Does Moore's Law Help or Hinder the PC Industry?

An anonymous reader writes to mention two analysts recently examined Moore's Law and its effect on the computer industry. "One of the things both men did agree on was that Moore's Law is, and has been, an undeniable driving force in the computer industry for close to four decades now. They also agreed that it is plagued by misunderstanding. 'Moore's Law is frequently misquoted, and frequently misrepresented,' noted Gammage. While most people believe it means that you double the speed and the power of processors every 18 to 24 months, that notion is in fact wrong, Gammage said. 'Moore's Law is all about the density...the density of those transistors, and not what we choose to do with it.'"

Comments

Both

[DaAdder]

I suppose it does both.

The drum beat of progress pushes development to it's limits, but at the same time hinders some forms of research or real world tests of computation theory, for all save the few chip makers dominating the market currently.

Re:Both

[ElectricRook]

We're also not paying US$800 for a 80387 math co-processor (only did floating point). Like a friend of mine did in the 80's. That would be about $US1,600 in today's dollars.

Re:Both

[arktemplar]

can I have that in libraries of congress please ?
*ducks*

I'm gonna vote for hurts - big time

[$RANDOMLUSER]

If only because it keeps us tied to the x86 instruction set. If we didn't have the luxury of increasing the transistor count by an order of magnitude every few years, we'd have to rely on better processor design.

Re:I'm gonna vote for hurts - big time

[bronzey214]

Sure, being tied to the x86 architecture hurts, but it's nice to have a pretty base standard as far as architectures go and not have to learn different assembly languages, data flows, etc. for each new generation of computers.

Re:I'm gonna vote for hurts - big time

[$RANDOMLUSER]

Game designers do it all the time. Compiler writers do it all the time. For 99.5% of the programmers out there, the underlying architecture is a black box; they only use the capabilities of the high-level language they happen to be using. But the final performance and capabilities of the system as a whole depend on that underlying architecture, which has been a single-accumulator, short-on-registers, byzantine instruction set (must. take. deep. breaths...) anachronism for far too long.

Re:I'm gonna vote for hurts - big time

[Chosen Reject]

And for 99.9% of users the underlying architecture is a block box. But 100$ of applications the underlying architecture is important, and if the application doesn't run then the user gets upset. Doesn't matter if the application only needs to be recompiled, even if the developers gave away free recompiles to people who had previously purchased the software, it would require users to know which architecture they have (already difficult for most people) and make sure they get the right one.

Have you ever seen people confused about which package to get for which Linux distro? They don't know if they should get the one for Fedora, Ubuntu, Knoppix, Gentoo or Debian, and then they have to decide i386, x86_64, ppc, or whatever else there is.

Yes, most developers would have no problem, and most users wouldn't care once everything was working, it's just getting things into a working state that would suck when underlying architectures are changing every few years.

Re:I'm gonna vote for hurts - big time

[TheLink]

So any recent benchmarks of how the latest T2 stuff does vs recent x86 machines in popular server apps like _real_world_ webservers, databases?

AFAIK, it was slower than x86 the day it was launched, and when Intel's "Core 2" stuff came out it got crushed in performance/watt.

Re:I'm gonna vote for hurts - big time

[sgt scrub]

we'd have to rely on better processor design. Not to mention we'd have to rely on better software design. The way Moores law effects software, by allowing it to bloat, is the anti technology.

Moore's Observation

[Edward Ka-Spel]

It's not a law, it's an observation.

Re:Moore's Observation

[jhfry]

If you think that Intel or AMD double the number of transistors in an effort to keep up with Moore's law than you know nothing about business.

No one does anything in an effort to prove Moore correct... they do it for their own benefit. Intel does it to stay ahead of their competition and continue to keep selling more processors. If they chose to stop adding transistors they could pretty much count on losing the race to AMD, and likely becoming obsolete in a very short time.

I agree that more transistors != better... however it is indeed the easiest way, and least complex, to increase performance. Changing the architecture of the chip, negotiating with software developers to support it, etc, is far more complex than adding more transistors.

Re:Moore's Observation

[TheRaven64]

It takes roughly 3-5 years to design a modern CPU. At the start of the process, you need to know how many transistors you will have to play with. If you guess to few, you can do some tricks like adding more cache, but you are likely to have a slower chip than you wanted. If you guessed too many, you end up with a more expensive chip[1]. Moore's 'law' is a pretty good first-approximation guess of how many you will have at the end of the design process. A company that can't make this prediction accurately is not going to remain competitive for long.

[1] There is no real upper bound on the number of transistors you can fit on a chip, just the number you can for a given investment.

No significances.

[Frosty Piss]

"Moore's Law" is not a real law. In reality, it is not relevant at all. It's kind of a cute thing to mention, but when it gets down to the real world engineering, it has no significances.

Re:No significances.

[truthsearch]

While you are correct, it has value as being accurate foresight. So the question is, was it just an observation or did it become a self-fulfilling prophecy? If it was a self-fulfilling prophecy then what other judgements can we make now that may drive technology in the future?

Answer to the question?

[titten]

If Moore's law is helping or hindering the PC industry? I don't think it could hinder it... Do you think we'd have even more powerful computers without it? Or higher transistor density, if you like?

Efficiency

[Nerdfest]

It certainly seems to have had an effect on peoples attention to writing efficient code. Mind you, it is more expensive to write code than throw more processor at things ...

