U.S. Invests $258M in Supercomputing Race with China

https://www.wsj.com/articles/u-s-invests-258-million-in-supercomputing-race-with-china-1497549600

by sew • 9 years ago

69 79 9 years ago

JosephLark• 9 years ago

Some non-paywalled information at Anandtech: PathForward: US Dept. of Energy Awards $258M in Research Contracts To Develop Exascale Supercomputer Technology [0]

[0] http://www.anandtech.com/show/11547/us-dept-of-energy-awards...

sitkack• 9 years ago

Pennies.

jayzalowitz• 9 years ago

They shoulda invested 256 million, woulda been more efficient.

Yahivin• 9 years ago

You mean $268,435,456?

YokoZar• 9 years ago

256 mibillion

vmarsy• 9 years ago

top500.org, which maintains the list of the 500 fatest supercomputers in the world also has more details on this [1]

The PathForward program [...] address the four key challenges: parallelism, memory and storage, reliability and energy consumption

Energy consumption is becoming a very important challenge, as the article says, current petascale supercomputers consume "enough power to run a small town". Major efforts will need to be made to have Exascale supercomputers that don't consume the equivalent of 10^3 small towns.

[1] https://www.top500.org/news/doe-shells-out-258-million-to-si...

davidf18• 9 years ago

Competition is good. For research, The Cold War with Russia and the resulting "Space Race" was good. In Israel, they spend a much larger share of their national income on R&D, largely to develop technologies to defend themselves.

Thriptic• 9 years ago

This. I'd argue one of the biggest problems facing the United States right now is that there is a (falsely) perceived lack of competition / existential threats which is making us lazy and complacent.

crb002• 9 years ago

That's like 1/5 the cost of one F35 Flying Turkey?

salesguy222• 9 years ago

Hahahha! Also roughly the price of just 2 Flying Raptors, which have the distinct honor of being nearly unfliable by even the most skilled pilots.

losteric• 9 years ago

And a mere 2% the cost of the USS Gerald Ford (which would probably be destroyed within 24 hours of active war with China)

vonmoltke4• 9 years ago

I would like to see the details if your tactical analysis on that.

adventured• 9 years ago

Your premise is comical. Any war that involves destroying a US carrier and killing thousands of soldiers in doing so, will move to a nuclear stand-off almost immediately thereafter. China and Russia have always understood that. For China, the ability to sink a US carrier only matters in instances of absolute war, where hundreds of millions of people will die. Being able to sink a carrier if you're a major military competitor to the US, is low on your list of actual concerns in the end (the threat is worth drastically more than the actual ability; if you use the ability and succeed, it's WW3). Carriers are for force projection, some intimidation, and fighting nations not named China or Russia.

unsupak• 9 years ago

why are Americans so hell bent on having a war with China? last I heard China doesn't have military bases around the globe like the US does.

bpodgursky2• 9 years ago

No, it's 2 F35s. Unit cost will be a bit under 100mm each.

trendia• 9 years ago

Does that price include oxygen? Not all of them do [0].

[0] https://www.theregister.co.uk/2017/06/13/f35_pilots_report_h...

dekhn• 9 years ago

Unit costs can only be determined after enough runs permit you to take the spent budget and divide by the number of units built. What you're describing is a projection and it only holds if the full run is built. If it gets cut short, the unit costs are higher because of the non-recoverable costs.

zt• 9 years ago

This is chump change, we should have invested $2.58 billion or more. China isn't taking half measures in their state's science and technology investments. They realize we're in a second cold war, just one where the two combatants have more economic interdependence. They are making investments like the ones that helped us win the last war.

MichaelBurge• 9 years ago

It's not like the DoE are the only ones with expensive computers: The NSA has their underground lair.

There's a commercial need for expensive distributed processing now, with Google/AWS/Azure spending a lot on video cards. The ML craze may make government spending on supercomputers less important, especially if people prefer to distribute calculations on many interchangeable computers.

I wouldn't mind a few NSF or DoD grants for novel computing architectures, but I'm not sure I buy that having massive supercomputers for economic forecasting is "strategically important for nations".

convolvatron• 9 years ago

these things get run in phases. phase 1 they insist on novel architectures, rethinking of basic premises, required involvement by academics.

by the end of phase 3 when they actually procure machines its mostly 'just give me one of what you're already shipping, and it better run MPI well'

doe exascale was supposed to be fundamentally different. because by the time you got there all the incremental improvements in power and latency management weren't nearly enough anymore...my guess is that they'll just make big gpu clusters and call it a day

its probably not that much money in the scheme of things, but the sad (or good) depending on your perspective is that HPC got so far out of the mainstream that when corporations finally got around to worrying about scaling, they just did their own thing. so I guess they can thank darpa/doe for infiniband? not really?

and they're still running fortran/mpi

nullnilvoid2• 9 years ago

They realize we're in a second cold war... For God's sake, don't start another cold war. This is an established incumbent, who has monopolized world power for decades, feeling threatened by a new kid in town. Any progress China has made has been labeled as threat, cold war, or dismissed as useless, or blamed on China's stealing US secrets. As an instance, China bought many Intel chips to create supercomputers. Then the U.S. government banned Intel from selling high end chips to China[1]. China had to create homemade chips for world's fastest supercomputer[2].

