Interesting post. The weak point seems to be the jump from Moore's law to a growth rate of 20% for "the computer sector". This seems like a figure picked ou of the air. I just got a new computer at work -- it is 3xfaster than the old one, and I'm more productive, but I'm not sure how much. Am I missing something?

>>As an economist....the noncomputer sector grows at 1 percent per year, and the computer sector grows at 20 percent per year<<

Brad has just answered this point (although it still amazes me how little basic economics you need to understand to call yourself an economist these days - well, come to think of it, I take that back, it doesn't amaze me, it explains how Glen Hubbard got to be where he is), while Morgan Stanley's Richard Berner saw it coming (see further down this blog):

" But there are three hurdles to strong growth in nominal IT company revenues. First, IT prices generally are falling as fast as volumes are rising, so IT revenues are barely increasing. For example, producer prices for computers and related equipment fell by 20.5% in the year ended in September. In addition, telecommunications companies glutted with capacity continue to slash capex. In the year ending in August, telecom equipment shipments fell by 20%, and new orders fell by 11.6%. Finally, overseas demand is faltering, adding to the pain."

And why, may we ask, are industries glutted with capacity and why is capex falling? One part of the story is the recent technology boom-bust cycle. But more long-term and more-preoccupying for those trying to decide investment strategies, is the fact that it's the very pace of the change, and the difficulty of predicting stable uses (ie the killer-app game, see comments on previous blog) that generates the uncertainty parameter that is very hard to handle.

>>Ray Kurzweil is one of the few people willing to take Moore's Law seriously as a long-term proposition. If he's right, then according to my arithmetic the economic boom that lies ahead will be stunning.<<

Nope. For the reasons given above by Brad and others these aren't the conclusions you necessarily need to draw. But the mention of Kurzweil is timely here. For those who don't know his arguments the following presentation charts might be a good place to start:
http://www.kurzweilai.net/meme/frame.html?m=10

All this reminds me of some of talk I've seen in the bio-informatics community. Apparently in gene and associated research it's a commonplace that it's more effective to teach molecular biology to someone with a computing background than vice-versa. Perhaps, looking at some economists as they try to wrestle with technology, it might also be true that it's cheaper and easier to teach good technologists like Kurzweil some simple economics and then listen to what they have to say.

If we showed Kurzweil some simple economics, he'd laugh at it the same way John Von Neumann laughed at econ. in the 1930's.

IF RK is right, the ontological and ecological implications of the computer may make economics unnecessary.

The pace of the development of computers is constrained by the way brains/bodies learn and share. Someday money itself will be a barrier to that process; lets use what knowledge of ecology, computers, economics, medicine etc. etc. we have now to bring that about.

I reiterate the same point I made in the last debate we had on this subject; the growth rate of clock speeds is a red herring. The very fact of Moore's Law guarantees its economic irrelevance.

Think about it this way; the "productivity" of a computer is the solution to a maximisation problem, the constraints on which are a number of technical factors; clock speeds, memory, disk access speeds. The fact that processor speed has increased so rapidly means that it is almost certainly not the binding constraint.

I'm not trying to argue that computer productivity isn't increasing incredibly quickly; just that Moore's Law is a really bad way to think about that increase.

In related news, I'd caution that the very fact that Intel is planning these things seven years ahead suggests that the pace of improvement might be slowing. Did anyone foresee the computers of 2002 in 1995? Or those of 1995 in 1987?

Intel was indeed planning 2002's computers in 1997. Chip design is chancy and expensive, so architectures are usually designed with at least a decade's worth of room to grow. (I recall seeing Alpha roadmaps in 1993 trade mags that went out to 2003.) "Room to grow" doesn't mean they design the chips ahead of time; it means they design the current chips so that there won't place any known-serious obstacles to improving them in the future. (There are frequently design optimizations that can improve the performance of a chip now, at the cost of making it harder to improve down the road.)

Sometimes the companies screw up really badly. A notorious case is the Intel IA64 architecture. Not only did Intel ship half a decade late, but several *other* chip vendors (SGI, HP, Digital/Compaq) decided that they couldn't compete with Intel and mothballed their own lines. A big oops, that left IBM and Sun laughing all the way to the bank. :)

Am I the only one puzzled by (1) what it means to speak of the "output" of the computer sector, (2) why that measure is connected with Moore's Law?

??

