Beta

Slashdot: News for Nerds

×

Welcome to the Slashdot Beta site -- learn more here. Use the link in the footer or click here to return to the Classic version of Slashdot.

Thank you!

Before you choose to head back to the Classic look of the site, we'd appreciate it if you share your thoughts on the Beta; your feedback is what drives our ongoing development.

Beta is different and we value you taking the time to try it out. Please take a look at the changes we've made in Beta and  learn more about it. Thanks for reading, and for making the site better!

Intel Aims For Exaflops Supercomputer By 2018

Soulskill posted more than 3 years ago | from the still-won't-run-crysis dept.

Supercomputing 66

siliconbits writes "Intel has laid down its roadmap in terms of computing performance for the next seven years in a press release; in addition, it revealed its expectations until 2027 in one deck of slides shown last week. The semiconductor chip maker wants a supercomputer capable of reaching 1000 petaflops (or one exaflops) to be unveiled by the end of 2018 (just in time for the company's 50th anniversary) with four exaflops being the upper end target by the end of the decade. The slide that was shared also shows that Intel wants to smash the zettaflops barrier — that's one million petaflops — sometime before 2030. This, Intel expects, will allow for significant strides in the field of genomics research, as well as much more accurate weather prediction (assuming Skynet or the Matrix hasn't taken over the world)."

cancel ×

66 comments

Is there an upper limit? (1)

MrQuacker (1938262) | more than 3 years ago | (#36525064)

At what point does more computing power not matter anymore?

Re:Is there an upper limit? (2)

buchner.johannes (1139593) | more than 3 years ago | (#36525068)

Exaflop computing is a requirement for the Square Kilometre Array. There is still a long way to go until there might be an upper limit, especially in Radio Astronomy.

http://www.skatelescope.org/ [skatelescope.org]
http://en.wikipedia.org/wiki/Square_Kilometre_Array [wikipedia.org]
http://www.ska.gov.au/ [ska.gov.au]

Re:Is there an upper limit? (0)

Anonymous Coward | more than 3 years ago | (#36525198)

I would LOVE to create full-hd 3d flame fractals in 200 fps...

Re:Is there an upper limit? (0)

Anonymous Coward | more than 3 years ago | (#36525080)

There is no upper limit.
You never run out of data to process, down to atomic level.
If that doesn't eat up all your processor, someone's gonna implement the simulation in javascript, and you're good for another 30 years.

Re:Is there an upper limit? (2)

AHuxley (892839) | more than 3 years ago | (#36525224)

Power supply can get be an issue. You can read about the NSA and the size and new power supply needs as it expands at Fort Meade:
http://publicintelligence.net/nsa-site-m-cybercom/ [publicintelligence.net]
No upper limits, just more land, power, cooling and smart people to keep the funding flowing.
What each chip can do is limited, but you just keep buying more :)

Re:Is there an upper limit? (0)

Anonymous Coward | more than 3 years ago | (#36530270)

Power supply an issue? How long before a collection spacecraft can be placed in orbit around the Sun outside the plane of the Solar System and send the electricity back to Earth from there? Might be feasible by the time a supercomputer needs that much extra power.

Re:Is there an upper limit? (0)

Anonymous Coward | more than 3 years ago | (#36525300)

> At what point does more computing power not matter anymore?

yawn... At what point does this stupid question not being asked anymore?

Re:Is there an upper limit? (1)

MacTO (1161105) | more than 3 years ago | (#36525306)

For home and office needs, we have been well beyond the upper limit for well over a decade.

For most business needs, we have been well beyond the upper limit for some time. But that does not consider the needs of all businesses.

For scientific, engineering, and military needs -- well, we have a bloody long way to go. Supercomputers aren't built for national prestige or any of that other nonsense, simply because they are too expensive and become obsolete too quickly. These computers are built to address current and pressing needs. As a scientist, I can point to processing the immense amounts of data that we currently create or the development of more sophisticated models of physical systems. Engineers would face similar issues. The military too, plus they have intelligence applications. So Intel will find ready customers if they are the leaders at that point of time.

Re:Is there an upper limit? (2)

Arlet (29997) | more than 3 years ago | (#36525468)

For home and office needs, we have been well beyond the upper limit for well over a decade.

For video manipulation, it can still take a unpleasantly long time.

