Slashdot Mirror
NSA, DOE Say China's Supercomputing Advances Put US At Risk (computerworld.com)
dcblogs quotes a report from Computerworld: Advanced computing experts at the National Security Agency and the Department of Energy are warning that China is "extremely likely" to take leadership in supercomputing as early as 2020, unless the U.S. acts quickly to increase spending. China's supercomputing advances are not only putting national security at risk, but also U.S. leadership in high-tech manufacturing. If China succeeds, it may "undermine profitable parts of the U.S. economy," according to a report titled U.S. Leadership in High Performance Computing by HPC technical experts at the NSA, the DOE, the National Science Foundation and other agencies. The report stems from a workshop held in September that was attended by 60 people, many scientists, 40 of whom work in government, with the balance representing industry and academia. "Meeting participants, especially those from industry, noted that it can be easy for Americans to draw the wrong conclusions about what HPC investments by China mean -- without considering China's motivations," the report states. "These participants stressed that their personal interactions with Chinese researchers and at supercomputing centers showed a mindset where computing is first and foremost a strategic capability for improving the country; for pulling a billion people out of poverty; for supporting companies that are looking to build better products, or bridges, or rail networks; for transitioning away from a role as a low-cost manufacturer for the world; for enabling the economy to move from 'Made in China' to 'Made by China.'"
Instead of billions on a stupid wall, invest billions in supercomputing tech. Hell, invest billions in semiconductor tech, cuz China is trying to take the lead in semiconductors big time.
Fucking Trump, trying to bring back manufacturing when he doesn't understand the concept of "robot".
copy paste
"Mr. President we must not allow a mineshaft gap!"
Alternatively, you could stop spying on everything everyone does, and use some of that money to cover your new toys.
Out here in the real world, we're about done writing blank checks for "national security" and "them terrists". No one would ever notice if you cut your mission in half.
Who is he talking about? Who's "us"?
Well, since it's the NSA and DOE saying this, I'd guess us=U.S.
If I can be modded down for being a troll, can I be modded up for being an orc, or a balrog?
China is taking a lot of very slow consumer chips and making a fast networked computer system.
Only the USA has the university graduates to design a real super computer with super computer ready CPU designs.
China is doing what contractors sold the NSA in the 1980's. A lot of desktop CPU's on a really fast network can do some super computer like work on a lot of data.
Great for the contractors selling a support service to the NSA.
An easy hardware solution for China to get fast results if the question can be worked on with a more limited set of maths on a networked system.
If the US wants to win the supercomputer race, just redefine what a real super computer is again.
A fast network of consumer cpu's is not a real supercomputer anymore.
With a new ranking standard the USA is number one again.
Domestic spying is now "Benign Information Gathering"
"We have to spend more money on this project or we will all die"
if your strength relies upon the weakness of others, then it isn't really strength. I know, "without light there is no dark" "without up there is no down" "without weakness there is no strength" - no. China is just 1 country, and has 6x the number of people. There are plenty of other countries that are in poor shape, if we (in the US) really feel we need to be "superior" to someone else. Until then, this kind of crap is a waste of tax dollars, under the thinly veiled cover of nationalism
US incompetence is portrayed as being a victim. As if China is a threat because the US has not done anything for decades. China deserves respect for getting their fingers out of their arses while the US is too busy wasting its money invading innocent counties. If we are to talk about risk discuss the real problem if we are talking about Chinese advances have the balls to show some respect.
I love stacking my barbecues in the shed at the end of summer - you can't beat a bit of grill on grill action.
I plan to snatch back the leadership back by assembling a few millions of raspberry pi. YEAH!
^(oo)^pig~
Since Donald Trump thinks China is the problem, claiming a lack of fund is going to let China overtake US seems a very good strategy to get funds.
China builds worldâ(TM)s fastest supercomputer without U.S. chips
Chinaâ(TM)s New Supercomputer Puts the US Even Further Behind
The next one is going to be 10 times as fast:
Tianhe-3 will be 10 times faster than the current fastest supercomputer in the world, the Sunway TaihuLight
The revolution for the NSA contractor was to offer the NSA desktop computer CPU's on a fast network.
The US collect it all policy resulted in data that needed to be sorted. A huge new network of desktop CPU's packaged to look like a classic super computer was a great idea to win a long term contract with.
The hardware was cheap, the new network support needed was expensive. Win, win for a private sector contractor.
