Slashdot Mirror
Moore's Law Disputed
Kumiorava writes "Transistors can be packed to same chip two times more in every 18 months. This Moore's law has been repeated already over 30 years. Computers become faster, IT economy grows, but Moore's law doesn't apply. That has been proven by researcher Ilkka Tuomi. You can read the research from First Monday article The Lives and Death of Moore's Law." 'tho, to be fair, it seems to me that Moore's Law has lasted a lot longer then the throng of people who keep predicting its death.
My guess is that the reports of the death of Moore's Law will turn out to be greatly exaggerated.
Pathman, Free (as in GPL) 3D Pac Man
First the 2nd law of thermodynamics fails, then Moore's Law... When will things start fulling upward?
"The present paper argues that Moore's Law has not been a driver in the development of microelectronics or information technology. "
A better title might have been: "Moore's Law - Not All It's Cracked Up To Be"
Work for Change & GET PAID!
It will stop, right? I mean, if the marathon record gets 10 minutes shorter every few years, for example, that doesn't necessarily mean that 100 years from now we'll be running a 20 minute marathon.
Aren't there limits to materials and stuff like that, or do we come up with Infinite Probability Drives, Dimensional Transfunctioners, Flux Capacitors, Heisenberg Compensators, Ludicrous Speeds....
It was never a law (as in operating principle of existence). It was merely a trend in manufacturing. Keen observers could probably make note of similar trends in other industries. I.e. gas mileage of cars, etc.
"I'm not impatient. I just hate waiting." - My Dad
Let me introduce the Slashdot law. This law is inversely proportional to the decline of Moore's law.
Already taking over 60 seconds to load up..
Bad boys rape our young girls but Violet gives willingly.
Lisa, in this house we obey the laws of thermodynamics.
Sorry, couldnt help myself
The level of Vodka in your blood doubles every 18 months
The oft quoted 'Moore's Law' as some have said before, is not in fact a law at all, but instead a theory proposed by Moore based on the economic and technological trends of his time. He by no means meant to imply that this measurement be used as a benchmark of the technology industry. The fact that is is not only known, but hotly debated in the industry shows not the accuracy of the 'law', but instead the success of the marketing campaigns based off that quote. To be quite realistic, some manufacturers have pushed out technology that has not been completely tested in order to compete in the marketing game of Moore's Law, and thus we have cheap, unreliable PC's. (Don't get me wrong, this is only one of many reasons for this effect!)
</RANT>
You can have it fast, accurate, or pretty. Pick any 2.
Moore's Law has never really been a hard and fast law. It's more of a rule of thumb... I've read a few books that mention it, and some of them even disagree on the time period in which the double takes place. Some say a year, while some say 18 months. I've also seen articles which claim as a part of "Moore's Law" that the prices also cut in half.
Defying Moore's Law isn't like defying gravity. We know that at some point, miniturization will no longer be possible. It's hard to double the number of transistors in one space when they're on the atomic level. Do you think we could do that in 18 months?
Uhh, it's not a logarithmic function, it's an exponential function. Exponential functions have no asymptote. Think about what you're saying exactly. Why should the number of transistors ever level off if the function specifies that it DOUBLE every year.
I'm not saying moore's law will last forever, but that's because of the physical limitations, not because the actual function hits an asymptote.
The number of people incorrectly predicting its demise will double every 18 months.
I thought the General version of Moore's Law was "The speed of a computer will double every 18 months or so".
Fine, originally it was "transistors" but I thought that if dual CPUs became a defacto standard in 12 months that would count towards Moore's Law instead of being illegal since the transistors aren't all on the same die.
It just sounds like nit-picking bullshit. I've always thought of Moore's Law as "the IT industry will find a way of doubling computing power every 18 months" not some stupid unit of measure.
Shit, if superior engineering can double computation with the same number of transistors (via better design) shouldn't that count? It just sounds like someone getting into a huff about it and having too much time on their hands to fiddle with Excel.
What is music when you despise all sound?
Every 18 months, computer software will be made to take twice the processing power for the same task.
Another factor is the great disparity between actual processing power (often measured in FLOPS etc) and the number of transistors on a chip. For a while, transistors numbers were doubling every 12 months, but computing power was only doubling every 24 months. Why? The need for pipelining and data management meant more and more of the chip had to be dedicated to pre- and post-processing of the actual calculation, along with intelligent caching and the related works of predictive streams.
An alternative approach has been to build specialized hardware to put all those transistors to use, at the expense of turning your general purpose computer into a very special purpose machine. This has been used, sometimes to great effect, in for example N-body calculations (GRAPE 1-6), yielding 50 or more TFlops of performance for the general computer cost of a 500 GFlop machine. It provides yet another example of the misappropriation of Moore's law.
I think while we may be starting to reach the point that the laws of physics may limit how much faster a CPU can go, don't forget that other parts of the computer are getting major speed boosts, too.
First, there is the connection between chipsets on the motherboard. AMD's Hypertransport and others could make big differences on overall motherboard speed.
Second, system memory speeds are getting quite a bit faster, too. Developments in DDR-SDRAM technology could eventually result in throughput 2-3 times what we have now with DDR333 technology.
Third, expansion slots are getting faster, too. There are now standards upcoming for both PCI and AGP that will substantially increase data throughput on expansion slots.