Re:Is there an upper limit? (1)

ChatHuant (801522) | more than 3 years ago | (#36532532)

For home and office needs, we have been well beyond the upper limit for well over a decade

What I think happened is the attention of the computing world and of the enthusiasts has been focused on the whole Internet thing. This has led to a slow-down on the home/office application development side; as a result, most things we use computers for in homes and offices are the same we had ten years ago, so of course current computers are powerful enough to handle them. But I can think of lots of new home apps that will need more resources; and I'm quite sure there will be another computer revolution (maybe soon) that will trigger another speed race. Could be robotics, or fully immersive virtuals, or direct computer/neural interfacing, or, more likely, something completely different :)

There is, but it is far off (4, Insightful)

Sycraft-fu (314770) | more than 3 years ago | (#36525644)

Basically a computer needs to be able to do anything a human can ask of it. Well we can ask an awful lot. I want a computer that can understand natural language and respond in kind. I want a computer that can render 3D graphics that look perfectly real. I want a computer that can accurately model the weather system of the entire planet, and so on.

We are still a long, LONG way from computing power not mattering anymore. Particularly at the high end where this is targeted. While the weather system modeling might be silly for a home user to say, it is something that we'd very much like a computer to do. However right now all the systems are crude models, very much is simplified because there just isn't the power to truly model everything down to, say, the molecular level.

However with enough power, such a thing could be done. How much I don't know, way more than we've got now, but it is perfectly possible.

So we really don't yet know what the upper bounds on what we might want in terms of processing power is. We won't really know until we start reaching it. We'll start having systems and any time we think up something new for them to do, they'll be able to do it with power to spare. Then we'll know "This is it, there really isn't a need for more processing power."

We may well hit physical limits before that though.

Re:There is, but it is far off (0)

Anonymous Coward | more than 3 years ago | (#36527194)

I remember hearing that most weather system simulations work on a granularity of 1km ? might have changed since then.

Still it shows just how far off we are from the improbable per molecule granularity.

Re:There is, but it is far off (1)

DMUTPeregrine (612791) | more than 3 years ago | (#36534718)

Not just improbable. Impossible with the current understanding of physics. Heisenberg's uncertainty principle would need to be false to allow perfect weather modeling. And you'd have to know the positions and momenta of all particles in the universe to infinite precision to get a long-term accurate model.

Re:There is, but it is far off (0)

Anonymous Coward | more than 3 years ago | (#36528236)

I want a computer that can understand natural language and respond in kind. I want a computer that can render 3D graphics that look perfectly real.

Both of those tasks might be within the capacity of current hardware - if we could only write the right software.

Re:There is, but it is far off (1)

insertwackynamehere (891357) | more than 3 years ago | (#36538592)

Well raytracing gets pretty close. Imagine extremely high end raytracing that can be done in real time for video games. Since this is just simulating the physics of light (if I'm not mistaken), perfecting this and having the hardware to do it very quickly would go hand in hand with amazing 3D graphics beyond any game we play now.

Re:Is there an upper limit? (0)

Anonymous Coward | more than 3 years ago | (#36527140)

No, there is no limit, as you can always easily double the computational power of the most powerful computer out there, just by putting another one next to it ;)

Re:Is there an upper limit? (1)

Mr.Z of the LotFC (880939) | more than 3 years ago | (#36531794)

Well, you eventually get a black hole if you pack them into a 2- or 3-dimensional array (not to mention heat problems in 3D). & if you put them in a line you start having problems with the speed of light. So miniaturization is important even if funding is unlimited.

Re:Is there an upper limit? (0)

Anonymous Coward | more than 3 years ago | (#36553258)

When Second Life no longer sucks.

Trickle down (1)

symes (835608) | more than 3 years ago | (#36525074)

While this sounds impressive, some of us would like to have something on our desktop that is capable of, perhaps, a petaflop or two. Undoubtedly, genomics research would gain massively but there are a lot more reseachers out there, I am sure, whose work would benefit from easier and daily access to such out of the box resources. Getting supercomputing resources to the masses would seem like a worthy goal, rather than just hitting exaflop headlines.

