The Future of Computing Performance: Game over or next level?

[zalbard]

Source: http://fudzilla.com/

A new report from the US National Research Council has warned that technology growth will grind to a halt unless a new generation of hardware and software is developed. The catchy-titled “The Future of Computing Performance: Game over or next level?” is packed full of bad news for the industry.

While microprocessors improved in speed by a factor of 10,000 during the 1980s and 1990s, two obstacles could mean computing power hitting a wall in the next decade, the report said.

While transistors have become ever smaller and more tightly packed, the speed at which microchips are clocked has levelled off, reaching around 3 gigahertz in 2005. That's because such fast chips generate too much heat to be used in smartphones and personal computers.

This plateau will kill Moore's law. While manufacturers have been fabricating two, four or eight microprocessor cores on a single chip to get around this hurdle, it is not enough the report warns. This is because the power efficiency of present transistors cannot be improved much more, and performance "will become limited by power consumption within a decade". It needs a yet-to-be-invented transistor architecture to save it.

Software is still behind hardware developments with multicore chips still not properly factored into designs. Software has to be designed to execute multiple tasks in parallel, rather than serially and few programs are up to this challenge.

But the report warns that converting the vast majority of software, written for serial execution, to work efficiently in parallel mode will be exceedingly difficult. It needs new software-engineering processes and tools. Programmers will need to be retrained, it warned.

The signs are pretty obvious, though, for a few years...

[FlawleZ]

Well I think we all knew that standard transistor based chips don't have much life expectancy left. I believe its around 11 or 12nm where the electrons tend to jump lanes because the distance between itself is so small.

[Johnny87au]

Kinda gathered this was happening, hope it doesnt happen in my life time though hehe, By the look of how things are moving im suprised it will hit us in the next 5 years! ... Im pretty sure with millions of transistors on a chip they can think of a new idea to be able to continue the fashion..

[mAJORD]

It's allready hit us..

There's not much 'current' computing you can't do with ease on a 5 year old dual core PC..

That was not the case in the 80s, 90s or early 00s..

[PatRaceTin]

hmmmmmmmmmm i will wait and see

[Eson]

Yes, lazy programmers free lunch is over. No more program-something-poorly-and-hope-computers-will-be-able-to-handle-it-when-its-released.
However if the code/problem is easy to parallelize there is still a free lunch but the portion will converge to a fix size with more cores according to Amdahl's law.

Parallel code might be new for software developers but its basic for electrical engineers.

Related: EE vs. CS

[Sn0wm@n]

has anyone noticed where this doom and gloom article comes from ???

[Olivon]

has anyone noticed where this doom and gloom article comes from ???

http://www.nap.edu/catalog.php?record_id=12980

[rcofell]

has anyone noticed where this doom and gloom article comes from ???

While I cannot say I have ever heard of the organization responsible for publishing the report, I can say with certainty that many of the committee members involved are what you'd considered experts in the field. Both sides, academia (William Dally & Mark Hill are those I'm familiar with) and industry (Intel/AMD/Nvidia/Google/etc.), are represented in the group.

Despite it appearing all doom and gloom to you, the issues at hand have been known for quite some time. The report is just summarizing them and recommending what could probably be defined as a national course of action, something targeted at policy makers. Mainly it comes down to advocating a better coordinated paradigm shift to (thread) parallelism, ideally through better education/research into algorithms, hardware and (power-)efficiency at all levels. Support is mentioned for specialized/heterogeneous computing resources as well.

Like some of the others have stated, nothing new here really.

[Easybeat]

Everything has an optimum size, power or usefulness to the consumer perhaps the PC has hit its optimum at least for the mainstream?

[Manicdan]

Everything has an optimum size, power or usefulness to the consumer perhaps the PC has hit its optimum at least for the mainstream?

a 600$ laptop will be able to do anything and everything quite well soon enough. gaming, htpc, portability, etc. and thats at 32nm. add in OLED, no need for more than 4 cores on the cpu, the rest of the power can go toward gpu which means a 40W total laptop package will be even more gpu based, so battery life gets better, and gaming gets better, and GPGPU makes everything feel so much smoother and faster. we really might hit the point in 4 years where the home pc is as good as we need it, and they last for a >decade instead of <5 years.

however outside of that the need for more is always going to be there.

[prava]

The article refers to the limits of our technology. Within a few years, when we get to it, we will need to find another path to keep pushing it. But that will require a huuuuge investment on many sides...