[1] http://wccftech.com/us-government-bans-intel-nvidia-amd-chip... [2] http://www.pcworld.com/article/3086107/hardware/chinas-secre...

myrandomcomment3• 9 years ago

I love this very naive view. China is 100% trying to beat and surpass the US as a superpower. They are not benevolent at all. Have a look at the ridiculous claims on the South China Sea for a start. I have spent lots of time in China. They still want payback for the humiliation they believe they endured in the 1800s. That is a one of the single most important driving factors in everything they do.

nitwit0052• 9 years ago

Honestly, faster super computers aren't going to affect the balance of power. There is a ridiculous amount of computer power available already.

jacquesm• 9 years ago

I'm not so sure about that.

Faster super computers allow you to do all kinds of things that would otherwise be quite hard. Think of supercomputers as weapons that you can fire very quietly, nobody even notices that you just tested some new (possibly nuclear) weapons system or that you designed some fighter jet, submarine or rocket.

Super computers are a game changer, not having them is as much as ceding territory, having them but not knowing how to use them has a similar effect. Just like computers were a game changer in the previous war so super computers will be a game changer in the next, with the added caveat that you might not even notice when it starts or ends other than that you suddenly find out that your counterparty runs the show because they cracked the codes you thought were safe.

davidgrenier• 9 years ago

It'll be a shame if they don't dump a chunk of that money on the Mill. However there may very well be some string attached unsuitable to Ivan Godard et al.

just2n• 9 years ago

I wonder if it being completely open source would have an impact on that.

deepnotderp• 9 years ago

The Mill isn't really very well suited for supercomputing workloads. It tries to do well on poorly written C++ code (aka "general purpose code") that OoO machines have historically don well on. By contrast, VLIW machines are perfectly suitable for supercomputing, and indeed, I think that Itanium did well in the HPC world until it was discontinued.

Something like the Rex Computing Neo would be more suitable for this.

zantana• 9 years ago

Does betteridge's law apply to conference talks? http://sc14.supercomputing.org/schedule/event_detail-evid=pa...

randcraw• 9 years ago

Here's the same content at AnandTech (w/o paywall):

http://www.anandtech.com/show/11547/us-dept-of-energy-awards...

killjoywashere• 9 years ago

USG has tons of compute. What we don't have is the tooling behind AWS or GCP (minus GovCloud which is mainly inaccessible due to infosec and budgetary insanity).

Our scientists look at what could be, look at their very interesting datasets, and then gnash their teeth for want of better tooling. People know this. Senior people from industry have screamed at senior government leadership that they are 20 years behind and falling faster. Nothing changes. Why?

Every time someone gets out from under their grad school payoff contract, they run to Google or Facebook.

Very few people are committing for the long term, and that's what's needed. Long-term commitment to steward a program. The people who remain often lack technical chops, have "certificates" in "cyber" something or other, but got through the Acquisitions school and now have authority to just pour money into blades, load Red Hat and mpi4py, and say "look, it's a supercomputer"!

They don't need to invest in iron. They need to invest in people. But they think they can just keep hiring new grads and expect magic.

Retric4• 9 years ago

AWS is a terrible match for super computing hardware. Sure, there are some embarrassingly parallel workloads that don't need fast interconnects, but mostly latency or memory becomes the issue.

cobookman• 9 years ago

Gcp has it's tpus though...

batbomb• 9 years ago

Dirty little secret:

Much the computing on these machines isn't supercomputing and doesn't need fast interconnects. In fact, many (most?) workloads would be best done in the cloud. Except we already bought these machines. So there's tons of frameworks around meant to use these machines like they are AWS pizza boxes. If it's inefficient due to an architectural mismatch, that's fine. You're still using the machine that was paid for, and the only numbers that matter is utilization, and those metrics aren't even defined in terms of CPU utilization. If you reserve 10 nodes, and only run on one node, it's still reported that a 10 node job ran for 6 hours, not that a 10 node job job ran at 10% efficiency for 6 hours.

