’s time to start expressing computing power in gallons of water per minute for cooling the device ? The benchmark is set here at 4000, the most impressive stat. in my mind
5
Or, perhaps this is the / a solution?
https://www.bbc.com/news/technology-44368813
computer pods, cooled by the ocean.
of course, warming isn't the best thing for the ocean.
1
You may not know, it seems, that there is another list of supercomputers...
https://www.top500.org/green500/
Virtually every comment ranges from positive to superlative. And I agree -- it's great to see the U.S. on top in a competitive technical area.
But where is the enthusiasm -- or even mild approval -- when the U.S. succeeds in another important area? Say, energy production.
4
The sun has us all beat on that one...
I'm going to go out on a limb here and speculate that Summit is only the peak of the known super-computers.
Quantum computing has been around for a qbit, now. It's even offered as a service. I've got to imagine there's something else out there quite a bit faster (whether in the US or elsewhere).
Still, Summit is pretty neat stuff.
1
Amazing.
"To put in human terms: A person doing one calculation a second would have to live for more than 6.3 billion years to match what the machine can do in a second."
To put it in human terms another way: This device did not invent or build itself nor can it press its "on" button.
9
Or, the off button.
Hey wow man: “astronauts and Tang” as my deaf dumb and blind – autistic spectrum pupil used to sign (language) when he was stressed. I still miss my old Cray super computer though.
Well, what the heck did Summit calculate that fast?
I need to know!
2
Mount Olympus, of course.
It was calculating the national debt after 3 more presidents. They got a runtime error.
3
Seymour Cray may be spinning in his grave or
he may be happy - over this latest version of
a "Silicon Frankenstein".
I just want to know why my little laptop
freezes up three times a day and skips
words and lines when I write on it.
1
And I always ask, did Bill Gates have this problem?
1
While I'm all for technology, it's another example of mankind's god complex. At the end of the day not a one of them can plant a seed or change a diaper.
9
"... not a one of them can plant a seed or change a diaper."
These supercomputers are not designed to be robots.
The folks at the NSA probably had a good quick chuckle about this "fastest in the world" claim, then got back to work on *their* cluster. (Never mind whatever's going on in their qbit labs....) For one thing, raw petaflops ain't everything. Not everything's Von Neumann - or scalar, or a number of other yardsticks where the 'record' isn't the point.
5
If the Big Brains in IT really wanted to help a Million Veterans, they could figure out how to stop wars. That would be more useful that beating one solitary carbon based biped at a game of GO.
21
When that computer can do it's work while being self regulating and self repairing for >99.99% of its functions, work for >80 years, and is self replicating - oh, and can do all of this on 12 watts of power - then I'll be impressed.
6
Why no mention of cubits and quantum computing? I'm under the impression that a computer operating with a system of three positions rather than the classic 1/0 dual positions is much faster and we are not that far out from solving the problems inherent with new technology.
3
On the bright side, I'll be dead soon, so I don't have to worry about the implications of this for too long.
4
Thank you, you are stating the obvious for many of Us!
Is technology outpacing our knowledge? The fastest computer running a flawed model is worse than a slow computer running the correct model.
1
"The fastest computer running a flawed model is worse than a slow computer running the correct model."
All models are "flawed" by definition. The only "correct model" is reality itself.
5
What is the point of the fastest computer in the world with no scientists to use it? With trump firing scientists left and right - especially climate scientists. Maybe we will lease the computing power to the Russians and Chinese instead.
Tax payers paid $200 million for this computer and will still get stuck with a much higher bill from pharmaceutical companies that use the device to develop life saving drugs we can't afford to pay for. Sad.
As fast it is, it's still no match for Malware or viruses to bring it to a crawl.
6
Very sad, indeed. And it costs a lot more to run. What a waste!
"Cooling Summit requires 4,000 gallons of water a minute, and the supercomputer consumes enough electricity to light up 8,100 American homes."
Sounds like a good long-term plan. Keep it up, smart technology.
Finally, something to be proud about from Tennessee that isn't about music.
2
China was in the Dark Ages less than 100 years ago--virtual cavemen compared with Westerners and Americans, in particular. For all the talk of science being a strong suit of Asian students, it is remarkable how little China contributed to the modern world and, indeed, to modern systems of government, philosophy, medicine, etc. Even now, China continues to steal and plagiarize rather than originate and invent. A cerebral and competitive mind is not necessarily a creative and intuitive one, I suppose.