Re:Trickle down (1)

IrquiM (471313) | more than 3 years ago | (#36525166)

That's what you have GPUs for. Ask Nvidia or AMD for that sort of things, not Intel

Re:Trickle down (1)

Namarrgon (105036) | more than 3 years ago | (#36525232)

Not all workloads translate to the specialised SIMD arrays that are modern GPUs. There is still a need for large arrays [intel.com] of more general-purpose computing, as well as a need for fast single core computers (not all jobs can be computed in parallel either).

I write visual effects software in C++ and in OpenCL. Where OpenCL is practical, GPUs can be 10x faster, but there are still many cases when it's too awkward or has too much overhead to use a GPU.

Why would the Matrix or Skynet... (2)

dohzer (867770) | more than 3 years ago | (#36525086)

...make accurate weather prediction any less necessary?

Re:Why would the Matrix or Skynet... (1)

Arlet (29997) | more than 3 years ago | (#36525450)

In the Matrix, the weather predicts you!

Re:Why would the Matrix or Skynet... (1)

Lord Byron II (671689) | more than 3 years ago | (#36530072)

Okay, that's the funniest "In Soviet Russia..." joke I've ever read. Mod parent +1 funny!

Re:Why would the Matrix or Skynet... (0)

Anonymous Coward | more than 3 years ago | (#36525466)

Uuum, how would you not have accurate weather predictions in the Matrix (it's just less work, and we know the machines are bad at creativity), and with Skynet, the sky would be black and radioactive, period. No going out in that weather. ^^

Re:Why would the Matrix or Skynet... (1)

Lord Byron II (671689) | more than 3 years ago | (#36530100)

I used to live in Phoenix. It's dry and it's hot. Phoenix is in the middle of a valley surrounded by mountains, so storm systems tend to wrap around the city rather than pass overhead. With the exact same weather every day, it's hilarious watching the weather forecasters try futilely to come up with new and exciting ways to describe that day's weather.

Re:Why would the Matrix or Skynet... (1)

progliberty (1530571) | more than 3 years ago | (#36553312)

Don't worry, you'll get lots of nice storms once they strip-mine a few of those mountains.

Outdone by the Japanese (1)

Kagetsuki (1620613) | more than 3 years ago | (#36525146)

Looks like someone is a sore looser.

Re:Outdone by the Japanese (1)

matty619 (630957) | more than 3 years ago | (#36525250)

Sore looser? That sounds.....kinky.

lol...sorry, Spelling Nazi at your service ;)

Re:Outdone by the Japanese (1)

antifoidulus (807088) | more than 3 years ago | (#36525294)

Why, did they take too much stool softener?

Ok, with that out of the way, unlike Fujitsu, who added on extra SIMD co-processors to their CPUs SPECIFICALLY for the K-computer, Intel does not currently do is offer CPUs designed specifically for HPC. Now of course a lot of advancements in general purpose CPUs can be applied to HPC. Overall the Earth simulator cycle seems to be repeating itself->a Japanese company creates an incredibly fast computer that makes heavy use of vector processors, then cannot sell enough of these to make keeping them up to date financially feasible, and eventually the commodity market outperforms them.......wait a few years and repeat.

Re:Outdone by the Japanese (0)

Anonymous Coward | more than 3 years ago | (#36525342)

Yeah, the Japanese sure are winning aren't they. That's why Fujitsu's last foundry (Mie) is shutting down permanently, and all fabrication moving off to TSMC. Intel should be so scared.

Try this memory tool... (0)

rts008 (812749) | more than 3 years ago | (#36525516)

"Loose as a goose, and you never lose."
That's how I learned it as a youngster. :-)

English can be tricky and crazy that way.

Congrat's on Japan taking the #1 spot!

Just think... (1)

PhunkySchtuff (208108) | more than 3 years ago | (#36525150)

Just think how much they could make mining Bitcoins on this thing... Oh, actually, hang on...

Re:Just think... (1)

metalmaster (1005171) | more than 3 years ago | (#36525262)

They crashed.....so theres only room for a rise now right? Never lose hope [/sarcasm]

It will use the CPU Lambardi chips (0)

Anonymous Coward | more than 3 years ago | (#36525164)

Intel plans to put 1,000 pentium pro's with SIMD per cpu for each GPU unit. There is an interview with one of the intel engineers [ebaumsworld.com] on it.

