Comments on: Moore's Law limit hit by 2014?

The high cost of semiconductor manufacturing equipment is making continued chipmaking advancements too expensive, threatening Moore's Law, according to iSuppli.

[coachgeorge]

OH NO!!
This would force programmers to write more efficient code!

[mbenedict]

Not really, we'll just rely on ever-increasing amounts of cores. ;-) So by 2014 expect 8-core machines to be the minimum requirement to run word processors without lag.

[ddesy]

I wish we could really expect this. If programmers still wrote efficient code the way they needed to when they were working with computers 20 years ago things could be done in the blink of an eye!

[dennisl59]

And this has been predicted for how many years by knuckleheads? What? 20?

[russkeller]

That's not how I remember it. It's power that doubles. Transistors are irrelevant if new technology like multi-cores hyper-threading ect that come along and change things. Or is this statement just another way of saying "Due to World financial concerns investments in future technologies will be put on hold."

[sythara]

Here is a direct quote:

"The complexity for minimum component costs has increased at a rate of roughly a factor of two per year ... Certainly over the short term this rate can be expected to continue, if not to increase. Over the longer term, the rate of increase is a bit more uncertain, although there is no reason to believe it will not remain nearly constant for at least 10 years. That means by 1975, the number of components per integrated circuit for minimum cost will be 65,000. I believe that such a large circuit can be built on a single wafer."

[fredtheviking]

Human innovation usually put these prediction to shame. But even if this would come to pass, it would not be so bad. It mean that making computers more reliable would add value (I talking about hardware here, not software). In Today's world, sadly making hardware reliable, at least on the PC side makes little sense. Why should computer makers make hardware last for more than 10 years, when the hardware itself 10 years from now would be so weak capable to a new machine.

But if the trend slows-down, having a computer for many years would make sense. Software could catch-up to the hardware and improved. Hardware crafting would improve. People could have computers for years.

But with quatum computers, will probably start this trend again. When data can be store at the atomic level, you only limit is the number of atoms you can fit and contol in a chip. Who knows, in future I suspect they will even figure out how to store information in black holes (the mini-ones that are about the size of an atom). Or some other great innovation.

[Dalkorian]

by fredtheviking June 16, 2009 11:35 AM PDT
Who knows, in future I suspect they will even figure out how to store information in black holes (the mini-ones that are about the size of an atom).

---------------------------------------------------------------------------

Oh great, so when that virus hits your new fista box it brings about the end of the world.

LOL - just kidding, of course. ;-)

[Seaspray0]

@Dalkorian. Respect must be earned. If you can't bother to spell a product correctly, you don't deserve respect.

[santuccie]

Considering that he's posting yet more anti-MS propaganda in a blog that has nothing to do with MS, I don't think he cares about respect. He's living a life of fuming bitterness, a half-life.

[gsigas]

Even if they do reach a minimum feature size limit I imagine they will continue to improve materials, manufacturing, superconductivity, cooling and packaging/wiring so you could end up with really thin stacked interconnected CPU's separated 20nm vertically and still potentially end up with thousands or millions of CPUs in the same space as a current CPU.

[santuccie]

I don't see anything like that happening anytime soon. I don't care how efficient processors are getting; if you sandwich thousands or millions of cores one on top of another, you'll be generating a LOT of heat, heat that has to be dissipated somehow.

[why do i need a name?]

Not surprising.. The economics are making it very difficult to move to new process nodes. Estimates of $30M to bring a 65nm part to market (masks, verification, design, etc). Costs are escalating to the point where 32nm is nearly $80M and going lower is going to cost even more. Sure you can put more on the part, but the design and verification costs are going to kill you.

Think of it this way. If you have to spend $80M on a device and you can make $5 on each one of them (easy for intel, very difficult in the mobile phone market) you have to sell 16M pieces over the life of the part. Given the pace of innovation, that is at most 2 years. And just how many design wins are there out there that give 16M pieces?


That said, the last time we faced this obstacle, it was the cost of bringing a fab on line when everyone owned their own. No one could justify the $B investements. The result: an entire new "fab" industry was born that amortizes those costs over numerous designs. The question now is: will someone come up with a way to turn the fixed design/verification costs into a variable cost like the fab guys did?

[tech_crazy]

Answers that would come to mind are general purpose architectures - either CPU/GPU, FPGAs or some such and then differentiated by firmware/software. ASIC costs are ever increasing. All the same, what used to be entire systems are now ASICs/SoCs, so in that sense the total cost/system (design, verification, masks etc.) has not changed much, albeit the entry level price tag has.

Appreciate you putting out real cost nos. out there but they may not be applicable en-masse. You can make a $5 profit on something that cost close to that. My recent work was on imaging SoCs that would make a profit if sold at 1 $ a piece. And yes, the mobile OEMs were asking for guarantees of us providing 5M+ parts/month. So, it seems that it is not only the cost per chip or only the volume, it is the percentage profit per chip that is key.

[CreativeMalcolm]

And by then maybe Microsoft will have figured out how to help the OS take advantage of more than one core.

[Erictheruler1]

Wow I just shot sunkist through my nose reading that and laughing you are Creative Malcolm hahhahaha Microsoft hahahhahahaha

[santuccie]

It sounds like you're insinuating that someone else already has. Grand Central Dispatch is Apple's first go at it, and that will be introduced in Snow Leopard. But just so you know, it's also coming in Windows 7. And since MS beat Apple to effective implementations of DEP and ASLR (among other things) by a long shot, I won't be mad if Snow Leopard hits store shelves a month before Win 7.