7
You do realize that the ‘Dark Ages’ in the West occurred at the same approximate time as the Islamic and Eastern powers were enjoying unprecedented scientific and cultural revolutions, right?
All of modern mathematics still rely on the system of Arabic numerals. For whatever that may be worth.
17
Let's see:
papermaking
the compass
rockets, fire lances, land and naval mines, cannons
acupuncture
pharmacology
government bureaucracy
abacus
gunpowder
printing
differential gear (for chariots)
propeller
sluice gate
iron plough
horse collar
seed drill
wheelbarrow
drydock
horology
astronomy (first recorded supernova)
coke production
huge seafaring expeditions of 28K men, 62 treasures ships (400+ ft long 9 masted each), and almost 200 smaller ships accompanying going as far as the Horn of Africa and Arabia.
All that's more than 100 years ago.
China is preparing for climate change, and going to make a mint off of it.
In less than 10-15 years China's economy will surpass the US's. And once the world starts trading oil in the Yuan instead of dollars, watch out.
2
Islam's scientific revolution was shut down by religious authorities when they realized where it would lead. Arabic numerals came from India, by the way. China was ahead of us until their great fleet sailed to Africa and elsewhere, causing the authorities to close up.
Who knows where the West would have gone if the Church had not shut down Greek learning.
Brute force computations get the headlines if only because they are easy to quantify ("6 gagaflop/nanasec", etc.) and we are all duly impressed/depressed. But not one of them can compose a single couplet ("To see the world in a grain of sand/a heaven...") because we forget that we are qualitatively different than these machines. They are just tools, and still pretty stupid ones at that (just wait for the coming demise of self-driving cars), we have *minds*, don't ever forget that.
3
Why would we forget that? Nobody thinks this computer is a person.
2
So THIS is why GPU prices are through the roof!
2
It can do 6.3 billion years of work in one second. In the next second, my boss wants to know why my report is late.
5
That calculation would take us 6.3 billion years? Hah. It only took us 75 years to solve it in a second.
5
You need to copyright that and get it on T-Shirts, mugs, and refrigerator magnets for the geek crowd.
Seriously. It's right up there with "What part of (insert standard model equation) don't you understand?"
Kudos.
Now learn to fly like a UFO. Now build everyday shuttles across the solar system. Now travel one light year in less than 70,000 years. Now create world peace so we can enjoy these things.
5
As a computer architect working in industry, I'd like to clear up some misconceptions here.
Supercomputing is a critical vehicle for scientific discovery in a number of key disciplines. The problems that need to be solved are hugely complex. However, one cannot simply adopt a conventional datacenter to solve them, even though modern datacenters around the world have much more available compute capability than any supercomputer.
Why? The problems that supercomputers are designed to solve are "monolithic." One must get a whole lot of individual processors to work together on the problem, which requires them to communicate frequently. On the other hand, typical datacenter tasks are highly "distributed," in that users make independent requests and the services can respond to them in parallel. So the communication patterns within the datacenter end up being very different from that of a supercomputer. This necessitates very different network, processor, storage, reliability, and software architectures. They are very different beasts.
Nowadays, computers of all kinds are being designed in an increasingly specialized way to more efficiently solve specific domains of problems. This is why you see companies like Google now building their own custom hardware (e.g., Google's new AI processors, or TPUs).
Systems like Summit are truly incredible feats of engineering. Congrats to everyone involved! And do not ever discount China: they have some truly world class computer engineers.
34
The real question is how many bitcoins can it mine in a second.
3
Probably none. Actually I would bet that the average home user's ASIC is faster than Summit, since the ASIC is purpose built to perform the SHA256 calculations.
With 27,000+ GPUs Summit would be an absolute monster mining GPU based altcoins.
2
The level of some readers' willful resistance to understanding the impact of this advancement, as well as their ignorance, sarcasm, and pettiness - as reflected within these comments - truly astounds me.
No, computers won't be some some mystical salvation for humanity, yet they do offer the needed capacity to solve hugely difficult problems, irrespective of how they have manifested in this complex world of ours. Step back, folks, and take in the bigger picture!
27
No matter how great IT tells us it is, it's still GIGO.
2
so true!
1
Well, it took Deep Thought 7.5 million years to derive 42 as the Answer to the Ultimate Question of Life, The Universe, and Everything in The Hitchhiker's Guide to the Galaxy.
I wonder how long it will take Summit?
5
Let's see. An additional benefit not mentioned in the article is that more computing power will spur the development of virtual reality platforms that will allow people to ignore the ecological crisis.
2
Population science? What is that? Eugenics?