Fourth, mass storage devices are getting faster, too. IDE hard drives have now reached ATA-133 speed, and future IDE hard drives using the new Serial ATA connection will eventually reach the equivalent of ATA-600 speed! SCSI interface hard drives are benefiting from Ultra 160 and Ultra 320 speeds, too. Even optical recorders are getting faster, too; we've reach 48X speeds for CD-R writers, and DVD recorders will go past 12X speeds some time in 2004.
Fifth, hot-docked external connections are getting faster, too. USB 2.0 support 480 megabits/second connections, and the next-generation of IEEE-1394 connectors will support 800 megabits/second connections.
Finally, graphics cards have seen VERY dramatic performance increases for 3-D graphics. Today's ATI Radeon 9700 Pro and the upcoming nVidia GeForce FX chipset graphics can achieve 3-D rendering that no one could have dreamed of even five years ago.
In short, CPU's will probably reach their limits before 2010 but overall system speed will still increase dramatically thanks to other system components speeding up.
However, many users still want functionality in Linux that closely resemble what you get in Windows XP. Unfortunately that will result in system bloat because of all the multimedia programs, web browsers, etc. you have to include.
The idea of the paper is to show that Moore's law can't be used to predict trends in economics.
So
a) "Moore's law" shows us the effect of demand vs. supply
b) It does not mean that the demand (or demanded quantity) would increase infinitely
c) You can not call it a law because the variations have been too big (first it was one year, then two, now 18 months) and as the formula is that of exponential growth, those variations mean huge differences at the number of transistors over a period of, say, five years.
In short, this article looks at the economics (as in macroeconomics) side of Moore's law. It doesn't claim that you couldn't pack more transistors or whatever on a microchip.
You could also claim that Moore's law might actually hinder economic development as Intel wants to obey the law. What results is that we are actually saying that "wow, Intel is keeping up with the R&D forecasts stated in their company strategy". Yipee.
Okay, a shitty explanation but please read the paper and look at the idea behind it before saying it's total bullshit.
However big, fast and/or powerful your computer is Adobe Photoshop will always take an age to start up.
Avantslash - View Slashdot cleanly on your mobile phone.
The guy isn't a person saying that Moore's law is doomed, he clearly points out that it doesn't exist in the first place. The claims of transistor counts doubling every 18 month, and processing power doubling 18 month, and the like are all historical inaccuracies, that Moore himself didn't claim. He also uses numbers to show that Moore's law has in fact NOT been valid.
It is also shown that Moore's law is often used as an reason by people who don't know better, and those who don't bother to verify their facts. The main point of the article though is that any Moore's law is not the driving force in the IT industry. It all comes to supply and demand. Unlike slashdotters, who seem to like pulling figures out of their ass, this guy actually has real and valid numbers which prove his point.
Before you make rediculous comments, please, RTFA.
The number of illicit, dishonest and monopolistic tactics employed by Microsoft doubles every 18 weeks :-| Have a Day
Ruby on Rails Screencast
"it seems to me that Moore's Law has lasted a lot longer then the throng of people who keep predicting its death."
You could say the same thing about Apple.
'tho, to be fair, it seems to me that Moore's Law has lasted a lot longer then the throng of people who keep predicting its death.
I just saw a throng pass away last week!
That doesn't sound so bad.
I suffer from attention surplus disorder.
Moore's Law is like Moore's love, hard and fast.
Did anyone actually read the damn article?
It's about how the entire concept of Moore's Law is vague and has been applied to all sorts of other things exhibiting exponential growth, even though Moore was not referring to them. And specifically Moore never gave the time frame of "18 months." He said "1 year" one time, then later said "2 years." And if you look at the data, the transistor count of chips doubles roughly every 26 months, not 18. The point of the article is that Moore's Law is more of a hazy myth than anything else.
For a second I thought that the headline read "Murphy's Law Disputed". I was going to argue it bitterly.
The global economy is a great thing until you feel it locally.
Actually Martyn is right. The actual evolution of microchip technology might (and probably will) eventually reach a physical limitation, which could be described as an asymptote.
Moore's Law, on the other hand, is merely a mathematical function, made to predict the evolution of microchip technology, and being an exponential one, it, per definition, does not have an asymptote.
You're falsely assuming that Moore's Law is an absolute reflection of the actual evolution of mcrochips, when it is in fact just a predition (although so far a pretty good one IMHO).
Murphy's Law, now that's a law.
- Health, Fitness, and Weight Loss News
Godwin's law fails!
USENET authorities are disturbed by the failure of a law that some thought to be a lynchpin of internet discussion: Godwin's Law. Simply stated, "As a Usenet discussion grows longer, the probability of a comparison involving Nazis or Hitler approaches one." Beginning last week observers began to notice something was wrong. Says one 'lurker', "I came across this thread on abortion, you see. I started reading--and that's when I noticed something strange. Every post in the thread simply got better and better as each participant read the other's arguments and replied calmly. It was then when it hit me--no Nazi references anywhere. I went back to read it again, and I was sure--Godwin's Law has been broken."