[Seaspray0]

@creativemalcolm. Microsoft has supported multiple processors since NT server. That's 17 years ago. If you can't keep up with the times, don't bother posting garbage like that.

[hmdz105]

I hope by that time we see applications that really take advantage of all those idle cores in a cpu. What ? 2014 ? seems like a short time from now!

[gertruded]

Not a chance. Applications mostly will not take advantage of multiple cores as most programmers can not program for multiple cores. The programming capability of most programmers has decreased significantly in the last twenty years. Creativity in programming has almost disappeared. Can we spell HI-b.

Yes, I programmed and supervised programmers.

[sythara]

Its just like with cars. When you increase quantity you lose the quality. Before, programmers were rare so they had to make efficent code. Now you can hire a graduate from India to write code for dirt cheap, and who cares of no documentation is made, for you can always make a patch that creates a solution on top of the problem rather than eliminating it.

[Been_there_Saw_it_before]

Multi-core programming is alive and well. You just need to look in the right place. Unfortunately, most of those places are where you are not allowed to peek. Wait five years until this filters out into the open.

[hmdz105]

On a windows 32bit system, 90 percent of my programs, at most, use one CPU core. It is silly that the operating system itself cannot cast/distribute the processing tasks between the cores if one application doesn't support multi-threading or multi-core usage.

@gertruded:
You are probably right. Since the invention of high-level programming languages such as Java / C#, programmers no longer needed to write near-assembly efficient code. But they still are required to make their applications as fast as possible to target a larger audience for their software. Look at what happened to Windows Vista; They (MS) are now releasing an OS (Windows 7) that has requirement less or equal to Vista, after 3 years.

@sythara:
That is what we call as Productivity v.s. Efficiency - Though I guess we are getting close to the point to have both together.

[Mr. Dee]

and I here I was hoping for 10 NM. Anyway, there is so much power in todays CPUs for software engineers to exploit and they have not started yet. The hardware seems to be about 5 to 10 years ahead of the software right now. When parallel programming becomes the norm, we should start to see some interesting experiences. Lets not forget about GPGPU too. The aim is to make devices smaller, by 2014, I don't think we will want anything smaller than an iPhone for our PC, anything other than that it will be lost.

[fokkwp]

OH NO!!
What if processors aren't fast enough 20 years from now to handle the software that we'll be writing for them?? FUN FACTS: the first nuclear missiles were guided with mechanical computers, since there were no useful digital ones at the time. Dangerous? you betcha. Feel safer, the faster computers get?

[LaTene_Man]

I don't see it slowing down at all, Moore's Law. We'll just go down a different path, getting our performance increases in another way. Quantum computers anyone?

[Pishkado]

As the article states, the limit on Moore's Law (it should really be called "Moore's Observation;" it's not a law in the sense of a law of physics or anything else) has been predicted any number of times due to something bumping up against some limit. Each time, people figured out a way to get around that limit. This will be no different. They have just identified another limit that we'll have to get around. I'm confident that we will.

[nicmart]

Nicmart's Law: Every two years it will be predicted that Moore's Law, for technical or economic reasons, will cease to be valid.

[santuccie]

I have used AMD processors, and would probably purchase an AMD-powered laptop again for economy's sake, but am not a big fan. That said, the last thing I would want is for AMD to go bankrupt, because Intel's prices would more than likely skyrocket again.

[Sabroson]

There is life after silicon ... how about organic based computing? Chemical based? I even saw a "laser" based computer being developed.

Then there is more efficient programming ... programers are spoiled by very fast computers and are not very efficient when programming ... Apple with it's new Snow Leopard has started to tackle efficiency on multi-core processors ... even using the graphics processor power for other computing tasks ...

[juggleufl]

Moore's Law refers to the capacity of computing power to double every two years. It talked about this in terms of transistors merely because that was the conceivable means of achieving this. Obviously they had no concept of quantum or optic computing at the time. Moore's Law, as it exists, will reach its limit, but only because it's part of a larger exponential curve, and once we reach the point of transitioning away from current computing technology then computing power and potential will grow much more quickly than Moore's Law predicts.

[kerogre256]

They definitely didn't hear about 'graphene' (check wiki) and ballistic transistors.

[petebloke]

I do not see what the issue is, why don't they just change the rule again. When Moore originally stated the "law" that component integration doubled approximately every 12 months, this then changed to 18 months and then 2 years. Moore's law is not so much a law but a sliding rule. Though you have to say it has giv en a lot of PR to both Gordon and Intel ... hmmm, there may be more to this ;o)

[erspark2009]

The larger concern is our ability to make any type of computer chip in the future. We're running out of the basic building blocks to make them. This will make creating any type of chip prohibitively expensive.
"both indium and hafnium - which is increasingly important in computer chips - could be gone by 2017."
<a href="http://www.newscientist.com/article/mg19426051.200-earths-natural-wealth-an-audit.html?full=true">NewScientist:Earth's natural wealth: an audit</a>
Make sure you've got a good pc in 2017, because it might be your last for a while to come. (And make sure you can run it on solar power as well! : )

[sagron]

There are several technologies upcoming that hold promise... quantum computing has been mentioned; there's also standard light-based or laser processors, spintronics, materials like bismuth telleride, graphene, nanotech, etc. - and those are just the ones I've heard of.