Slashdot Mirror
Breaking Supercomputers' Exaflops Barrier
Nerval's Lobster writes "Breaking the exaflops barrier remains a development goal for many who research high-performance computing. Some developers predicted that China's new Tianhe-2 supercomputer would be the first to break through. Indeed, Tianhe-2 did pretty well when it was finally revealed — knocking the U.S.-based Titan off the top of the Top500 list of the world's fastest supercomputers. Yet despite sustained performance of 33 petaflops to 35 petaflops and peaks ranging as high as 55 petaflops, even the world's fastest supercomputer couldn't make it past (or even close to) the big barrier. Now, the HPC market is back to chattering over who'll first build an exascale computer, and how long it might take to bring such a platform online. Bottom line: It will take a really long time, combined with major breakthroughs in chip design, power utilization and programming, according to Nvidia chief scientist Bill Dally, who gave the keynote speech at the 2013 International Supercomputing Conference last week in Leipzig, Germany. In a speech he called 'Future Challenges of Large-scale Computing' (and in a blog post covering similar ground), Dally described some of the incredible performance hurdles that need to be overcome in pursuit of the exaflops barrier."
They're building a god.
How is exaflop a barrier? Is there some atypical difficulty in exceeding an exaflop?
My take away from reading this and the blog post is that, while NVIDIA may consider graphics to be their bread & butter, it looks like they're looking at this space (HPC) very seriously in the long term--perhaps they even think they can dominate it. This is a big difference from the other players: IBM isn't bothering to throw POWER at it, and AMD/ATI is only present on older machines; ATI in particular seems more interested in going after the mobile space rather than HPC. I don't know what to make of Intel other than they know they're the choice for the non-GPU side and are at the top of their game.
One problem I see is that NVIDIA is still a fabless house and has performance limitations tied to whatever fab they partner with; perhaps this is why they downplay process gains in the blog post.
Of course, if the conspiracy theorists are to be believed, NSA and friends already have this 10-years-into-the-future technology...
Dunno about "top secret", but the DoE puts a huge amount of computing resources into physical simulation. Check out some of the NERSC projects (GTC, for example).
Hmm, Mr. Fusion is due in a couple of years...
“He’s not deformed, he’s just drunk!”
Oops, sorry,
Should have used "tera" in place for "giga" ...
Muchas Gracias, Señor Edward Snowden !
Imagine a beowulf cluster of.... What? All supercomputers are basically beowolf clusters now? Umm...Ok, is Natalie Portman still topical?
Well I don't know anything at all about nuclear simulations and fluid dynamics modeling...
But for pure benefit to mankind I'd say folding@home is a pretty worthy project. It's been running for years and has helped make actual discoveries and raised understanding of protein folding's effects.
According to Wikipedia it was running at 14 Petaflops when last updated. Would taking that up to an exaflop be a huge benefit? You bet!
How about being able to simulate an entire life cycle of a human body at atomic scale? That would gain us tremendous understanding of well... EVERYTHING.
Most definitely there are worthy projects that have a real need for exaflop computing and it's not a waste of time.
You remind me of my friend who years ago said that his 802.11b wireless network was as fast as he'd ever need. Guess he didn't plan on people watching multiple HDTV streams throughout the house.
Cwm, fjord-bank glyphs vext quiz
So if we JUST put roughly 30 of the Tianhe-2s or 500,000 nodes with 100,000,000 computing cores in one big system, we'd have our exascale computer!
Anyone want to venture a guess how long it'd take Intel to make 1,000,000 Xeons and 1,500,000 Phis?
I can't wait to see the day, but me thinks we have a long way to go!
I can't believe some folks thought the Tianhe-2 was going to be the one to break the exaflop barrier! OOPS, only made it 3% of the way there...
Cheers!
humans will even worship a rock or lump of baked clay, for something to be a "god" only requires worshipers.
you'll be glad too know I use money and consumerism only to worship myself.
Does anyone have an idea of what these extremely expensive systems are even for? And don't say password cracking/NSA, because both of those tasks are "embarrassingly parallel", so that you can use a cloud of separate computers rather than a tightly interlinked network like a supercomputer.
Are there real world problems right now where another 100x more CPU power would make real, practical differences? (versus making the algorithm more efficient, etc)
we all know Chinese numbers represent a value exactly 14% less than what the rest of the world agrees on.
But for pure benefit to mankind I'd say folding@home is a pretty worthy project. It's been running for years and has helped make actual discoveries and raised understanding of protein folding's effects.
According to Wikipedia it was running at 14 Petaflops when last updated. Would taking that up to an exaflop be a huge benefit? You bet!
While not wishing to critisise folding@home specifically, we should be careful not to assume that there is an automatic progression from data to knowledge to understanding and hence to benefit. And with rising costs (both financial and environmental) we should not blindly assume that building huge supercomputers or running millions of inefficient home computers 24/7 is an inherently good idea.
So, you are claiming that Obama is really a puppet animated by a DoE supercomputer, running software written by the NSA?
Sleep your way to a whiter smile...date a dentist!
It won't work. Captain Kirk will just ask it what love is and it'll blow itself up.
systemd is Roko's Basilisk.