>>Am I the only one puzzled by (1) what it means to speak of the "output" of the computer sector, (2) why that measure is connected with Moore's Law?<<

No; the author of the piece Brad quoted is clearly also confused, though he may not realise it.

Why isn't the obvious comparison made with the oh-so-wondrous storage and telecoms industries, which had higher growth rates (100% per year for storage, far in excess of that for telecoms).

Look at teh vast amount of money seagate and IBM storage made in the past year, on top of their 100% per year growth. Look at the huge demand increase by comsumers. The mass capacity increases were clearly followed by huge demand for services.

Those industries did so well, tied to equivalent metrics, that there is clearly nothing but blue skies ahead for the computer industry.

B

Please set me right....

Intel and Hewlett have been preparing for a new generation of chip beyond the pentium. My understanding of the project was that is was much delayed, far more costly than hoped, far more energy intensive than might be worth while.

Will this new generation of chip be a success simply because it is faster? Suppose companies decide to use more of the pentiums to grow rather than a more costly, energy hogging, chip?

There may be ways to gain speed at lower cost. What then?

Please excuse the note from a philosopher.

Intel's Huge Bet Turns Iffy
By JOHN MARKOFF and STEVE LOHR (NYT)

"Google, the Internet's leading search engine, powered by an arsenal of computers with 15,000 microprocessors -- should be a premier customer for Intel's new Itanium 2 super-chip.

"Itanium, a joint project of Intel and Hewlett-Packard, Silicon Valley's two largest companies, has been in the laboratory for more than a decade. Itanium is designed to excel at a sweeping array of advanced computing tasks, from solving grand scientific challenges to rendering complex graphics to slicing through vast databases. With more than 200 million transistors on each chip, it is designed to process data in big bites -- 64 bits, in chipspeak -- at blazing speeds."

Suppose Google were to find an alternative to Itanium....

Actually, Moore's Law refers to number of transistors (not clock speed). For most purposes, we can think of it as computer processing power. That said, however, this discussion only speaks to the limits of silicon IC processors.

Certainly, if the discussion is phrased in terms of processing power, Moore's Law can continue indefinitely --it might even be pessimistic. There will be breakthrough ways of processing information (non-silicon IC) using technologies will come on line within ten years --count on it!

There will be DNA-based processors, chemical-based processors, etc. There will be methods of dividing problems and finding more efficient solutions to them in the future. These solutions are only beginning to be considered today.

I'm certain that with more processing power, we will be able to do things like develop treatments for diseases in near real-time compared to what we're doing today. Think of the capability to generate models many orders of magnitude more complex than today. We will be finding relevant data that we don't even know enough to use today. And we will be able to process it for real useful applications: economic models, weather models, global climate models!

To follow up on Daniel Davie's point, everyone here is taking it as given that
(a) Intel's chief technology officer is telling the truth and
(b) that 15GHz in 2010 means 5x as fast as 3GHz now.

I would point out, wrt the first point, that Intel's primary concern these days is not what they'll be doing in 2010, but how to deal with the disaster that Itanium has become, and how to deal with AMD's upcoming Opteron. Responding to these issues requires a whole lot of convincing the world that "even though we look pretty clueless today, we have a SECRET PLAN". In other words, these pronouncements are worth no more than the pronouncements of an analyst two years ago pushing WorldCom.

As for the second point, recall that the P4, while indeed running at a much higher frequency than the PIII, is for many uses a disappointment. The 18 FO4 timing of each clock stage limits the smarts one can perform at each stage to try to hide latency issues, perform superscalar work, and predict branches. Intel presumably expect to reach 15GHz through a combination of a slicing of each pipeline stage in two to 9 FO4, along with process improvements. The process improvements will be real (though perhaps limited in impact by the speed of RAM), while the new micro-architecture will prove even more disappointing than that of the P4 to anyone running non-regular code.

A couple of points, from a former tech jockey and first-year econ grad student...

Re: Kurzweil, whoever mentioned sitting him down and teaching him economics is right on the money. I largely agree with his technological predictions, but there are a few out-and-out economic howlers in his writings. The main one I'm thinking of is his prediction of massive, continuous deflation brought on by moore's law-like growth in the rest of the economy.