The violation of Godwin's law is hailed by some as a doomsday scenario for USENET. "These threads will just keep going and going forever! There is nothing to stop them. Eventually it'll all just reach critical mass and collapse in on itself," says a popular USENET troll. Others don't see it as Godwin's law fails! USENET authorities are disturbed by the failure of a law that some thought to be a lynchpin of internet discussion: Godwin's Law. Simply stated, "As a Usenet discussion grows longer, the probability of a comparison involving Nazis or Hitler approaches one." Beginning last week observers began to notice something was wrong. Says one 'lurker', "I came across this thread on abortion, you see. I started reading--and that's when I noticed something strange. Every post in the thread simply got better and better as each participant read the other's arguments and replied calmly. It was then when it hit me--no Nazi references anywhere. I went back to read it again, and I was sure--Godwin's Law has been broken." The violation of Godwin's law is hailed by some as a doomsday scenario for USENET. "These threads will just keep going and going forever! There is nothing to stop them. Eventually it'll all just reach critical mass and collapse in on itself," says a popular USENET troll. Others don't see it as cataclyismic, put painful all the same. "World War II is a large part of the world's history--I don't want to see that forgotten," reads one post to alt.military.history., put painful all the same. "World War II is a large part of the world's history--I don't want to see that forgotten," reads one post to alt.military.history.
Looks to me like some jackass with no credibility is trying to make a name for himself by "publishing" a junk article in a "peer-reviewed" online journal by "proving" that Moore's law isn't a fundamental phenomenon. Well, duh. Hell, I wouldn't be surprised if he posted his own article to /.
-Looking for a job as a materials chemist or multivariat
It looks like about 3 people so far, but some read it more carefully than others. Please everybody who is reading this: read this article article because it is very important. Again, though, even people who have read the article (or skimmed it) appear not to have gotten the full message. So Junks Jerzey writes:
It's much worse than that, actually. When he really pulls the gloves off and looks at the hard data over the entire 43-year history of the industry, he finds *no* simple doubling time for almost any measure of interest that has been claimed to be Moore's Law or any folk version of it. Even for transistor counts. What you can sometimes sort of show is iffy exponential fits to the data for 5-10 year periods. Strikingly, though, the doubling rates for several of the measures the author investigates have *slowed*. Improvements do keep on happening, but the pace of the improvement is not as consistent or rapid as you might have expected.
Now the big deal about this is simple. Anybody who tries to project that our problems will be solved when X doubles in Y months is really walking on thin ice. It is also important because chip technology has often been held up as some special and amazing business whose success should be inspirational to us all, since it improves so fast. Clearly, improvements in raw components have been rapid (although not as rapid as you might expect), but the Big Changes caused by technology are rarely tightly coupled to the speed of improvement in underlying technology. Hey, the *big* change of the last decade is that your grandma now probably has email. I'm not sure it makes sense to calculate how many transistors that took.
Babar
but unfortunately it's been, well, you know.
i could live a little longer in this prison
Since it's a rule of thumb and not a law, why should we care if it's *about* to be broken? Let me know when you have 6 months or maybe 2 years of data showing that the cycle is lengthening or shortening. And since we know that progress will eventually double capabilities, isn't the length of the cycle the only thing that can change?
If Moore's Law does official die, will there be sightings of it afterwards like that of Elvis?
You're right, that's true that the title was that way. But is it not also true that if Moore's Law were not being actively met, we would likely have cooler-running, more efficient (yet faster than the previous) processors designed more along the lines of Astro and Crusoe rather than P4 and Athlon? Personally, I'd like it if chipmakers would strive for some efficiency and cooler-running chips. As much as I love Athlons (i run two at home), I'd rather heat my house with the central heating system than with my CPUs. As it is, since I spend all my home-time with the exception of sleep and bathroom time in my computer room anyway, I don't even have to turn on my heater.
In SOVIET RUSSIA... erm...NSA AMERICA, the Internet logs onto YOU!
the point is: Moore's law only applies as long as chip development is treated as an clock speed race rather than as an overall improvement (in more ways than just speed) push.
In SOVIET RUSSIA... erm...NSA AMERICA, the Internet logs onto YOU!
No. Wrong. Sorry, try reading the *whole* article again. The BIG major point of the article, which he point out at the very beginning, by the way, is just this:
Moore's Law has never really existed in any form that is consistent or interesting to us.
It isn't "just" that the doubling times was fudged (although when you're talking about a presumably exponential process a little fudge goes a *long* way). The above bold point really breaks up into three major claims:
Seriously, it *is* a really big deal when an idea as big and as potentially important as Moore's Law turns out to have little or no substance. It is always a rude awakening when you find out that a growth process that appears to be exponential has hit some limit. It may be worse in some ways to find out that not only were you not looking at some coherent or unitary process, but that none of the obvious possibilities really ever seemed to show an exponential growth curve for more than 5 years or so.
I don't think you read this very carefully. I don't think the author cares at all about fundamental phenomena, just whether there is any testable content to various formulations of Moore's Law, and if there is something you can test, do the empirical data fit the law. Very, very embarassingly, (in my opinion) nobody much bothered to do this before, and the actual data lend very little support to any statement more concrete than "technology has improved significantly and rapidly since the invention of the IC".
Babar
That has always been my favorite.
The article, although very long and intense is well-written and very educational in the many interpretations of Moore's Law. A good read.
Relive the BBS Past - One Byte at a Time! www.ssabbs.com
I agree that the law/behavior you suggest occurs. If you read the article, you'll see that this is really a different sort of article however.
The article doesn't say that Moore's law won't continue. It says, and attempts to show empirically, that the ill-defined Moore's Law never really was in effect to begin with; that the data in many cases doesn't really support Moore's Law(!) This is a new and distinctly different sort of claim.