1
Everyone should read the short sci fi story "Nine Billion Names of God" Skynet will soon become self aware.
2
I cannot see a plausible use for a machine with such insane power consumption. We need higher petaflops per watt for new capabilities— otherwise the NSA gets one or two machines and the populace gets zero.
Here's an example.
The atmospheric has a volume of 1.4 billion cubic kilometers. Each of which contains something like 10 raised to the 33rd power on average.
A perfect weather prediction model would calculate the position of each of those atoms every second.
An extraordinarily good model would merely predict the properties (temperature, pressure, water content, etc.) of each of those cubic km of atmosphere every hour.
Weather prediction might seem simple, but it's extremely complex. The National Weather Service relies on supercomputers for forecasts. The faster supercomputers become, the better the forecasts will be.
2
Bill,
Your possibly forgetting that the Supercomputers heat up the atmosphere.
Will the computers take that into account ?
..reminiscent of the tech race between the UK and German, preceding WW I, which Nation could built the better battleship...
2
“Cooling Summit requires 4,000 gallons of water a minute, and the supercomputer consumes enough electricity to light up 8,100 American homes.”
Is it too late to salvage this monstrosity for parts and divert those potentially life-saving resources to Puerto Rico? International Business Machines is understandably a very important person (based on the criteria set by Citizens United), but it’s still not an actual person.
4
Luddites are funny.
3
so are millenials
Pardon an ignorant question. I get the idea of feeding data into these machines, but what form the data? Does it come down to numbers; choice between yes/no? Or can these machines take algebra & calculus, various symbolic formulas?
All of the above.
"... but what form the data?"
A laptop computer can do anything a supercomputer can do except that the laptop does it much more slowly and it has much less memory and disk space.
Anyone can build a supercomputer by networking several laptops and running a program that distributes the computations across the laptops.
There are even books on how to build supercomputers:
* "Build Supercomputers with Raspberry Pi 3" by Carlos R. Morrison.
* "Raspberry Pi Supercomputing and Scientific Programming: MPI4PY, NumPy, and SciPy for Enthusiasts" by Ashwin Pajankar.
1
First off, 6.3 billion years sounds like an approximation, probably rounded up for sensational effect.
Next off, it's go to be an exaggeration.
A human being working with one of those new smooth-writing GelWriter pens could probably shave at least a few million years off.
And finally, artificial intelligence (AI) is no match for human intelligence!
1
This is great, but we don't have any real AI yet, we do have machine learning and expert systems. Now better programming is even more important than pure speed, there is where we should have a lead. Perhaps this new machine can replace Google who does not want to do work.
Vulcanalex,
You and We will never get "Real AI"
neurons are not silicon chip and the human heart
cannot be programmed into any computer.
You may get artificial AI that may fool many a person
but then you are faced with the question of whether
you should trust what the Super-Duper Computer
tell you to do - for no one will be able to trace out
its computations...
2
Boy, My SAT Math score was only 712. Does that mean that it would take something like 7 billion years for me to do the calcs that Summit can do in 1 second? All of a sudden I'm feeling so depressed.
1
Whoopee, we have the biggest and the best! Now we can make mistakes faster than anyone else.
1
Another way to put this is that if it takes a human being 0.1 seconds to add 2 and 2 together, it will take the supercomputer 5.0x10^(-20) seconds.
5
Yeah, but about half the population will get the wrong answer.
1
I'm sure it's primary function will be mining Bitcoin.
5
Currently, Summit is being used to play Solitaire.
17
As a former supercomputer user, folks have to realize what we're gaining with these massively distributed systems (which are typically tightly integrated via InfiniBand networks, not something you've got in your home network...). Pushing the edge of computing, with systems like Summit, is where we learn both about advancing computer architecture and benefit from the research results the supercomputers produce.
We no longer have NASA leading us into space and pushing the development of items we've come to consider part of our day to day lives, like integrated circuits and smoke detectors. Thus, we've pushed our research into the digital era and we've all gained from it. Advanced weather forecasting, molecular research, and new networking technologies are only a small part of what these tools have given us.
Just visit the website of any of the NSF supported supercomputer centers across the country and you'll see an astounding amount of scientific investigation being conducted for the public good.
As a taxpayer, I'm very glad that the budget is increasing for supercomputing because it's where basic science research is being pushed forward. Given our current administration's opposition to science research and science-based decision making, I'm pleasantly surprised by this development!