The other emerging issue for the NSA is Australia, Canada and the GCHQ can buy the same methods and do not need NSA staff to help other 5 eye nations with bespoke US branded NSA export grade super computers.
So other nations are collecting all, breaking codes and using low cost computer networks regional that no longer need constant direct NSA oversight and support.
The other issue for the NSA is US universities are just giving every US CPU design away as its created by US graduates.
A lot of other nations university students get to study into the USA. Some hold over cold war State department policy to make the US look nice?
The US university system is addicted to the profits from such students? The CIA hopes to turn some of the students from Chain in the USA so that spy policy is worth the total loss of decades of US super computer plans?
Only the NSA can see the epic fail in giving decades once secret US CPU plans to any other nations students who attend a top US university.
Did the CIA actually turn a few students from China over the decades and the results from a few very well placed US spies in China that counter the NSA/GCHQ digital collection results?
The more public question of a super computer gap will restore a funding balance. Think of the Army and Navy code breaking of the 1930's. Two sides of the US gov/mil working on the same issue and not sharing.
The super computer gap news is just the public face of such issues.
Domestic spying is now "Benign Information Gathering"
All non-military, non-entitlement budget entries are going to be dramatically slashed in the current president's upcoming budget.
#DeleteChrome
That's often the case, but not always. In some applications you have a deadline -- or a deadline/precision trade off (e.g. weather prediction).
Post may contain irony: discontinue use if experiencing mood swings, nausea or elevated blood pressure.
The article, accurately summarized and absent any clickbait titles: "They have a faster supercomputer than we do. That means they are ranked higher, and are faster than ours. We want the fastest supercomputers. Whoever has the fastest supercomputer can solve all our problems, but that person only. It should be us, so we need the fastest supercomputer."
Slashdot: Playing Favorites Since 1997
I don't know of any such applications that would be best described as a national security issue.
Fucking Trump, trying to bring back manufacturing when he doesn't understand the concept of "robot".
Perhaps, but I'm pretty sure he understands the concept of unemployment.
In your opinion, is supercomputing more important than bringing back jobs?
What's your stance on H1B then?
I don't know of any such applications that would be best described as a national security issue.
I certainly can, although I don't know whether they're what the author has in mind. The two applications that come immediately to mind are naval: high precision weather forecasting; and submarine identification from sensor data.
Post may contain irony: discontinue use if experiencing mood swings, nausea or elevated blood pressure.
"Build better bridges".
Not really. The better we've become at engineering, the more we cut the bridge designs from "massively overbuilt, in such a way as to endure they never fall apart" to replace them with "barely overbuilt, in such a way as the first storm slightly out of the overage tolerance we've allowed will cause everything to be destroyed".
Seems stupid.
Rather than trying to figure out how to cut our tolerances as close to the bone as possible, we should probably go back to massively overbuilding things -- and then use our knowledge of tolerances to *ensure* they are massively overbuilt.
If we did that, we wouldn't have things like the 2007 I-35W bridge collapse happening. The bridges might sink into the ground under their own weight, but they wouldn't be collapsing.
These HPC people are also glossing over the issue that for most important problems, parallelizing over commodity CPUs connected by commodity networks (i.e. the cloud) is far more cost-effective than the "big iron" shared-memory HPC systems,
Nope there's a ton inaccurate about that. Firstly, supercomputers haven't been shared memory for decades. They're built with commodity (usually) CPUs with phenomenally expensive interconnect hardware. Sometimes expensive off the shelf, sometimes proprietary. Occasionally built right into the same die as the CPU.
A super computer lives and dies on the quality of the interconnect. Using someone else's server with normal networking hardware simply won't work well for many HPC problems because they'd spend most of the time waiting for events.
Ethernet first appears at position 94 on the top 500, with a machine using 10G Ethernet. It so happens that it has the dame gen xeons, running marginally faster. It has over 3x the number of CPUs to reach the same speed on the LINPACK benchmark, one known for not stressing the interconnects all that well.
Not that Google sms Amazon use entirely normal networking hardware, but their datacenters are optimized got throughout, not for machine to machine latency and bandwidth.
SJW n. One who posts facts.
China is taking a lot of very slow consumer chips and making a fast networked computer system.
That's not right. It's not even wrong.
You don't seem to understand what modern supercomputers are, what China are doing or what the hard bit is.