--LP
P.S. I hate to bitch. Well, not always. But sigh: "2002-12-14 19:29:50 Moore's Law: the data doesn't fit (articles,hardware) (rejected)"
Moore's Law, Moore's Law, Moore's Law...Christmas!
What "Law"? How about "Postulate", or "Theorem"...no, not theorem...how about "conjecture"? How about "cockamamie bullshit"? You could probably make a similar "law" that describes the performance of light bulb technology over the last 100 years..."well, lightbulbs become (sort-of) 5% less expensive to make and 5% brighter every decade! Whoopee!!!"
Everybody's seen the graph. It's not linear. It's not exponential...it's just up. Hit or miss. No "law" involved here at all.
The whole idea that Moore's Law is a Law is stupid from the get-go. Damnit, I wish I could remember the name of the...oh yeah, the IgNobel. They should give the IgNobel to the cat who disproved Moore's Law. I mean, come on people, duh!
This is almost as stupid as those clone-aid wackos...
It's really just "Moores Semi-Accurate Observation That We Can Use To Help Figure Out How Fast Things Are Changing".
Moore's Law has never really existed in any form that is consistent or interesting to us.
Right...but since nothing else was ever claimed for Moore's law by anyone with intelligence, I hardly see the point. Yes, I read the article. Yes, what you say is right. Moore's law has never been strictly correct. I'm kind of surprised you thought otherwise.
Hell, it's never been a law, in that there is no fundamental, scientific *reason* for there to be *any* link between the number of transistors on a chip, processing power, or whatever, and time. Intel *could* have ratcheted up the doubling times if they wanted, say in response to competition. Like what's happened in the last ~4 years thanks to AMD. That alone should have made it obvious that Moore's law is bunk.
Very, very embarassingly, (in my opinion) nobody much bothered to do this before, and the actual data lend very little support to any statement more concrete than "technology has improved significantly and rapidly since the invention of the IC".
To me, that's like saying it's embarassing that no one has ever done a test to prove that concrete is harder than styrofoam. No one bothered because it's so trivially obvious. The only people who considered Moore's law to be anything but a marketing construct over the last 30+ years are journalists, most of whom have no tech training.
It is always a rude awakening when you find out that a growth process that appears to be exponential has hit some limit.
Now, *that* wasn't in the article. He just proved that Moore's law never really had a point. He gives *no* technical reason why whatever validity it has now will cease to be. Nothing regarding power consumption/loss, tunnelling across junctions, etc. In fact, I saw nothing technical in the "article" whatever. Partially, that's fitting, since Moore's "law" isn't technical. But for the claim it has some technical, fundamental limit, such proof is needed.
So I'll stay with my original point - this article used 10 pages to prove the mundane. Also,what most people will assume the article proved wasn't in the article at all.
-Looking for a job as a materials chemist or multivariat
This question has been puzzling me for awhile... because, unlike other 'laws' that have fallen into disfavor, we never even expected Moore's "law" to ring true for even a relatively short duration. That we would call this set of observations a law in the first place strikes me as odd, considering that it's expression is dependent on so many other socio-economic factors.
>Why should the number of transistors ever level off if the function specifies that it DOUBLE every year.
The function doesn't take real world problems into account.
Eventually the size of transistors will reach a near molecular level and be too expensive or impossible to make any smaller. (we are no where near this point yet)
OR
Eventually the transistors will be small enough for an arc to bridge them, even at low voltages. Then it goes from being a transistor to being bridged. This isn't good for logic circuits: )
These are two good reasons why the number of transistors you can squeeze into a given area of real estate is finite.
Of course, you can simply make the die bigger and lower the voltage if necessary. Even this has practical limitations.
Problem:
heat becomes an issue and the wires need to get bigger as the current rises (due to lower voltage and the higher current that results).
While the size of the die is not limited, eventually to keep up with moores law, the chip would get too big to be practical once the transistor minimum size limit is reached, and a couple of generations of the device had passed.
Don't take my word for it, I am not an EE or computer scientist. I am simply a professional programmer/hobbyist (with an electronics background) who likes to read a lot.
While we may not run into these issues in the next ten years, or even in your lifetime, it is a mathematical certainty that we will eventually get there. This is the fundamental problem with moores law: transistors can only get so small.
Of course by the time we reach this point, we will have found a better control device than the transistor, and a better logic device than computers and chips as we know them today.
Biocomputing comes to mind(no pun intended).
l8,
AC
"The number of /. readers that don't read the article will double every 1 to 2 years"
The Kruger Dunning explains most post on
Yes, and no.
If you only load those programs when the users wants to use them, you will help minimize system bloat, and if theuser wants all that stuff, then bloat would not matter.
Of course for me bloat has always meant how tight the code is, not its size. Clearly the more you want to do, the larger the code will be.
The Kruger Dunning explains most post on
Intel itself has already said that Moore's law is over, explained in slashdot here. Of course, other people are always predicting the end of it as well. Then again, some people think it will continue.
I really wish people would get over Moore's prediction and talk about relevant stuff. There is no way to predict how long unknown scientific breakthroughs will allow Moore's Prediction to remain true. There is one absolute though, the end will come some day, you can only store so many atoms in a certain amount of space according to the rules of quantum physics - that is the absolute barrier.
Until it is actually abandoned I could do without hearing more of Moore's law.
