One day it will end. Planes today aren't any faster than they were 30 years ago, if not slower. There are plenty of industries that have had rapid growth which ended.
They're already running into the physical constraints imposed by the speed of light.
At a clock speed of 3GHz or 3 x 10^9 cycles per second, an electric signal travelling at or near to the speed of light (3 * 10^8 metres per second) along a wire can only travel 10cm in one cycle. You have to start looking at multiprocessors and parallel computing to speed things up.
Look at the 'slowing down' of CPU clock speed increases
1981 PC - 5 MHz my first PC - 25 MHz around 1991. 1995 - Pentium 100MHz 2000 - AMD 1GHz 2002 - P4 3Ghz 2011 - 8 GHz, but an AMD special with nitrogen cooling that you can't buy in the shops. Fastest clock speeds you can buy are around 3.6 GHz - hardly any change from 2002.
Would love your hear your investment implications.
I think even speed did collapse we will see more ambitious progress in distributed core computing allowing computing per se (in flops for example) to increase. We may not reach supercomputer speeds in our desktops and mobiles but it can go up significantly.
So the hardware part will ease up and distribution part will take hold.
It's not all about clock speed. In actual fact, applications these days aren't demanding in terms of hardware, with the exceptions of latest games and things like CAD software. And on aggregate, it is difficult to see where computerization of our world will end, but we have a long way to go. PCs sales may be flat, but tablets, phones are growing, cars have up to 30% of electronic components.. But inserting microchips into other white goods is only starting, in toys too, in clothes etc.. And 3d printing could also push adoption further. Of course, all these things will make it hard for existing chip behemoths to continue being profitable as they have been because these trends imply disruptions.
The law is named after Intel co-founder Gordon E. Moore, who described the trend in his 1965 paper.[2][3][4] The paper noted that the number of components in integrated circuits had doubled every year from the invention of the integrated circuit in 1958 until 1965 and predicted that the trend would continue "for at least ten years".[5]
There was always the atomic/quantum limit looming. But I am very impressed by the engineering ingenuity involved in the march towards 22nm on such a regular pace as if it were a after dinner stroll. Also interesting that Moores law stalls around the same time that baby boomers enter retirement. Surely they are pure coincidences. Yet the implication for growth can't be overstated. There are things in the world that are very hard to predict. There are others that are inevitable even if only in the distant future.
Only a limited set of users will need any more processing power. I'm writing this on a secondhand laptop that I've been using the last three years. For most of my computer use this vintage Compaq nc6220 is sufficient. It was probably retired by some big corporation because it has number tag 27692. I can even program python on it and do sound editing. Probably it will harm the cloud computing centers that need massive power.
A discussion is going on about this on slashdot: http://tech.slashdot.org/story/12/07/12/1737237/pc-sales-are-flat-lining
Re whitewashed windows and vacant stores : there's a real balance there.
a) internet means you can compare prices, get best deal...
b) but the item's got to go in a wagon to a distribution centre then in a van to your house. All costs money - I wonder some retailers are making anything from online sales.
There's a petrol/labour price at which you might do better shopping once a week in town. Now if I could just work out what it was ...
As Dan said Moore's Law was never really about the clock speed but about the number of components (gates) you can get per given area of Silicon. Since the increase in number of gates has been down to the ever-shrinking geometries we have, as a kind-of bonus, also got increasing speed.
Since e.g. 2002 the processes' ``feature size'' has come down from 180nm to 22nm.
The trouble is that at some point, about the 65nm -> 40nm transition, the power consumed suddenly shot skywards due to increasing leakage; and this increased leakage is independent of clock speed.
Since power also increases as the speed increases there's a max speed beyond which the chip gets too hot to function and an increasingly large part of the temperature ``budget'' is being used up by the leakage currents.
While the demand for raw processor speed may plateau, there'll be a continued push for greater processor efficiency (as AAPL is doing now with their iPhone/iPad processors). Given the large real world demand for this (longer battery life, thinner packages with lower operating temp chips) there's still large pricing power for such efficiency improvements (much like Boeing and Airbus being able to sell boatloads of marginally more efficient airplanes). Intel has more competition on this front, but it'd be at worst a rearranging of that sector for at least another decade or so. After that the processor side may well "decontent" and lose the ability to differentiate between products, which would be a wholesale change.
Further downstream AAPL is a "post Moore's Law" company - their big profit driving devices (iPad, iPhone) deemphasize raw power and put an effort on integrated package. The demand for upgrades is as much fashion as upgraded capabilities. This is even more apparent on the PC side where people and enterprise just don't care to upgrade much anymore, certainly not like 10-20 years ago where there were real improvements in even the span of a year.