As far as why the accounting is this way, it's assumed that users will attempt to maximize their usage since allocations are defined. But even then, you can blow through your allocation and generally ask for more. But that's not necessarily what happens.

In summary: Because we bought these big computers, we have to use them, even if the workload isn't a good match.

steve371• 9 years ago

only 200M? Am I the only one think it is a small budget for supercomputing?

meric4• 9 years ago

I've spent a lot of time in country X. They all want Y.

Okay. That's the kind of crap that was in the 1800's, that's still in the 2010's, unfortunately.

Your words sound like something from a grown up bully jealous of their victim's achievements in life.

myrandomcomment3• 9 years ago

How do you know I am not Chinese or my wife is not? Or my best friend? Or someone in my family?

I wish the best for China and the Chinese people. I hope for a peaceful and prosperous future. That however does not involve the threat to take Taiwan by force or claiming all of the South China Sea as theirs.

What I was trying to say is that my opinion is not informed based on what I have read in the press but based on the fact that I have spent time in the country and have read the local press, and spoken with the local people. The humiliation faced by the Chinese at the hand of the western powers is a common theme of books and movies and TV shows in China.

meric4• 9 years ago

Well, for one, I am Chinese. My parents are Chinese, so are my grandparents, and their grandparents.

I've listened to my grandmother talk about her grandparents, who are from the 1800's.

Of course we're drawn to movies and books and TV shows from that era. Same reason why teenagers in Europe are interested in Medieval stuff and Roman empire stuff. It's part of their history.

Is China taking Taiwan by force? Taiwan has been developing peacefully for almost 70 years since the civil war, I don't see any violence over there.

..."claiming all of the South China Sea as theirs"

Lots of governments do lots of crap, does that mean we can say the whole country and all its people want it?

And what does U.S. has anything to do with any of this stuff, anyway. Sure I understand if China was doing things in Mexico or the Caribbean, but I don't understand why they want to meddle in Ukraine, Syria, Afghanistan, and South China Sea, and then call Russia, Iran, China as agitators.

Do I think that's what the U.S. public support? Meddle in countries half way around the world? No!

And thanks for your best wishes.

dang• 9 years ago

Your words sound like something from a grown up bully jealous

This crosses into incivility and personal attack, and those things are not allowed here. Please edit all that out of your comments when you post.

We detached this subthread from https://news.ycombinator.com/item?id=14565533 and marked it off-topic.

mbloom19152• 9 years ago

Why is this in the DOE budget?

jarmitage• 9 years ago

This is where the US' climate change budget is going [citation needed]

Guess this now needs reconfiguring http://worrydream.com/ClimateChange

mbloom19152• 9 years ago

DOE? that cant be right

evanb• 9 years ago

Energy, not Education.

dis-sys3• 9 years ago

This is just stupid. US basically sent out a signal to its competitor to demand them to invest more heavily on their own R&D capabilities- what Chinese universities, research institutions and companies are going to do with such news? They take it to the Chinese central government to ask for more R&D funding.

Think about it - what China can possibly lose in such a race? When Americans are trying to maintain its No.1 ranking in the field, Chinese are just trying to further modernize its industry which stands at less than $10k GDP per capita, you can probably beat the next generation Chinese supercomputer, but hey, you are pushing for the creation/improvement of the Chinese supercomputing community!

Think about it again - what China can possibly lose in such a race? America won't lose its overall dominance just because some other countries have faster supercomputers, America is losing its influence around the global when it stupidly using its borrowed resources to force its main competitor (also No.1/2 debt owner) for more R&D spending.

nippples• 9 years ago

Son, you're not making too much sense. Of all things to be mad about, tech investments isn't one of them.

dis-sys3• 9 years ago

If supercomputer is a race and the race is between the US & China, then such investment from the US federal government using borrowed money is _not_ doing any favour to the US - it might build a few faster supercomputer in the end, but in the long run, such a direct head on competition which fuels Chinese R&D spending in arguably one of the most cutting edge fields is damaging US interests.

Why Chinese are building supercomputers on their own? Because the US laws prohibit any US supercomputer export to China, including the export of core components that can be used to build such supercomputers.

In case you are not familiar with the matter - the above mentioned core components are not some fancy parts guarded by your NSA - we are talking about parts that can be freely purchased on ebay for $300-1000 each.

adrianN• 9 years ago

More R&D spending is usually good for humanity, no matter what country foots the bill, so I'm okay with this development.

albertTJames• 9 years ago

Tweeting will be so fast, it is going to be great

U.S. Invests $258M in Supercomputing Race with China

Related Stories