The Tianhe-2 (current #2) uses Xeon and Xeon Phi chips like many of the other top 500. Xeon Phi is the Xeon equivalent of a GPU which is more or less a large bunch of rather slow, FP heavy cores connected on die with a good interconnect. They couple the whole show together with their own custom interconnect, which is one of the hard bits. Intel spent $125 million to buy a previous generation interconnect from Cray, because that was easier than rolling their own.
But anyway the Tianhe-2 is like many (the majority) on the top 500, COTS CPUs, COTS coprocessor cards and some sort of cool interconnect.
And then there's Sunway TaihuLight, the current #1. That's another ball of wax. That's a bit more like the K-computer (#7) which has a large number of custom CPU dies with relatively low clocking, high floating point density cores with a networking interconnect right there on the die for extra low latency. The Sunway CPUs are a bit like the Cell in archiecture. Either way, they've got a huge amount of floating point grunt and are not remotely consumer chips.
Only the USA has the university graduates to design a real super computer with super computer ready CPU designs.
Actual verifiable facts disagree with you. The #1 position is a custom design made in China. The former #1 (now #7) is a custom design made in Japan, and has the best peak to max ratio on the list.
The US currently dominates the list, but is not the only player. Your claim is nothing but mindless jingoism fuelled by an ignorance of how supercomputers work.
A fast network of consumer cpu's is not a real supercomputer anymore.
A very fast, low latency network of CPUs is the *only* thing a supercomputer is anymore.
With a new ranking standard the USA is number one again.
Right so if y'all simply define yourselves to be #1, then youre #1. What does that achive?
SJW n. One who posts facts.
The revolution for the NSA contractor was to offer the NSA desktop computer CPU's on a fast network
Stop banging that drum: it's just not true. I think the last single CPU supercomputer was the Cray 1. Even Cray himself who rather disliked multiple CPUs (would you rather have 1024 chickens or a strong ox) had to put 8 in the Cray-2.
From then the number only went up. Even Mr Cray started going the massive parallelism route before he died. The problem is that you can't cram the power dissipitation or memory bandwidth into a really really small space. And you can't get the clock speed up that high so you have to do stuff in parallel as well. Once you start going wide you hit the wall after a while and need multiple streams of execution otherwise too many dependencies build up.
There's no grand conspiracy from the NSA here, just a mix of physics and computer science. The first #1 to start going to large numbers of CPUs (still vector) was IIRC a Japanese machine. The last gasp of mega CPUs was also incidentally Japanese, the Earth Simulator which used Vector processors, pushing out an RS/6000 system, though even that had tons of CPUs. Just not as many (each vector CPU was about 6-10x faster than the RS/6000 server/desktop processors).
After that it was clusters all the way. The next time Japan hit the #1 spot was with the K-computer which had piles and piles of SPARC CPUs, customised with extra floating point, but the big thing was the interconnect on die to make the network extra extra fast.
So, this is no NSA conspiracy unless you think they've managed to persuade Japan to ignore the "true path", even though they clearly had and have cutting edge supercomputer tech there. It's a combiation of physics and computer science.
SJW n. One who posts facts.
Nope - they've published video of the factory creating the trash can case and everything. They built a very expensive facility to do it.
It's a shame they aren't using it to make a computer that anyone actually wants to buy.
Slashdot still doesnâ(TM)t support Unicode after it was added to the HTML standard in 1997.
re "Stop banging that drum"
I was thinking more of what the NSA learned around the Project THOTH decade, 1984-88.
Domestic spying is now "Benign Information Gathering"
That is what happens when you educate foreign nationals in your universities and make all of your stuff in their country.
Go suck a crocodile cock, spamhole.
I've calculated my velocity with such exquisite precision that I have no idea where I am.
Anyone who has that kind of money in China has "ties to leading families of the Communist Party" - that's how they got the fucking money, or at least got to keep it.
Well, duh. That's why they went into power in the first place, and continue to do so in most, if not all, countries. Even the west they throw out tidbits to the population, but behind closed doors, money changes hands (unofficially, or officially through donations) and this or that regulatory burden or law gets changed. That's why they go into power everywhere.
This worldview has yet to fail in its massive explanatory and predictive power.
(-1: Post disagrees with my already-settled worldview) is not a valid mod option.
I don't know of any such applications that would be best described as a national security issue.
One of the biggest government uses for super computers is modeling nuclear weapons. These models are used to create smaller, more powerful warheads and in the absence of actual tests (due to treaty bans) these are how we now maintain and upgrade our nuclear stockpile. Since the DOE is mentioned in the article, I'd imagine that is the national security issue they are concerned with.