The market valuations of players somewhat reflect this, and it would seem practically that Moore's Law already broke. Which doesn't matter because Moore's law mattered more to the press and analysts than anyone technical.
Even if Moore's law stopped tomorrow we can probably get another factor of 10 to 100 improvement in cost per flop for compute intensive programs compared to current leading desktop chips through better chip design and programming.
we are already there. The implications of the stop of Moores law is that 1) Alternate designs would carry more weight (rise of the foundry/ASIC house) 2) Software would be much more valuable than the underlying CPU/hardware.
Both of these are now demonstrably true. The relative value add of hardware is small. Even in mobile, the mobile CPU itself is a small fraction of the total power demand, and a sufficient mobile experience is now cheap.
All the important factors in Moores law are done. We are all jus twaiting on ~4 years for DRAM to enter the same state.
MY investment thoughts: DOWN on the major equipment vendors, which are essentially investors in the 'churn' of technology.
There is lots of money in the electronics business. Right now, apple makes all of it....that won't last, and renogotiation on the part of the major manufacturers will happen.
Moore's Law is alive and well, but old software can not automatically capture the gains. There is a shift underway but it is "under the hood" and only in applications where performance matters. As that happens, the dividends from Moore's law are being paid out as decreased power consumption. Not a bad thing.
BTW, Moore's Law is a factor of 100x per decade. Any graph of CPU performance will show a faster trend from 1994-1997 and a blow off in 1998-2001. Long term, we are on track and I am 100% certain we will get our lunch this decade.
Luckily, now that we have tablets and smartphones, moore's law can start all over again!
As I type this on my laptop, the main frustrations are not with my processor speed but rather with slow internet and inefficient programs (and r/w speeds), so I doubt even a diminished moore's law would really effect my life.
This seems relevant: http://www.nature.com/news/a-giant-bid-to-etch-tiny-circuits-1.11060
"“The situation has changed from, ‘Can we make it work?’ to ‘We have to make it work’,” says Stefan Wurm, director of lithography at Sematech, a semiconductor R&D consortium in Albany, New York. ... ASML has just a few years to reach its goal. Intel says that it can use the current technology for two more generations of microchips, but only by adding expensive patterning steps and buying more lithography machines. Benschop admits that EUV technology should have been ready “yesterday”."
I've been thinking about this for a while (daydreaming comes naturally) and it hit home a little more when I was driving the other day. There seems to be a somewhat analogous situation in the case of cars, where rapid development of internal combustion engines has resulted in significant advances in torque/horsepower which are arguably hitting up against the limits of human cognition and reaction times (case in point: try having all the world driving a Ferrari F12 Berlinetta which is 544kW and 690 N.m and still maintaing a semblance of road safety. Many consumer cars today are more powerful than Ferraris of 20 years ago).
I think in our lifetime we will start to see real questions asked about the virtues of some technical advancements and how these are able to be applied successfully by humans - consider the KCG debacle from the past few days, which has in large part come about as a result of cheaper CPU cycles. I don't want to sound tin foil hat but we're letting some technology genies out of a bottle, so the advancement of Moore's law in infinitum may start to bang up against some human cognition limitations.
nb: Thanks for a great blog, hope this stream of consciousness post at least made some sense.
The content contained in this blog represents the opinions of Mr. Hempton. You should assume Mr. Hempton and his affiliates have positions in the securities discussed in this blog, and such beneficial ownership can create a conflict of interest regarding the objectivity of this blog. Statements in the blog are not guarantees of future performance and are subject to certain risks, uncertainties and other factors. Certain information in this blog concerning economic trends and performance is based on or derived from information provided by third-party sources. Mr. Hempton does not guarantee the accuracy of such information and has not independently verified the accuracy or completeness of such information or the assumptions on which such information is based. Such information may change after it is posted and Mr. Hempton is not obligated to, and may not, update it. The commentary in this blog in no way constitutes a solicitation of business, an offer of a security or a solicitation to purchase a security, or investment advice. In fact, it should not be relied upon in making investment decisions, ever. It is intended solely for the entertainment of the reader, and the author. In particular this blog is not directed for investment purposes at US Persons.
31 comments:
Now you know how believers in peak oil feel every day.
One day it will end. Planes today aren't any faster than they were 30 years ago, if not slower. There are plenty of industries that have had rapid growth which ended.
and I dreamt of main street's whitewashed windows and vacant stores:
http://www.slate.com/articles/business/small_business/2012/07/amazon_same_day_delivery_how_the_e_commerce_giant_will_destroy_local_retail_.html
They're already running into the physical constraints imposed by the speed of light.