Moore's law predicts that the "factor-of-33" will be bridged in about 10 years. There is only a factor of 20 to the "peak performance", so about a year before that, peak performance might topple the exabyte "barrier".
(Some people plug in different constants in Moore's law. I use factor-of-1000 for every 20 years. That's 30 every 10, 2 every 2, and about 5 every five. This has never failed me: it always works out).
artificial intelligence will never match natural stupidity
You know, it's comment like this is why I rarely bother coming back to slashdot anymore.
You know, it's comments like this is why i think you're a total dweeb, and everyone knows you really can't get enough of /. while you're sitting there cooped up in your "command center" in your mom's basement stuffing your pizza face with McDonald's fries.
And no doubt behind Slashdot you have a bunch of tabs with Google image searches for "boobies".
Those are called "sheeple" by those who truly understand the distinction
no, they're called apple customers
So, you are claiming that Obama is really a puppet animated by a DoE supercomputer, running software written by the NSA?
fuck no... obama isn't that smart... he's more like a commodore 64 with a virus
That reminds me, what exactly IS that top secret program they use the Department of Energy's super computer for?
virtual porn... the government felt it would be a little unethical to use pixar's infrastructure
We're at 5.4% of exaflop scale. Somehow I don't think this is a 2013 / 2014 goal ;)
Some developers predicted that China's new Tianhe-2 supercomputer would be the first to break through.
Wait... *what* uninformed developer(s) predicted that? The previous record (six months ago) was set by Titan, at 17.59 Petaflop/s. So to pass the exaflop barrier this time around would require over a fifty-fold improvement -- something never before seen in the history of the Top500 list. Did someone *really* make this prediction, or is author Kevin Fogarty just making shit up?
Tom Geller
Unless the expensive Exadata box we just bought isn't capable of the exa-stuff they promised.
none
A more pessimistic estimate would say Moore's law only gets you a doubling every 3 years nowadays, so a factor of 32 would take 15 years to work out. See the troubles there were for e.g. TSMC moving to 28nm, and now 20nm.
An exaflops supercomputer would still be possible, with a 10x boost from Moore's law over 10 years and building a 3x bigger supercomputer.
Mainy important science problems are at least 4D in nature: three space dimensions and time. These include weather prediction, seismic prospecting, fluid dynamics, etc. 5.6 to the fourth power is one thousand in creased cost. You either get a finer grid or larger grid.
I heard a NOAA talk in Boulder about the erroneous Hurricane Sandy prediction. The "European" weather model correctly predicted the rare west turn of the northeastern hurricane while the US models did not. The Europeans used a 20 km would wide grid model and the SU a 32 km grid cell. The Europeans had more powerful computers. The US runs more frequent "incremental" model updates, while the Europens run fewer full calculations.
They are much lower power and somewhat lower speed than desktop CPUs. You'd have to use many more of them. Some projects are trying this.
The Jaguar/Titan system mentioned in your link is used for unclassified scientific computing. The NSA is building a computer facility at ORNL, but that's a different system (and was never claimed to be for stockpile stewardship). They don't put classified jobs onto unclassified systems.
The human brain can perform around 200-2,000 Petaflops (0.2 - 2 Exaflops) - when we compare it to computers.
The problem is that we can only access the conscious part, which is probably in the range of 100-200 Flops (Note, no mega, giga, or tera).
The subconscious part is where the real processing power lies. If we could simulate that in a computer, it would be tremendous in maybe understanding how it works.
For example: the human brain has the ability to "foresee" the future within a timeframe of around 0.2 seconds (or less). How does it do this? How does that work? Is it only a result of huge processing power or something else? Does two exaflops result in consciousness?? Questions after questions...
This comment is more insightful than funny. I doubt a machine could ever reproduce this condition. When is the last time you met a stupid person? Sure, at the time I'm it was very frustrating. But even to this day, you remember a lot of them, and it makes you feel good. Doesn't it?
nonsense, man is the measure.
Damn you Kirrrrrk!
And yet when it comes to getting elected, he seems pretty clever, or at least more clever than the competition.
The advantage of Xeon Phi cards is that parallelization on those cards works similar like classical parallelization on supercomputers.
Not really, no. Classic supercomputers were vector machines whereas Xeon Phi is wide SIMD.
You just use MPI
MPI is equally applicable to GPU or Xeon Phi, it operates at a level above the raw computation. In both cases you have controlling CPUs with accelerators attached (GPU in once case, Xeon Phi in the other). MPI is used to manage the data flow between these units but has little to do with the architecture of those units themselves.
For GPUs, on the other hand, you have to adapt a lot of code.
You have to adapt code either way:
For GPU you express the problem as a scalar kernel that is executed in parallel. You have to make sure that the work doesn't overlap but you only have to consider one element at a time.
For SIMD you break your problem in to SIMD-width chunks that are computed in parallel. It is easier to synchronise operations but you have to fit the problem into chunks of the right size.
Xeon Phi has an advantage where you have existing SIMD code (e.g. SSE), but if you are starting from scratch then there is no clear winner. And HPC code is increasingly being written in languages like OpenCL and CUDA which are designed for GPU rather than SIMD.