Only problem is, deflation (like inflation) is a monetary phenomnon. The general price level can only go down if there's less money in circulation - and that has to do with governemtn policies, not technology. Increased productivity in some sectors leads to a change in realtive prices, but that's not the same thing. Good bet: proces of manufactured goods will drop to virtually nothing, while human labor and land will continue to take a larger share of nominal economic output. (In other words, dump the stocks and buy houses - just be sure to put 30-year roofs on 'em...)

Which beings me to my other point - as it happens, I was just recently (for a midterm I just took today) going over Scraffa's "Production of Commodities" input-output model. The way I understand it (and since I don't think I did all that well on the midterm, maybe I don't...) since computer power can be understood as a "basic good", used in the production of all others, isn't tracking growth sheerly by "growth in the computer sector", well, wrong?

After all, according to the model, increases in efficiency in one inductry do not neccessary redound to the benefit of that industry - the relusting surplus can be distributed elsewhere, wccording to market pressures. The conputer industry could be driving growth in the whole economy while accounting for a constant, or even shrinking, share, and earning profits no greater than any other...

Hi Maynard, while I agree with you that the Itanic has sunk, I don't think that necessarily implies that Intel's numbers are complete hokum. They need to come up with a set of numbers slightly better than what the tech press thinks AMD can do, but not so much better that everyone thinks they're blowing smoke. In that sense, it should represent a slightly upward-biased estimate of how well they think the industry can do.

Besides, Intel has so much cash they could afford half-a-dozen IA64-class disasters before they start seriously hurting. Surely one of their skunkworks projects is going to work.

Jimbo

>>The main one I'm thinking of is his prediction of massive, continuous deflation brought on by moore's law-like growth in the rest of the economy.<<

This isn't a howler; it may be a mistake, and I personally think it is, but it's not a howler, and under the title of "Dutch Disease" (referring to a period in the Netherlands when the hypergrowth sector was natural gas extraction), it's gained quite a few adherents. I very much dispute whether labour will extract much of the value from productivity growth in computers; the experience of the 1990s would certainly suggest the opposite.

On the other hand, I wholeheartedly approve of the fact that your university is teaching Sraffa to first year grad students, and that you're applying it. I seem to remember that Robert Vienneau, the author of a number of very good online papers about Sraffa's model, is also a computer scientist by training.

>> This isn't a howler; it may be a mistake, and I personally think it is, but it's not a howler, and under the title of "Dutch Disease" (referring to a period in the Netherlands when the hypergrowth sector was natural gas extraction), <<

It's seems more a question of semantics. Doesn't Dutch Disease refer to a situation in which an abolute advantage in the production of one good hurts the economic competitiveness of a country otherwise by bidding up the price of universal factor inputs?

>> Intel is going to keep Moore's Law going...<<
~~~

Scientists Shrink Computing to Molecular Level

[S]cientists have created what they say is not only the world's smallest logic circuit, but also possibly the smallest that could ever be made.

The entire circuit covers less than a trillionth of a square inch. The equivalent circuit made from state-of-the-art silicon transistors takes up 260,000 times as much space.

"That gives you an idea for what incredible potential there is for miniaturization," said Dr. Andreas J. Heinrich, a physicist at I.B.M.'s Almaden Research Center in San Jose, Calif., and the lead author of a paper that appears today on the Web site of the journal Science.....

The technique is far from practical.... Still, scientists marveled.

"The beauty of this experiment is opening our eyes," said Dr. Wolf-Dieter Schneider, a professor of physics at the University of Lausanne in Switzerland. "There might be new possibilities in computing...

http://www.nytimes.com/2002/10/25/technology/25COMP.html

"I very much dispute whether labour will extract much of the value from productivity growth in computers; the experience of the 1990s would certainly suggest the opposite."

Note: saying that "labor intensive industries" will account for a larger share of the economy does not mean that the workers in those industries see any of the loot. I mean, workers can afford more toys as those toys get relatively cheaper, but their wages relative to capital (i.e., who gets the surplus) are determioned outside the model.

Sadly, Intel strikes me as fair-valued to over-valued even before options costs are included in earnings. What sort of growth rate can Intel expect, what sort of p/e would be fair value even if they refuse to count options cots against earnings?

Daniel:

You ask, "Did anyone foresee the computers of 2002 in 1995? Or those of 1995 in 1987?"

Well maybe -- things are more evolutionary than revolutionary now. But when Intel started planning for the 80286 processor in the 70s (the one used in IBM's third-generation PC, the PC-AT), they asked marketing to identify possible markets for the thing. The list marketing came back with didn't even include computers!