At a clock speed of 3GHz or 3 x 10^9 cycles per second, an electric signal travelling at or near to the speed of light (3 * 10^8 metres per second) along a wire can only travel 10cm in one cycle. You have to start looking at multiprocessors and parallel computing to speed things up.
Look at the 'slowing down' of CPU clock speed increases
1981 PC - 5 MHz
my first PC - 25 MHz around 1991.
1995 - Pentium 100MHz
2000 - AMD 1GHz
2002 - P4 3Ghz
2011 - 8 GHz, but an AMD special with nitrogen cooling that you can't buy in the shops. Fastest clock speeds you can buy are around 3.6 GHz - hardly any change from 2002.
http://en.wikipedia.org/wiki/Clock_rate
Laban
Would love your hear your investment implications.
I think even speed did collapse we will see more ambitious progress in distributed core computing allowing computing per se (in flops for example) to increase. We may not reach supercomputer speeds in our desktops and mobiles but it can go up significantly.
So the hardware part will ease up and distribution part will take hold.
It's not all about clock speed. In actual fact, applications these days aren't demanding in terms of hardware, with the exceptions of latest games and things like CAD software. And on aggregate, it is difficult to see where computerization of our world will end, but we have a long way to go. PCs sales may be flat, but tablets, phones are growing, cars have up to 30% of electronic components.. But inserting microchips into other white goods is only starting, in toys too, in clothes etc.. And 3d printing could also push adoption further. Of course, all these things will make it hard for existing chip behemoths to continue being profitable as they have been because these trends imply disruptions.
But that is Moore's dream from 1965:
The law is named after Intel co-founder Gordon E. Moore, who described the trend in his 1965 paper.[2][3][4] The paper noted that the number of components in integrated circuits had doubled every year from the invention of the integrated circuit in 1958 until 1965 and predicted that the trend would continue "for at least ten years".[5]
(from wikipedia)
Quantum / optical computing anyone?
Moore's Law has not yet collapsed, but it has become completely irrelevant for ordinary users.
The surprising thing is that Intel's margins have not collapsed. The PC makers just keep buying the latest Intel chips and paying a premium for it.
There was always the atomic/quantum limit looming. But I am very impressed by the engineering ingenuity involved in the march towards 22nm on such a regular pace as if it were a after dinner stroll.
Also interesting that Moores law stalls around the same time that baby boomers enter retirement. Surely they are pure coincidences. Yet the implication for growth can't be overstated. There are things in the world that are very hard to predict. There are others that are inevitable even if only in the distant future.
Perhaps Metcalf's law is the more important one at this point.
Only a limited set of users will need any more processing power. I'm writing this on a secondhand laptop that I've been using the last three years. For most of my computer use this vintage Compaq nc6220 is sufficient. It was probably retired by some big corporation because it has number tag 27692. I can even program python on it and do sound editing. Probably it will harm the cloud computing centers that need massive power.
A discussion is going on about this on slashdot:
http://tech.slashdot.org/story/12/07/12/1737237/pc-sales-are-flat-lining
Re whitewashed windows and vacant stores : there's a real balance there.
a) internet means you can compare prices, get best deal...
b) but the item's got to go in a wagon to a distribution centre then in a van to your house. All costs money - I wonder some retailers are making anything from online sales.
There's a petrol/labour price at which you might do better shopping once a week in town. Now if I could just work out what it was ...
You can sleep easy, PR and Marketing will find a way around it...
Must admit my first reaction was "so what?" But I'll bite... What investment implications do you see?
damn right, buy asml:
http://www.asml.com/asml/show.do?lang=EN&ctx=5869&rid=46711
go asml..
Implications?....peak margins?
As Dan said Moore's Law was never really about the clock speed but about the number of components (gates) you can get per given area of Silicon. Since the increase in number of gates has been down to the ever-shrinking geometries we have, as a kind-of bonus, also got increasing speed.
Since e.g. 2002 the processes' ``feature size'' has come down from 180nm to 22nm.
The trouble is that at some point, about the 65nm -> 40nm transition, the power consumed suddenly shot skywards due to increasing leakage; and this increased leakage is independent of clock speed.
Since power also increases as the speed increases there's a max speed beyond which the chip gets too hot to function and an increasingly large part of the temperature ``budget'' is being used up by the leakage currents.