Don't mess with mrs. Moore.
The Kruger Dunning explains most post on
There are more factors to Moore's Law then just how many transistors can be put on piece of silicon. It's also based on how fast:
* Motherboards are created to use the new higher speed processor
* RAM bandwidth (compare SIMMs performance to DIMMs)
* Software that utilizes the newer/more advanced features (8bit vs 16bit vs 32bit vs 64bit)
* Additional load placed on a processor by a new GUI's look feel (eye candy slows your machine down)
* Lack of advancement of storage devices (slow drives = slow machine)
* Lack of advancement of IO ports and devices (My ISA video card isnâ(TM)t as fast as my PCI card)
Processors do get faster. Nevertheless, other factors limit them.
Try this. Compare the startup time of a fresh install of Windows 95c to a fresh install of Windows XP on the same hardware. You will find that the 95 system is much faster then the XP startup because there isn't nearly as much OS baggage slowing it down.
You say things that offend me and I can deal with it. Can you?
What's most amazing is how many people believe that their boom/bust cycle is the only one - I see at least five in that graph :-)
"It's not that transistors use less material, but that the atoms themselves in the material shrink!" declares Special Agent Fox Mulder, expert in conspiracy theories.
"For eons we've wondered how come Dinosaurs were so much larger than modern mammals, but it's because the closer you get to the Big Bang, the largers those atoms were. I have something in my pocket that will astonish you..."
Agent Mulder removes from his pocket an atom the size of a tennis ball. "This is an atom from the Dawn of Time itself. The Al Queda has been trying to get there hands on this puppy, because you can split it with a butter knife."
(Portions of this post were lifted from a bit of Fan Video called "The Fed-EX Files" produced by a film crew in Montreal, Canada.)
There were all these irregular looking graphs. But when I looked at them they had axii with titles like
... difficult to interpret. It could be sloppy labeling, but I'm afraid that after looking at a couple of those graphs I estimated that the article wasn't worth reading.
"average cost of chips" by year.
This is great is you want to guess what people are paying, less good if you are trying to estimate what they are buying. "Cost per chip" without identifying the chip is
I think we've pushed this "anyone can grow up to be president" thing too far.
Seriously, it *is* a really big deal when an idea as big and as potentially important as Moore's Law turns out to have little or no substance.
Is this a joke? Moore's law isn't E or the speed of sound: It's a general hypothesis about the rate of technological progress. No one expects there to be an absolute correlation, and really any correlation that there has been has largely been perceived as humorous in the context of the "law" (it isn't a "law", of course, but is rather an "observation").
Should we go back and re-engineer all of the processors because of this amazing new research into Moore's Law?
Used to be that the stock price of MicroSoft stock doubled every two years, hence a split. Bill Gates would get ten times richer every five years, with predictions of hime becoming the first trillionaire sometime in the first decade of the 21st century.
Well, MSFT has been stagnant for the past four years. Bill gave over a third to charity and he's been stuck at $30-$40 billion for a while.
The author messes up by paying too much attention to the constant: that is, whether the doubling time is 18 months, 2 years, or some other number. He also worries too much about whether it's an exact exponential or not. It's not. So what? The most amazing thing is that a doubling time exists, meaning that we have exponential growth.
Moore's Law should be read as saying that various measures of transistor density on chips grows as O(exp(t)); this has held for 40 years. Of course, no exponential growth can continue forever.
Much of the recent history of the electronics industry has consisted of treating Moore's Law like a human law, that is, it is the marching order for the entire industry. Everyone from the fabs to the electronic design software houses to the microprocessor manufacturers to the systems houses plans in terms of generations of exponentially increasing density. Even the computer science notion "all problems can be solved by adding an extra level of indirection" implicitly assume that since the processors are getting faster all the time, we can make the code slower if we get more function out of it.
Keeping this exponential scaling process going is a massive undertaking; those interested in the problems at the cutting edge might want to look at the International Technology Roadmap for Semiconductors.
In any case, Moore's law is doomed in the long term. I think it's got another decade or so of life, though, as the researchers have a pretty good handle on the next couple of generations of scaling.
At one point, Intel actually could do this. They could literally pace their processors so that processor speed would "fit" with software speeds so you'd need a "fast" (and expensive) processor. That pretty much was the situation up to '99, when AMD released their Athlon. Since then processors have been speeding ahead of software due to competition. If that hadn't come I think we'd probably hit the GHz barrier about now, certainly not 3GHz.
Kjella
Live today, because you never know what tomorrow brings
How can one deal with this computers on a regular and continuing basis and still come out with a jack ass statement like:
Moore's Law has never really existed in any form that is consistent or interesting to us.
Pardon me? When new procesors are announced do they have performance increases that are proportional to their existing speed or are they some lesser quantity. For example when 300 MHz processors were current the next interesting products were hundreds of MHz faster. When processors were 1 MHz (the original 6502 in the Apple II) the improvements were 2 and 4 MHz. Now with GHz processors we look at new generations measured in multiples of existing speeds, not a few MHz or even hundreds of MHz faster.
A similar story is true of memory density. The original Mac had 64K of RAM. Each generation wasn't just larger, it went up by an order of magnitude. Now we can easily afford and use a gigabyte of memory.
Hard drives started around 5 to 10 megabytes. The build up to current 100's of gigabytes of storage (all at static price levels) was obviously exponential.
