Computing power: A deeper law than Moore's? | Hacker News Reader

Computing power: A deeper law than Moore's?

economist.com

14 pointsColinWright13 comments

This Koomey guy better have a conversation with Ray Kurzwiel before he go's around naming this "Discovery" after himself. Considering Kuyrzwiel's book "The Singularity is Near" already covered this phenomenon 7 years ago. I didn't see anything in this article that I hadn't heard before. That chart is nearly identical to ones in Kurzwiel's book.

Moore's law was simply an observation or a prediction about the computing market at that time: That the number of transistors on a chip would need to double every 18 months if the chip manufacturers hoped to get people to buy the next gen. He argued that people/companies wouldn't upgrade their existing equipment for a measly 10-50% increase in performance. They'd need to double performance if they wanted to get rapid turn over to the next generation of products.

tl;dr: Some researcher does the math and finds that computer chips are becoming more efficient over time, i.e., electrical efficiency doubles every 1.6 years. He wants to call it Koomey's Law.

Neither Moore's or Koomey's "Law" are very deep and don't really rise to the level of a true physical law. Its been known for a long time that Moore's "Law" was going to eventually run into the real laws of physics and that seems to be happening now.

>Its been known for a long time that Moore's "Law" was going to eventually run into the real laws of physics and that seems to be happening now.

I have been hearing people say this for a very long time, but I've never seen good evidence that it was true. I mean, at any point in time, the state of the art has "run into the laws of physics" - that's why it's the state of the art. At any given moment, the best we can do is limited by some physical property. Processors don't get faster because Intel just decides to turn the clock speed up. They get faster because researchers find ways around the old physical limitations.

And I think that's the fundamental misunderstanding people have about Moore's law and similar consistent growth trends. People see a consistent curve and they think "Oh, this is just the natural course of progress, and so it shall always be." But that's not how it works. The reality is that each point on that curve is the result of team of very smart people making a breakthrough.

For comparison look at the increases in aircraft speed. Many limits where broken, but eventually progress stopped.

One of the real issues has been the cost of fab over time increasing to the point where only few are able to keep up with the current processes by it's self. TSMC will be investing 9.3 billion dollars in its Fab15, and again Intel is still fine but the ever increasing capital costs have keep narrowing the field. And, soon it's going to be 50 billion just to stay in the game.

Compare how long it took to go from 1Ghz to 2ghz (about 2.5 years), to the current situation.

I bought an i7 920 (2.7ghz) in late 2008.

Now 4 years own and speeds have crept up to about 3.4ghz.

For the last few years AMD and Intel have just been side-grading by slapping more cores on, instead of actually making faster chips.

Yeah, but you can't really just go by clock speed.

If you have a bounded system growth will eventually level out. Physical considerations makes it unfavorable to go below the atomic level for feature size. Therefore growth in transistors per area is bounded and Mores law will eventually stop being true and therefore it is nothing but a phenomenelogical law.

I think Kurtzweil has suggested going into 3D for CPU design with multiple layers of circuits, but I have seen no serious attempts to realize this so it is a bit of a long shot to depend on to keep the pace of growth intact. However, it still does not resolve the fundamental limitation of Moores law, it only extends it for some time.

There has been some progress in 3D CPU design. Ivy Bridge, for example, uses a 3D transistor design although the CPU itself is still flat.

> The reality is that each point on that curve is the result of team of very smart people making a breakthrough.

My view is not meant to disparage the work of the people who keep the progress going. My point is that at some point Moore's "Law" has to breakdown since nothing real can double forever but perhaps you don't find philosophical/metaphysical arguments compelling.

As way of analogy, look at electric power generation based on a heat engine cycle. Ultimately you run into theoretical limits based on the Second Law of thermodynamics. On top of these physical constraints are Economic Law such as marginal utility which determine if it is really economically efficient to spend x dollars to gain y increase in efficiency or to stop short of the theoretical limit due to diminishing returns on investment in efficiency gains. These are real laws that computer chips will eventually run into.

My understanding of computer chips is that they are running into thermal limits and also bumping into quantum tunneling effects that are interfering with the deterministic nature of logic gates. I am sure people are working on ways around these limits and they may very well may find new and brilliant ways to continue the increase in processing power. It may even be true that theoretical limits to computation are far far away from where we are now and Moore's "Law" will continue to apply for decades but it still would not make it a actual physical law in the proper sense of the term.

Yes, there is a point where the doubling has to stop, but what I'm taking issue with is that people tend to point to current physical limitations as evidence that Moore's law is already over, or nearly dead. And at any given time in the history of computers, you could look at the fastest processors and say "they're bumping up against limit X". But until X is Bremmermann's limit[1], that is not an indication that progress will slow.

Incidentally, if Moore's law were to continue and it were as simple as "speed per mass doubles every 18 months", it would still take somewhere on the order of 300 years for us to approach Bremmermann's limit.

[1] http://en.wikipedia.org/wiki/Bremermann%27s_limit

This seems like a straightforward corollary of Dennard's power scaling law which says that the power of a transistor should be proportional to its area. http://en.wikipedia.org/wiki/MOSFET#MOSFET_scaling

"Some researchers are already building devices that run on “ambient” energy harvested from light, heat, vibration or TV transmitters."

Very cool.

Reply on news.ycombinator.com