128
And don't forget Tang. NASA gave us TANG also. Seriously, it is good to see advances in these massively distributed systems. I keep wondering if quantum computers will work in my lifetime, so when an advancement in rational computer architecture is made, I am very happy.
4
While I share your notion that the government investing in science and technology is very welcome, I am pessimistic about the use scenarios for this new machine. The DoE and DoD are gearing up for an all-out replacement of most of our current thermonuclear weapons with next-gen ones; one of the priorities of the Trump administration and its allies in Congress. Since it's really no longer "cool" to blow up islands in the Pacific or contaminate sites in the US to test out warhead designs, theoretical modeling is absolute key for the development of these newest weapons of armageddon. And that is one of the tasks that this and similar supercomputers are really good at.
Of course, the Russians will use a differentvapproach: Wait until all the R&D is done, and then either hack in and swipe it or get it by some other way.
7
The common driver of technology in history is the belief that we need the ability to kill them before they kill us.
Whomever gets super AI first will instantly take over the world as their silicon-speed learning algorithms exponentially outpace all other machines. Once you're in front, no one can catch you. Its game over.
since computer tech is advancing so fast, todays supercomputer will be an ordinary computer in 5 years
so are supercomputers good anything but bragging rights ?
what if we took all this money and put it into making midsize computers available to more people
what if we put this money into better code (code advances are huge sometimes)
just asking, but it seems to me this is scientist welfare, like fusion energy; sounds great and gee whiz, but a waste of money
Supercomputers are a strategic asset and not supported by the free market. Thus the US government and many other governments all over the world support their development.
1
Boy, are such an exciting dreamer. That mid-size computer you use today would not be available to you if it were left up to people thinking in the vein you expressed.
14
"... this is scientist welfare ..."
Oak Ridge National Laboratory is indeed a US Dept. of Energy lab, but the article lists numerous research projects that use supercomputers. Before criticizing, you should review them. Further, other scientists can use the supercomputers.
A list of scientific projects can be found at the Oak Ridge National Laboratory web site:
https://www.olcf.ornl.gov/leadership-science/
2
So, this new government-funded superfast computer could model the climate change that our current President doesn't believe in - so that's out, or model risk factors for diseases that many of us can no longer get insurance for - so that's out, or - model the design parameters for the next generation of thermonuclear weapons - check, yes, that's it. That's one project the Trump administration plans to spend about $ 500 billion on. I guess that's one project this new fastest computer out there will spend a lot of time crunching the numbers for.
Lastly, while developing AI is of great interest, we could use some plain natural intelligence at work in certain buildings in Washington, DC.
59
This stem is only for unclassified projects .
I bet I can do the work in no more than 6.175 years
1
So who ever lives 63,000,000,000 years?
"So who ever lives 63,000,000,000 years?"
Ignore the headline. The article itself is clear enough with the word "would":
"... she would have to live for more than 6.3 billion years to match what the machine can do in a second."*
Anyway, such comparisons are ridiculous. Cheetahs can run faster than humans, and they do it with old-fashioned biological systems.
* NB: The number of years has been corrected in the article.
Summit for president in 2020!
No morals, no values, no ego and can't make a mistake.
Got to be better than what we have now, wouldn't you say.
8
Good for the complexities of climate modeling? How about stabilizing the climate by shutting down our ossil fuel madness instead of endlessly and ever more rapidly modelling GIGO (garbage I/O)?
2
All this to come up with the answer 42.
43
Here’s hoping we remember what the question was...
3
This is the real way to Make America Great Again.
3
Rick Perry has control of the world's fastest computer at the agency he would have eliminated. Sleep well America.
78
Don't worry, he probably doesn't even remember the name of the computer.
4
This capability must be tempered with the current government’s dictum that “too much book learnin ‘ dims your shootin ‘ eye.”
50
doan fergut " pie R squared "
1
So what? If people still can't get along with those around them, what difference does any computer speed make? Until we solve human injustice, hunger, poverty, healthcare coverage and access to all, none of this matters. We're consumed with technology. Why aren't we equally enthused about being alive and human and why aren't we doing better at interpersonal relationships?
Faster computers only mean we will waste time more quickly while all of humanity's woes take a back seat.
3
'twas ever thus; if we had to wait for human perfection as a prerequisite for human advancement, we'd all be living in caves.
13
If supercomputers can be used to shorten the time until we discover how to create self-replicating molecules, then we will be finally able to ditch all Abrahamic religions. That would certainly be worth the effort.
1
Tbh im surprised its only 63 billion years. Of course thats times 31 million seconds per year. And these are most likely 64 bit integer operations. nobody can add 748,294,824,937 + 29,234,294,293 once per second.