Just how dim do you have to be to miss the fact that fixed price improvements come as multiples of existing levels which is what defines exponential growth which is the actual content of Moore's Law?
Did I read the article carefully? No, because I quickly reached the conclusion that the author was a moron with the ignorance that is only obtained via studying for an advanced degree in one of the less rigorous academic fields. If you are looking for insight read some of the observations of Carver Mead who obtained the fundamental insight back in the 60's and helped to create the phenomenon that has transformed the world (as dimensions get smaller almost every performance aspect improves so we are on a roller coaster until we butt up against fundamental limitations which themselves seem to recede as ever more human ingenuity is applied to every detail).
When I was an undergrad I worked for a physics professor whose research included experimental measurements that pertained to general relativity. He showed me the archives of the kooks that are invariably drawn to the challenge of refuting general relatiivity. He was always getting these silly things in the mail and few of the authors had any sort of understanding of physics at any level. The same seems to be true of Moore's Law today. But in this case you can't have a requirement that a disputant at least have some knowledge of differential geometry. You just have to be able to spell the names.
There are both vertical and horizontal asymptotes. Look at the Sine, Cosine, and Tangent graphs.
:-}
In the case of an exponential function you have slow growth, acceleration, then a rush towards infinity. That final part is the vertical asymptote as you cannot progress further along the horizontal axis.
Ouch. My brain is now full of dust from those old math books...
--- I wish I could hear the soundtrack to my life. That way I'd know when to duck.
The author of the article seems to misunderstand completely the intent of Moore's law. The article notes a few things:
...Moore's law was always intended as a rough rule of thumb that applies relatively well over a long period of time. If anything, the article buttresses Moore's law. The article notes that the original micoprocessor in 1975 had 2,500 transistors, and that the P4 has ~40 million. If we assume a doubling time of 2 years, then Moore's law was substantially correct, within a 10% margin of error. This was far more accurate than I was expecting, and far more accurate than Moore was expecting.
1. Increases in transistor count do not precisely follow an exact, continuous, exponential mathematical function. Some years it grows faster, others slower, etc. WELL FUCKING DUH. The article seriously thinks this is original and insightful, but actually it was known to everyone. OBVIOUSLY, Intel releases new processor architectures on some years but not others, therefore the increase in transistor count will be faster on those years and slower on others.
2. A few journalists have misrepresented Moore's law, by publishing versions that were not identical with what Moore actually said. AMAZING. A journalist misquotes, or misunderstands a technical issue? Who would have thought it possible? I'm glad we have this article to expose such shocking truths.
Well put... Enough squabbling over this 'it's a bad paper'/'it's a good paper' nonsense. The paper correctly points out that Moore's law is poorly named. To most people this doesn't really matter. Your point - O{exp{t}} - is a good one. It should be noted though, and the paper seems to miss this, that the real take home message of Moore's law isn't that there will be a single exponential curve stretching back to the dawn of time, its that computer power (measured in an ever changing way) seems to double every couple years. This is useful if you take current circumstances into account to plan for the next few years.
Yes, Moore himself seemed to use an ever changing time constant and an ever changing metric of what was doubling, but it was being used as a way of predicting marketing trends. (it's the cost-optimum chip's power, now it's the total number of elements used in the cost-optimal chip, now its the feasible memory size, its the processor speed, its the transistor count, now we take into account the mythical "engineering cleverness" etc.) The problem is that the people the rule of thumb was created for (marketing folks) took the silly thing too seriously. It should be, and usually is, treated as a flexible rule of thumb. There will be periodic sweeping advances. There will be periodic lulls. It's not like it would be at all practical to use Moore's law to predict how long it would take to increase computer power by 10%, it is entirely dominated by local variations, or to predict the time for a 10x boost in power to the exact month. As long as a few months and as soon as three years are reasonable answers, trying to nail it down exactly is certain folly. Which seems to be the trap that the paper authors fall into.
If the paper helps a few people to set a project deadline realistically great, otherwise, it's just sort of a re-statement of the obvious. Think back 4*18mo (6yrs). Taking into account drive speeds, memory speeds, memory size, processor speeds, processor architecture, bus-speeds, addition of GPU's etc. do you really think that today's computers are only 16x faster than they were in 1997? Given the nebulousness of the definition (its cast as a limit in the paper, not the average) does it count as a doubling if someone decides to tie 2x as many processors together to make a more complex cluster? Alternatively, would you believe me if I said that computers would double in power sometime in the next 3 years? I get to make up the definition of double in power three years from now though. For those saying that people are basing multi-billion dollar decisions based on Moore's law... crap... I hope you're wrong.
If not, then in reference to a comment above, maybe we ought to give the hampsters a chance.
I frankly don't care to get into the discussion of whether you should have called it a "Law", "Hypothesis", "Theory", "Observation", "Note", or anything else. In any case, what you would expect is that there was some concrete statement being made about technological progress (i.e., "X doubles in Y months") and that if you checked it out, you would find basic agreement of fact with the statement. As it turns out, you find that the history behind Moore's Law is really murky, and the fits to data aren't really even close for any concrete statement. Now, this would be okay if, as you say, people really did suggest this was all just a joke. But I'm not sure that it was, and, if it was just a joke, I can assure you that way too many people took it pretty seriously without checking it out. You see, the problem goes beyond whether or not the doubling time is 11 months or 42 months or what have you. The point more seriously is that the existence of something like Moore's law, which was the observation of an exponential growth process over the course of decades, gives people license to believe (and they have) that technology really will always be there to step up, or (alternatively) that (chip) technology provides a model for what other fields could hope to accomplish if only they could {fill in the nostrum here}. I find the fact that there is not now nor was there ever any big reason to believe that anything like Moore's Law fit tech data of any kind to be...striking.