@Manicdan: don't think so. A mobile gpu rated @ 50W, to be as powerful as a GTX580 these days, will require an awesome integration node. Keep in mind that nowadays we can have more or less the power of a 4870 as a mobile gpu, but not much more. If you want to double that, or even more, we will need probably less than 22nm. So...not happening within this year (2011) nor probably next.

[Manicdan]

@Manicdan: don't think so. A mobile gpu rated @ 50W, to be as powerful as a GTX580 these days, will require an awesome integration node. Keep in mind that nowadays we can have more or less the power of a 4870 as a mobile gpu, but not much more. If you want to double that, or even more, we will need probably less than 22nm. So...not happening within this year (2011) nor probably next.

i was referring to what will be available for a laptop for mainstream that can work well for 90% of the population. a desktop will still be based on a max power envelope above 1KW where people can make a real gaming PC, i never inferred that a desktop would be useless after 22nm laptops come out

[Hannibal Lecter]

You know from time to time we get this kind of article. I have great faith in human ingenuity and I am certain that we will not hit a wall. This is not something new.

As ever, new discoveries will be made, new manufacturing processes will be invented. And don't forget as was mentioned above that there is a huge performance reservoir than can be tapped simply by writing better, a little more optimized software. Hardware has made such a rapid progress that there wasn't any incentive to optimize software.

Time will tell but I doubt we'll hit a technological wall for the foreseeable future.

[god_43]

we will move from silicon to something else once that happens, i think that its past 15nm we begin to reach the limits of silicon.

[VulgarHandle]

molded diamond transistors grown in vaccum FTW!

[445533]

So that means that programmers will have to code more efficiently instead of writing 600mb drivers and 4GB word processing applications.

[phantomferrari]

i think the age of silicon computing is about over but the soon to be future holds so much promise. replacing silicon with gallium arsenide, then eventually maybe even graphene in a decade. if you have been following IBM's recent breakthrough with photonics then we might not even need graphene in our chips

[Dimitriman]

I wish the software industry would drive the hardware industry instead.
Wouldnt it be nice if we had software capable of being executed in 24 threads but not yet the processor to do it?
Then we could just be counting the weeks till a 24 thread processor came out to provide all that power.
I like that world more than 'we have a 24 thread processor but we need to make each core as fast as possible so we still run those 1 threaded codes faster'

[Mechromancer]

molded diamond transistors grown in vaccum FTW!

^THIS! Carbon based circuits are cheap and durable. They'll care less about modern heat and voltage constraints if what I read about them at Science Daily is correct.

[Movieman]

I always chuckle when I see people bring up "Moore's Law" as if it were holy grail to the industry. It isn't and like everything else there will come a point where it's irrelevant.
Don't get me wrong, Mr Moore has my utmost respect and admiration but this "law" was from a certain timeand although technology has moved forward in that timeline there will come a time when it doesn't.
As to the future of computing I wish I was 40 years younger so I could get into the design end myself.
Fascinating business to be in. Challenges the mind to see what can be done.
I look at the guts of a 32nm chip and it boggles my mind to see what is possible.

[Manicdan]

I always chuckle when I see people bring up "Moore's Law" as if it were holy grail to the industry. It isn't and like everything else there will come a point where it's irrelevant.
Don't get me wrong, Mr Moore has my utmost respect and admiration but this "law" was from a certain timeand although technology has moved forward in that timeline there will come a time when it doesn't.

they should call it "Moores Average" how hard is it to count transistor increases of current tech and say the last few years was X, so the next few years can also be X

[god_43]

I always chuckle when I see people bring up "Moore's Law" as if it were holy grail to the industry. It isn't and like everything else there will come a point where it's irrelevant.
Don't get me wrong, Mr Moore has my utmost respect and admiration but this "law" was from a certain timeand although technology has moved forward in that timeline there will come a time when it doesn't.
As to the future of computing I wish I was 40 years younger so I could get into the design end myself.
Fascinating business to be in. Challenges the mind to see what can be done.
I look at the guts of a 32nm chip and it boggles my mind to see what is possible.

40yrs younger eh? so that would make you like 60 right?

[zalbard]

Was nice to see you around, god_43...

[god_43]

Was nice to see you around, god_43...

lol i forgot about the banhammer...i plead temporary insanity....



on topic:

i see the future of computing as very bright, and i think visual computing is teh future.