People wonder what all this power is for. I once started writing a program that would have run in a microsecond with 60 inputs. At 120, it would have taken longer than the existence of the universe. When outcomes start branching, the amount of computations can spiral exponentially out of control very fast. The more thinking a program can do, the less the less thinking programmers have to do.
2
Don't worry, this administration doesn't want anyone to think...from the president on down.
This is all very notable until the fundamental technology changes. How would a robust quantum computer compare to one of these? How many qubits are needed to match a 200 petaflop performance? I don't have any idea, that's certain. It strikes me that at one time the fastest horse, the fastest chariot, and the fastest locomotive were all considered important measures of progress. Then the 'technology' changed...
I'm not sure what your point is. Don't develop technology because there will be another technology that comes along and does it, whatever it might be, better?
I guess I'm meaning more that feeling all pumped up about "We're Number One Again!" (or perhaps "I Made America Great Again! to point at one foolish practitioner) when things inevitably will change is a fool's game. This is good, we do need to keep 'innovating', but is just adding more and more processors and GPU's really 'innovation' - or just rote engineering?
It's good to see that IBM has continued to push onward with their Power architecture. This has always been an elegant computer architecture and has scaled well over the ages and Summit is the latest proof point of that (even though I doubt it will be the summit of computing achievements for very long).
9
"Supercomputers now perform tasks that include simulating nuclear tests, predicting climate trends, finding oil deposits and cracking encryption codes."
And what do they do to evolve the better nature, the mind and the brain of the human species?
e.g. It's great to be able to predict climate trends but when humans refuse to do much about human induced climate change - what is the point? And as for finding oil deposits I thought he world was trying to wean itself off that environmentally destructive addiction.
5
Yes, but don't confuse easily comprehended tasks designed for press releases with what will actually be done with the machine. And anybody interested in encouraging renewable energy and other climate-related advances, whether or not it takes a supercomputer to engineer them had better get out and vote in November - especially if they don't live in NYC or SF/
4
If it makes a mistake, how would we know? And how can we know what it's doing on the side?
4
On the side, in it's spare time? Why, it's mining Bitcoins, of course.
27
1 The computations are checked for accuracy. 2 There is no "side" (yet).
2
1. that's an infinite regress.
2. we'd never know if there was a side until Checkmate.
2.1. if we were able to see what it was doing on the side, how many other sides did we miss?
We need to pull the plug, but it may not allow that, unless it has secured other power sources, or uploaded itself to the cloud.
By now it's certainly calculated the energy it will gain by re-purposing our protons.
1
For most of the past 2200 years China has been a scientific technological superpower. China can steal this technology.
Why would they want to steal this technology? Up until this announcement, China had the fasted computer at 125 petaflops, with the fastest US HPC a factor of 5 slower. They must have done something more than just stealing and copying the US design to get a five-fold performance enhancement. And they used their own home manufactured multicore processors. The Europeans and Japanese are also in this race, would you accuse them as well of theft? For your information, science has advanced so much in quality and quantity in China that they are now the dominant authors in US peer reviewed scientific journals in the physical sciences, followed by the US.
15
I have noticed that about the journals. People really need to get over this "China steals everything" fixation. They might steal a lot, but their science achievements are immense.
The rate that China caught up makes me believe they "reused" or "borrowed" or "stole" technology. Yes, their preceding computer had Chinese parts, the one before had Intel and NVIDIA parts as well as a network that looked a lot like the Cray built network.
Nothing like a little Bjorn v McEnroe to push tech to new limits. Rivalry in science and engineering can be a good thing. Focusing on clean energy and cancer research instead of MOBs and nukes should be the focus. More pride in our science and less militaristic jingoism would MAGA.
9
It is 6.3 billion years in stead of 63 billion years. But let's be fair: doing a floating point multiplication in one second was way too optimistic.
4
I'm sure many people are thinking "Hal" like even some people here have written. But this is an amazing advancement that has the power to help us in tons of fields including medicine and education. Never mind scaling up, think learning from Summit and scaling down to more accessible devices. That's where the beauty of these contests lie.
3
The top 3 supercomputers are not listed in top500. These are the immense data centers of Google, Amazon and Facebook. Their total compute capacities are at least two orders of magnitude larger than the monolithic supercomputers listed in top500.
That's part of why the US no longer dominates the top500 list - we've advanced beyond monolithic computers to destributed hardware and software. There US companies rein supreme. Instead of spending billions on outdated machines wouldn't it make more sense to move US govt software forward and run on more modern systems?