Or, let me put it another way. Worldcom almost brought the entire telecomm industry to its knees by playing a similar stunt, suggesting that lots of serious investment in infrastructure was needed because the growth rate of the internet was such that the number of {bits, packets, whatever} transmitted was doubling every {small time period}. Billions in capital were essentially lost because nobody poked around hard enough to find that the data Worldcom used to support their case covered about six months of UUNet growth in the mid-90s. Why were people so completely fooled? I'd suggest part of the reason was that we had been accustomed to the notion that technological progress could be exponential, since, hey, hadn't this guy Gordon Moore predicted that the number of transistors you could put into a given area of silicon would double every 18 months?
Babar
Sometimes you make the next step by pushing the same limits you did before (either on your own or using equipment other people are making to keep up with the Joneses in their part of the industry), and sometimes by finding a different way to do things. For instance, even if you can't make transistors much smaller in two dimensions, but at some point people will figure out ways to go 3D, stacking a bunch of transistors vertically. That turns out to have annoying manufacturing and heat dissipation issues, but huge speed gains if you do it right, because distances become much shorter.
The fun change that's happened recently has been the recent jumps in disk drive price and performance - it's been more like quadrupling instead of doubling the last couple of years, and neither Bill Gates nor Linus have managed to fill it (though Linux can build file systems well enough, and Grandson-of-Napster can give you things to fill it with...)
Bill Stewart
New Fast-Compression-only CPR http://preview.tinyurl.com/dy575ks
Within a decade, that technology hits a wall - atoms and electrons are too big. That's the ultimate limit for photolithography on flat silicon. We may hit a fab limit, a device physics limit, or a power disspipation limit before that. Right now, power looks like the limiting factor. We're headed for hundreds of amps at fractions of a volt going into physically small ICs. Heat dissipation per unit area is approaching levels normally associated with cooking equipment. But somebody may find a way to get power dissipation down; it's been done before.
Even after the size limit is reached, it may be possible to push on cost. IC cost per unit area has increased over time as fabs became more expensive. New fab technologies, or improvements to existing ones, might improve the situation. It's of course possible to build physically bigger parts, as well. (Wafer-scale integration turned out to be a dead end. You can make a RAM chip several inches across, and it's been done. But the chip, plus its massive stiffener, is bigger, more expensive, and harder to cool than the current packaging systems.)
Alternative IC technologies are possible, but none of them seem to provide a lower cost per gate. Gallium is too rare. 3D layering doesn't bring cost down and makes cooling harder. Quantum computing is a long way from the desktop. Nanotechnology is still vaporware. Some of these technologies may eventually work, but to keep digital logic on the Moore's Law curve, they'd have to be further along than they are now.
It's much like aircraft, circa 1970. Aviation people were talking about bigger supersonic transports, hypersonic transports, suborbital ballistic transports, and large VTOL craft as near-term possibilities. None of them were feasible. 30 years later, aircraft are about like they were in 1970.
We're going to see a slowdown in IC progress within a decade.
Anything that can go wrong will go wrong.
vs
Number of components in chip dubble each year
Eq
The number of defects dubbles each year.. errr 18 months. Is it still 18? Murphys hangling for 6? I see..
I don't actually exist.
On the contrary, Grab's Law states that any new PC CPU you can buy is adequate to run a PC for normal home use.
Grab.
Absolutely; and we know what happens next. Somebody comes up with a new and improved process or a better substrate and, boom, another round of improvement. Then the same thing happens N months down the road. Now, I think the fact that many people are missing here is that the *perception* of what is going on here can be different from the reality. The perception is that the people claiming limits we're about to hit will stop Moore's Law dead are seen as Chicken Littles who are then steam-rolled by the onslaught of new technology. In other words, the naive view is that Moore's Law is repeatedly tested, and passes the test each time. The truth is, people have such vague and inconsistent notions of what Moore's Law really is that there is no real test. An even more striking truth, made in the article quite clearly, is that there is no long-term exponential growth seen in any more concrete formulation of Moore's Law. I think some people would be surprised that Moore's Law isn't right even for the specific original case of the number of components on a minimum cost per component silicon chip.
I remember those days, too. To my eternal chagrin, I even believed some !@#$!@#$ pundit and bought a 486/33 system on the premise that faster systems would be unstable and that chip prices would go up and stay up. Oops... Now one point here is that Moore's Law (called that) seems to be an idea that took off in the 80s, and instantaneously morphed into a bunch of non-equivalent and inconsistent statements. Even the transistor version is not the original one! And the computing power version (which the article covers in detail, by the way) is a real stretch.
See my reply above; I really do think many or most non-engineers do have the kind of magical thinking you suggest is ridiculous. In particular, it is the fact the constant cries of the death of Moore's Law *seem* to be just cries of "wolf" that make the "truth" of Moore's Law appear stronger to many people, particularly including the pundit class (e.g., check out the writings of Cringely) and also to many economists (see the paper for more cites). Now the real news here is that when you do look carefully at any concrete data set, you do *not* see an exponential growth curve. In other words, Moore's Law has not really been tested since it was never really true at all.