Wasn't 2011 supposed to be the year Netburst (1)

matty619 (630957) | more than 3 years ago | (#36525206)

Scaled to 10 GHz? [geek.com] (the comments are fun to read)

It's hard to take a claim like that seriously since that famous prediction. Oh well. At least they redeemed themselves with the Core architecture. In my little book anyway.

Re:Wasn't 2011 supposed to be the year Netburst (1)

Your.Master (1088569) | more than 3 years ago | (#36525482)

I love the comments there. Compared to the consensus of the commenters (circa 2000, anyway -- by 2005 they started approaching conventional wisdom), Intel was extremely conservative.

Re:Wasn't 2011 supposed to be the year Netburst (0)

Anonymous Coward | more than 3 years ago | (#36525788)

I wonder is the Apple user still buying Apple now that they aren't running PPC :P

Looking at the other predictions thats pretty good (1)

nzac (1822298) | more than 3 years ago | (#36525764)

When you look at the number of people who still believed in the clock frequency version of Moore's law intel actually got it pretty accurate.
They are only out by a factor of two comparing that to the other numbers 128GHz!!!! Though doubling it again for consumer chips is going to take at least 10 years more.

When they are promising those kind of improvements for 2018 I'll let them be an order of magnitude out.

Re:Wasn't 2011 supposed to be the year Netburst (1)

somersault (912633) | more than 3 years ago | (#36526468)

Got to love all the idiots lambasting Brian L's sarcasm as if he were serious.. *facepalm*.

Re:Wasn't 2011 supposed to be the year Netburst (1)

CecilPL (1258010) | more than 3 years ago | (#36529720)

Well Netburst has been pretty much replaced with Core, but the new 10-core Xeons [intel.com] run at 2.4GHz, so that's 24GHz of computer power on a single chip.

Die size in square meters (2)

Ceriel Nosforit (682174) | more than 3 years ago | (#36525260)

If my calculations are correct, 1 exaflop with today's best hardware translates to roughly 2000 m^2 worth of CPU area. Reportedly a little less than half an American football field.

Still I'm not sure if it's enough for what I want, namely to create new lifeforms, artificially evolved to say live on a partially terraformed Mars. - That or give teens genetic upgrades for irises which change colour with their mood. The latter of these seems more likely to be met with commercial success.

Re:Die size in square meters (1)

rbrausse (1319883) | more than 3 years ago | (#36525528)

Reportedly a little less than half an American football field.

or - in a more useful and recognized measurement - almost 0,8 % of the LoC's floor space

Re:Die size in square meters (0)

Anonymous Coward | more than 3 years ago | (#36529318)

How many square Sagan ångströms would that be?

Re:Die size in square meters (0)

Anonymous Coward | more than 3 years ago | (#36526394)

Personally I'd like to see teens that listen to their parents, but well go for it with the iruses thing!!!

Re:Die size in square meters (0)

Anonymous Coward | more than 3 years ago | (#36526966)

Always listen to authority figures.

First retail quantum computer. (0)

bejiitas_wrath (825021) | more than 3 years ago | (#36525360)

http://venturebeat.com/2011/05/27/first-quantum-computer-sold/?utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+Venturebeat+(VentureBeat)#bXpulseX [venturebeat.com]

This is the first quantum supercomputer, a 128 qubit machine that is already here. I am not sure how it compares to the machine planned in the OP, but it would be quite good for cracking a password hash or any other number crunching task for sure. Obviously if this is available to Lockheed Martin, then I am sure that the NSA have already taken ownership of machines like this as well. But this could simulate the weather better than existing machines and improve long term forecasts as well as predict financial data.

The future is here.

80 cores (1)

Lord Lode (1290856) | more than 3 years ago | (#36525572)

In 2007, they were saying intel would have 80-core CPU's in 2011.

It's 2011 now, where can I buy one?

Re:80 cores (0)

Anonymous Coward | more than 3 years ago | (#36525798)

They only install them in flying cars.

Re:80 cores (0)

Anonymous Coward | more than 3 years ago | (#36527328)

You can get a 80+ core chips at nearly any computer hardware store, but not from Intel. Instead look for the boxes that say nVidia or ATI...

Re:80 cores (0)

Anonymous Coward | more than 3 years ago | (#36528672)

I heard you could get one at Knight's Corner.