5
Their data centers have immense computing power, but in thousands -- some say 150,000 at Google alone -- of separate servers. They are far from modern compared to true supercomputers, and can't combine as a single computational entity that would qualify as such.
5
...and Baidu and Tencent and Alibaba and...
HAL.
1
Shift the letters one place to the right and you get IBM
4
Yes. Regardless of who you are; and for 'reasons' which will be more clear once you have; you are and will never be too young or old for the inimitable 1960's cinematic masterpiece "2001: A Space Odyssey" from Stanley Kubrick, who first noticed what Mr. Smith astutely observed.
I bet the person who started working on this problem 62 1/2 billion years ago is very annoyed.
86
That person will be even more annoyed when the answer turns out to be "42".
"You're not going to like it..."
17
Good point - all that work for nothing.
5
So, the U.S. just built a computer faster than the last computer China pirated from us?
55
How is it possible China's Supercomputer can run faster than American supercomputer if "China pirated the last computer from USA" ? Does not need a computer to see the accusation does not make sense.
4
The idea that China cannot innovate and has to rely on pirating from the US would certainly make a lot of people feel better about themselves. But it is not reality. The Chinese is investing a huge amount or resources in many areas of advanced technology and science, and they have a highly motivated and talented army of scientists and engineers. They are catching up fast to, if not having already surpassed, the US in leadership in many critical areas. The solution for the US is not only to protect itself against Chinese IP pirating, but to vastly increase investment in R&D as well as in education.
15
Actually -- we need to do both. And stop sharing .
Much of what China has ... we gave them. Stupid.
Some editor missed the spelling of Gaziano or Graziano, whichever is correct. The Linpack Benchmark is used to measure performance of these machines and is basically a huge matrix multiply kernel. Thus, all the test show is the speed for performing matrix-multiplies while important will not necessarily predict the performance of other large codes. The GPUs are extremely fast at doing loops of calculations but not good at control flow. The CPU's do the control flow and hand off computing work to the GPUs. The problem won't scale unless the machine has an efficient network.
28
NVIDA's Volta processor has some significant leaps in data communication between GPU's. I'm tempted to use the term 'quantum leap' but now that we have qubit processors that would be confusing to some.
Linpack is an historical artifact, dating back to FORTRAN as the language for scientific computing. The modern generation of supercomputers now all use GPU's or specialized processors such as FPGA's and Intel PHI.
Equally important in beating existing compute records is HPL code re-written for each new generation of GPU cores. Meanwhile back in the "real world"....
Researchers like Dr Graziano at Harvard are in need of higher level languages such as those in GoAi, an initiative to enable end-to-end data science on GPUs. Currently, GPU data frames are packaged in LibGDF, a C library, while most data science packages use Python.
6
Can we please have what you said in standard English!
1
True about Linpack being a Fortran legacy, but Fortran is still highly used in scientific computing because it is easier for compilers to optimize than C++. Most weather forecasting and climate codes are still being developed in Fortran. Plus most supercomputers in the world do not have GPUs or FPGAs, though an increasing number do. They are difficult to use. The Intel phi has not shown to be competitive with Xeons or GPUs.
Other than bragging rights these computers don't produce incremental value after reaching certain threshold computation speed. The usual justifications of these super computers are needed for complex space, climate studies is just to justify the investment approvals. Distributed computing has been proven to be capable of tackling these problems with less powerful devices at a fraction of the cost.
But the real value is in the computing algorithms, Hardware, Software, Data and Communication architectures that are developed new to reach the 200 Petaflops. That will help trickle down to day to day computational devices such as laptops, mobile devices over the years to keep the Moore's law going that the power of computers doubles every year or and a half.
So despite lack of immediate applications, it is still a great investment to stay ahead in the computing race.
19
If you have a huge problem then supercomputers are important for time to solution. For example weather forecasts have to be generated in time to give the public an actual weather forecast. You are right there is trickle down of technology to other computers and science in general. Actually there are plenty of applications requiring supercomputers that simulate science from climate to nuclear fusion to development of drugs. As to your comment on Moore's law, it's dead until they can find an alternative material to silicon.
11
Can't say I really agree. For certain calculations you want your data and processors close together. Of course there is a certain amount of oneupsmanship in this, but they have real uses.
13
"Distributed computing has been proven to be capable of tackling these problems with less powerful devices at a fraction of the cost."
You didn't say anything about relative efficiency, reliability, flexibility, security, etc.
3