I browse on +1 so AC's need not respond, I won't see it.
Capitalists spent the better part of a half-century driving out skills based work
Not surprisingly, hundreds of thousands of highly educated people decided to go home with their training after their individual H1-B indentures expired and compete with their old bosses
With EVERY product being offshored, the design work of man-millennia has become Chinese property
Thank you very much Capitalists, for making America wilt
The next generation of 'enemies' will now have superior tech to ours...and will win wars, unlike our Pentagon.
Hit it on the nose. Even in Mandarin, (typically) when you speak of communism as the M.O. of the country, you refer to it as "Chinese style communism" (loosely translated).
The kicker is, as you mentioned, the centralized planning. With a five-year plan--which can fluctuate to extreme degrees responding to acute problems--one can obfuscate earmarked spending and hide actual power structures holding up the government, hide failures/give the party momentum to spin those failures into neutrals or successes.
I find 'Chinese-Style Communism' quite fascinating.
OMG facts!
What do you expect, when you outsource hi-tech and manufacturing? Of course they are going to surpass you someday.
Tell him that you need all these new supercomputers to make his tweets the bestest, and most read. If that doesn't get you a $10B or so then nothing will.
Except it's not about some vague notion of the "strength" (whatever that means) of the US or the "weakness" of everyone else, or being "superior" in terms of some sort of abstract `virtue'. It's about security vs. vulnerability, which are in fact appropriate to define in terms of the capacity of others to do you harm.
The concerns here are things like:
* China being more capable than the US of running the sort of simulations involved in developing advanced weapons systems, such that China could become invulnerable to US weapons tech while simultaneously meaning that the US is would become increasingly vulnerable to Chinese weapons tech.
* China being more capable than the US of running analytics such that they can circumvent US surveillance capabilities while simultaneously surveilling the US and its allies in ways that the US cannot circumvent.
We're not talking about being on the beach and being jealous of the guy with the big muscles because all of the girls are fawning over him, and figuring that if he'd just leave and let you be the only guy there then surely all of the girls would be fawning over you. It's more like thinking "that guy with the attitude problem has been working out and turning into a real bruiser; pretty soon I might not be able to keep him from kicking my dog anymore".
-rozzin.
China is nearly half a century behind us in military tech. And currently, their budget is a quarter that of the US. I'm not exactly sure what you think they're going to do to us in this or the next generation, but by the time they actually catch up - even if we merely matched their budget - we'd be back to where we were 50 years ago with Russia - neither defensively able to counter the other's offensive capabilities, thus "mutually assured destruction" being the deterrent. On the other hand, maybe the US doesn't need to project military force across the entire farking planet? What craziness could we accomplish if we were just using our military to defend ourselves, not police the planet? (prior military, btw)
You seem to have mistaken me for one of the people in the article. I was merely explaining to article to someone who had misunderstood it (actually, a apparently a number of people who had misunderstood it--seeing as the mistaken comment was highly rated).
I have not done my own analysis of Chinese vs. US capabilities, and even if I had it wouldn't have been based on whatever these people supposedly saw at the conference (since I wasn't there). Maybe the observations expressed in the article were correct, maybe they were mistaken, or maybe they were even lies.
When you write...:
Considered that perhaps the people behind the statements in the article may simple not agree with your premise that cold war is actually a desirable state of affairs. And that's probably in some part due to the fact that "mutually assured destruction" only works as a deterrent on people who aren't suicidal or otherwise find with their victories being pyrrhic; consider, for example, the conversation that Robert McNamara says he had with Fidel Castro in `The Fog of War'--something like:
Another consideration is that 50 years can actually be either an incredibly short or incredibly long period in terms of political relations, military technology, and military capabilities (where "capabilities" isn't actually quite the same thing as "technology"). And part of that consideration is that you may actually not get to choose whether 50 years is `short' or `long'.
And yet another is that, once you get to that `equilibrium' of mutually-assured destruction, even if you assume that everyone else with heavy arms is perfectly sane, it's still not a stable state. e.g.: what if the other if both you and the other guy have enough nuclear ICBMs to ensure mutual destruction given current launch- and early-warning technology, and current levels of strategy and tactics..., and you're not actually at the absolute pinnacle of all of those things yet? What if ICBMs aren't actually the quickest way of obliterating your enemy that will every be possible, and if there are new breakthroughs in strategy or tactics possibly just around the corner?
-rozzin.