Now, this is not to say that we will not continue to have progress, or that progress will not be at some times fairly rapid (but note that progress to date has been far less smooth than many people assume). I think it is important to note that it does *not* make sense to make assumptions about the future rate of progress.
I'll explain the segue more carefully below. But please read the article in question; I do not really believe your contention that Gordon Moore was trying to be either "somewhat humorous" or "loose". In the presentations he made on the subject, there were real graphs with real data; in the years after, he has made a point of correcting people's statements about the law (e.g., he never said anything doubles in "18 months" but rather first "1 year" and then later "2 years". I think he is whimsical about it being called a law and named after him.
OK, there are two points going on here. One is that Worldcom, the company, was a corrupt and fraud-ridden organization whose stock price went way up in a bubble economy. And you're right, that part has nothing to do with my argument. :-)
That said, there were huge numbers of companies, old and new, that suddenly saw what they believed to be a nearly limitless opportunity to build out national (and global) optical fiber networks, and (separately) others saw a huge market to work on "last mile" solutions. Now, the only way the huge fiber roll-outs would make sense is if internet traffic was increasing really, really fast. In particular, these business plans were toast unless there was a period of fast exponential growth. Worldcomm certainly did make this claim. You could see this in a number of places, but let's go for this one in Worldcomm's 1999 annual report:
Interestingly, that was a true statement at one time...but I believe only for a six-month period in about 1994 or 1995. The doubling rate has slowed way down now (which means, of course, that we are not talking about an exponential growth process, and probably never were). But, as they say, the rest is history. Now I claim the connection between Moore's Law and internet traffice (or networks) was made early and often, and you can google up the number of hits for "moore's law" and any of "optical fiber", "internet" or "last mile" and get thousands of hits. Try it and see. (Don't follow any of the links if you have a weak stomache...)
Now, wasn't this all particularly witless? Sure it was, since nobody much bothered to pay attention to the actual traffic data and say "wait a minute...". But, irrational though it may sound, people thought they had a previous piece of empirical truth about the *possibility* of exponential growth in technology, namely Moore's Law, which seemed to hold true (NOT!) for 40 years. Heck, you could even be "sophisticated" about it and claim that the fastest doublings would only hold for a period of 5 years...but of course it didn't come close to that. That's the problem with assuming an exponential.
Anyway, I hope I've cleared some stuff up. People have big troubles reasoning with exponential growth scenarios. They underestimate how badly they are being hurt by credit card debt, but do not appreciate how rare true (even temporary!) exponential growth is in many other situations.
Babar
Let's go back to the points from the original paper (as layed out it this one). In 1965, putting planar circuits together was in its infancy, and Moore correctly asserted that there was a lot of room for improvement without even doing anything fancy. Later as certain limits were approached, even these were pushed much further back. The paper is correct in pointing out that the original paper doesn't factor in things like design cost, complexity and other effects. With the exception of memory technologies, modern devices are more and more dominated by interconnect. Wires are components too, even if they are pretty much passive. Moore even makes the important point that costs are dominated by packaging and external interconnect, particularly at the lower intergration levels at that time. These costs are still very significant if no longer dominant in all cases.
I would assert that whenever a new information technology emerges, it sparks a new and more rapid period of near exponential change, both in society and in human intellectual advances as well. In evolutionary time scales, WWII to the present (the whole history of electronic computing fits in this period) is a blink of an eye. In the fossil record, it is impossible to sort out changes that happen at that scale, so it is just about impossible to sort out the internal structure of how it happens.
With modern technology, we have records that give a pretty accurate picture of the evolutionary process in action. Changes is discontinuous; a new process emerges and there is rapid progress as the process undergoes variation and selection at a rapid rate, then the limits of that process is reached and things may settle down for a while. Or the new position may spark additional new processes of a fundamentally different nature and continue the progress for a few more technological generations.
We know that we have only scratched the surface of what is possible with information technology, even if chip technology stagnates, which it won't for a while yet (to 2016 at least according to the experts, which is suspect is the limit of their willingness to attempt to map it, not the possibilities or even the most likely path).
The really frustrating thing about this paper is that it comes very close to making some very good points, but then just drops them. From here on out, technological progress is going to be all about managing extreme complexity, and finding appropriate uses for technology. It is clear that we will be able to design technology to accomplish just about any task we set our collective minds to, and the more important questions become why and how (socially, more than technically).
Without going into all the arguments about why, I will assert that it is the power of sharing ideas and information as embodied in the Free/Open Software movement that has the potential to lead the way.
BTW, in this line of argument is also the best way to refute all the MS FUD on the subject of the GPL. It's all about doing more for less. If you build it they will come, although it may take a little longer than first thought. For simple circuits, you can ignore the design/complexity costs, and nobody has a monopoly on good ideas, so exponential growth is a natural consequence. Things get bottled up when one player tries to monopolize a particular idea (or expression of one). OSS commoditizes software solutions for well known problems, and having a large library of off the shelf solutions to build on is the only way to extend our reach significantly. The control tactics of MS and companies like that are holding back progress, not the other way around as they are now trying to assert.
LOL!
Funny is even funnier when the moderators don't get it.
Hint, to those who modded the parent down: "Moore's" sounds like "S'mores".
Liberty uber alles.