Re:80 cores (1)

DanLake (543142) | more than 3 years ago | (#36534622)

In 2007, they were saying intel would have 80-core CPU's in 2011.

It's 2011 now, where can I buy one?

Intel showed a working 80 core research processor in 2007. Maybe that's what you heard about? It was used by researchers in the fields of distributed and parallel computing to develop new programming and compiler optimizations and operating system enhancements.

https://secure.wikimedia.org/wikipedia/en/wiki/Teraflops_Research_Chip [wikimedia.org]

Intel has since released a second generation research processor for distributed computing and cloud research called SCC, formerly know as Rock Creek. It is also available to universities and researchers but not a consumer product.

Disclaimer: I am an Intel employee but the statements and opinions here are my own based on the referenced link(s).

Exaflop vs. exaflops (0)

Anonymous Coward | more than 3 years ago | (#36525728)

1000 petaflops (or one exaflops)

Grammar nazi mode: it's "one exaflop", not "one exaflops". Yeah, one exaflop is several operations, but it's still "exaflop", for the same reason it's, say, "one kilometer" and not "one kilometers".

Re:Exaflop vs. exaflops (2)

Noughmad (1044096) | more than 3 years ago | (#36525954)

It is "one exaflop" and "1000 petaflops". FLOPS still meanst FLoating point Operation(s) Per Second. http://en.wikipedia.org/wiki/FLOPS [wikipedia.org]

Re:Exaflop vs. exaflops (1)

White Flame (1074973) | more than 3 years ago | (#36534740)

"one exaflop" means 10^18 floating point operations, with no time basis.
"1000 petaflops" means 10^18 floating point operations _per second_.

The 's' is not a pluralizing suffix, it's part of the FLOPS acronym. In this usage, it should be "one exaflops", or "one exaFLOPS" to capitalize properly.

Not very likely (1)

Anonymous Coward | more than 3 years ago | (#36525818)

http://spectrum.ieee.org/computing/hardware/nextgeneration-supercomputers/0

Re:Not very likely (0)

Anonymous Coward | more than 3 years ago | (#36533288)

Article should be prefaced with "Assuming total technological and algorithmic stagnation," then it would be honest.

Funny (0)

Anonymous Coward | more than 3 years ago | (#36527604)

It's not really such a wise-crack joke, but you had me laughing for a minute with the Skynet / Matrix comment :-)

4 Exaflops? Damn that's impressive! (0)

Anonymous Coward | more than 3 years ago | (#36527650)

It's almost enough to run Vista!

This is the NSA (1)

decora (1710862) | more than 3 years ago | (#36528002)

"In 2008 James Bamford's The Shadow Factory reported that NSA told the Pentagon it would need an exaflop computer by 2018."

from wikipedia.

80 Plus Platinum certified PSU? (0)

Anonymous Coward | more than 3 years ago | (#36528592)

You're going to need a small fission reactor to be packaged with your exaflopper.
As well as a few phallic cooling towers.

This is the anti-Moore's law (or the Anonymous Coward law) of computing.

So they can do what? (0)

Anonymous Coward | more than 3 years ago | (#36529488)

Play chess & jeopardy?

Intel just follows trend predictions from top500 (0)

Anonymous Coward | more than 3 years ago | (#36530004)

If you refer to the projected performance from the top500 (top500.org), you'll see that everybody will be competing to get the Exaflop by 2018 http://www.top500.org/static/lists/2011/06/perfdevel/Projected_Performance_Development.png.

For those who have some curiosity about what an Exaflop machine will probably look like and what it will be able to do, you can have a look at the IESP roadmap :http://www.exascale.org/iesp/Main_Page

never enough (0)

Anonymous Coward | more than 3 years ago | (#36539732)

CPU's are never going to be fast enough.

Check for New Comments
Slashdot Account

Need an Account?

Forgot your password?

Don't worry, we never post anything without your permission.

Submission Text Formatting Tips

We support a small subset of HTML, namely these tags:

  • b
  • i
  • p
  • br
  • a
  • ol
  • ul
  • li
  • dl
  • dt
  • dd
  • em
  • strong
  • tt
  • blockquote
  • div
  • quote
  • ecode

"ecode" can be used for code snippets, for example:

<ecode>    while(1) { do_something(); } </ecode>
Create a Slashdot Account

Loading...