So, umm... In your dream Moore's Law's gone leaving a state of Law-Less-ness?
http://arxiv.org/abs/1207.1463
I too had an incredibly precise dream. Murphy's law had collapsed, also having all sorts of investment implications.
Everyone should do a little research on spintronics. We will move from electron charge to electron spin.
While the demand for raw processor speed may plateau, there'll be a continued push for greater processor efficiency (as AAPL is doing now with their iPhone/iPad processors). Given the large real world demand for this (longer battery life, thinner packages with lower operating temp chips) there's still large pricing power for such efficiency improvements (much like Boeing and Airbus being able to sell boatloads of marginally more efficient airplanes). Intel has more competition on this front, but it'd be at worst a rearranging of that sector for at least another decade or so. After that the processor side may well "decontent" and lose the ability to differentiate between products, which would be a wholesale change.
Further downstream AAPL is a "post Moore's Law" company - their big profit driving devices (iPad, iPhone) deemphasize raw power and put an effort on integrated package. The demand for upgrades is as much fashion as upgraded capabilities. This is even more apparent on the PC side where people and enterprise just don't care to upgrade much anymore, certainly not like 10-20 years ago where there were real improvements in even the span of a year.
The market valuations of players somewhat reflect this, and it would seem practically that Moore's Law already broke. Which doesn't matter because Moore's law mattered more to the press and analysts than anyone technical.
Even if Moore's law stopped tomorrow we can probably get another factor of 10 to 100 improvement in cost per flop for compute intensive programs compared to current leading desktop chips through better chip design and programming.
we are already there. The implications of the stop of Moores law is that
1) Alternate designs would carry more weight (rise of the foundry/ASIC house)
2) Software would be much more valuable than the underlying CPU/hardware.
Both of these are now demonstrably true. The relative value add of hardware is small. Even in mobile, the mobile CPU itself is a small fraction of the total power demand, and a sufficient mobile experience is now cheap.
All the important factors in Moores law are done. We are all jus twaiting on ~4 years for DRAM to enter the same state.
MY investment thoughts:
DOWN on the major equipment vendors, which are essentially investors in the 'churn' of technology.
There is lots of money in the electronics business. Right now, apple makes all of it....that won't last, and renogotiation on the part of the major manufacturers will happen.
Moore's Law is alive and well, but old software can not automatically capture the gains. There is a shift underway but it is "under the hood" and only in applications where performance matters. As that happens, the dividends from Moore's law are being paid out as decreased power consumption. Not a bad thing.
BTW, Moore's Law is a factor of 100x per decade. Any graph of CPU performance will show a faster trend from 1994-1997 and a blow off in 1998-2001. Long term, we are on track and I am 100% certain we will get our lunch this decade.
Luckily, now that we have tablets and smartphones, moore's law can start all over again!
As I type this on my laptop, the main frustrations are not with my processor speed but rather with slow internet and inefficient programs (and r/w speeds), so I doubt even a diminished moore's law would really effect my life.
trust your dreams. but often they are not precise enough, even if they seem so, so handle with care. g.
PS: I love your blog! thanks
This seems relevant:
http://www.nature.com/news/a-giant-bid-to-etch-tiny-circuits-1.11060
"“The situation has changed from, ‘Can we make it work?’ to ‘We have to make it work’,” says Stefan Wurm, director of lithography at Sematech, a semiconductor R&D consortium in Albany, New York.
...
ASML has just a few years to reach its goal. Intel says that it can use the current technology for two more generations of microchips, but only by adding expensive patterning steps and buying more lithography machines. Benschop admits that EUV technology should have been ready “yesterday”."
I've been thinking about this for a while (daydreaming comes naturally) and it hit home a little more when I was driving the other day. There seems to be a somewhat analogous situation in the case of cars, where rapid development of internal combustion engines has resulted in significant advances in torque/horsepower which are arguably hitting up against the limits of human cognition and reaction times (case in point: try having all the world driving a Ferrari F12 Berlinetta which is 544kW and 690 N.m and still maintaing a semblance of road safety. Many consumer cars today are more powerful than Ferraris of 20 years ago).
I think in our lifetime we will start to see real questions asked about the virtues of some technical advancements and how these are able to be applied successfully by humans - consider the KCG debacle from the past few days, which has in large part come about as a result of cheaper CPU cycles. I don't want to sound tin foil hat but we're letting some technology genies out of a bottle, so the advancement of Moore's law in infinitum may start to bang up against some human cognition limitations.
nb: Thanks for a great blog, hope this stream of consciousness post at least made some sense.
Post a Comment