Re:Efficiency

[Kjella]

It certainly seems to have had an effect on peoples attention to writing efficient code. Mind you, it is more expensive to write code than throw more processor at things ...

Well, you can have software that's feature-rich, stable, cheap, fast or resource efficient, pick any two (yes, you still only get two). Let faster processors handle speed and GB sticks of memory handle resource efficiency, and let coders concentrate on the other three. The margin between "this will be too slow it doesn't matter what we do" and "it's so fast noone cares" is usually very slim (unless you're talking about major changes like using smarter algorithms, pushing heavy processing out of a loop etc. in other words a smarter design, not assembly hacking).

Murphy's law...

[jhfry]

is more important to nerds than Moore's law anyway. Where's the /. article about it?

Murphy tells us that more bugs will be found on release day than any day previous. That your laptop will work fine until the very minute your presentation is scheduled to begin. And that backup generators are unnecessary unless you don't have them.

Who cares about Moore's law... it's just prophecy from some Nostradamus wannabe.

density of transistors?

[Red Flayer]

While most people believe it means that you double the speed and the power of processors every 18 to 24 months, that notion is in fact wrong, Gammage said. "Moore's Law is all about the density...the density of those transistors, and not what we choose to do with it."

Hmmm. Seems to me Gammage might have it backwards, the misunderstanding of Moore's Law by most people is due to the density... the density of those people.

Re:density of transistors?

[hAckz0r]

My first own version of Moore's Law states in rule one; that the 'density' of the sales force is inversely proportional to the 'core size' (N) of the sales force times e^2. [eg. 1/(N*e^2)] That is the only "density measurement" worth paying attention to when buying any new computer equipment.

My second law of 'density' states that that the PR intelligence quotient is randomly modulated by Schroedingers' cat in the next room, and is only measurable when not actually listening to it.

Re:density of transistors?

[Red Flayer]

My second law of 'density' states that that the PR intelligence quotient is randomly modulated by Schroedingers' cat in the next room, and is only measurable when not actually listening to it.

Wow, you deserve a Nobel Prize. You've figured out how to directly measure a null value!

In My Opinion, It Isn't a Law

[eldavojohn]

I always viewed this as an observation or rule of thumb, not a law.

Moore (or Mead for that matter) didn't get up one day and declare that the amount of transistors on a square centimeter of space will double every 18 to 24 months. Nor did he prove in anyway that it has always been this way and will always be this way.

He made observations and these observations happen to have held true for a relatively long time in the world of computers. Does that make them a law? Definitely not! At some point, the duality that small particles suffer will either stop us dead in our tracks or (in the case of quantum computers) propel us forward much faster than ever thought.

Why debate if a well made observation hurts or hinders the industry when it's the industry doing it to itself?!

Better Summary

[Palmyst]

The core of their argument is that instead of actually delivering same performance at lower prices, Moore's law delivers more performance at same prices. i.e. you can buy Cray-1 level performance for $50, but you can't buy Apple I level performance for $0.001. The second level of their argument is that this march of performance forces users to keep spending money to upgrade to the latest hardware, just to keep up with the software.

Here's a more interesting question:

[spun]

Does Cole's law help or hinder picnics?

Discuss.

Definately Both

[john_is_war]

With companies driving to increase transistor density by decreasing process size, the speed we can accurately use these methods is slowing. With each decrease in process size, a lot of issues arise with power leakage. This is where multi-core processors come in. These are the future because of the speed cap of processors. And hopefully this will spur an improvement in microprocessor architecture.

The Real Story

[tomkost]

The real story is that Moore's law describes the basic goal of the semiconductor industry. Perhaps there are better goals, but they tend to get swallowed up in the quest for smaller transistors. The other real story is Gate's law: I will use up those extra transistors faster than you can create them. My hardware OEMs need a bloated OS that will drive new HW replacement cycles. I also seem to remember Moore's law was often quoted as a doubling every year, now I see some saying 18-24 months, so I think in fact the rule is slowing down. We are pushing into the area where it takes a lot of effort and innovation to get a small increase in density. Even still, Moore's law has always been a favorite of mine! Tom

Why?

[malsdavis]

Why do computers in general need to get any faster these days?

Ten years ago I wouldn't believe I would ever ask such a question but I have been asking it recently as my retired parents are looking to buy a computer for the web, writing letters and emails. I've told them specifically "DO NOT BUY VISTA" (why on earth would anyone want that ugly memory-hog?), so I just can't think of a single reason why they need even one of the medium-spec machines.

Personally, I like my games, so "the faster the better" will probably always be key. But for the vast majority of people what is the point of a high-spec machine?

Surely a decent anti-spyware program is a much better choice.

It could be...

[paladinwannabe2]

I've heard that companies plan, design, and release new processors based on Moore's Law. In other words, if it doesn't keep up with Moore's Law it's discarded, if it goes faster than Moore's Law its release is delayed (giving them more time to fine-tune it and get their manufacturing lines ready). If this is the case, then it could be hindering developement in new ways of processing (that have a payoff that takes more than 3 years to develop) and we might even be able to beat Moore's Law rather than follow it. Of course, Moore's Law is awesome enough as it is, I don't feel the need to complain about how it takes two whole years to double the effectiveness of my hardware.

Cost of fabs...

[kebes]

"...Every 24 months, you're doubling the number of transistors, doubling the capacity," he said. "But if you think about the process you're going through--they're taking a wafer, they put some devices on it, they cut it up and sell it to you--the cost of doing that is not doubling every 18 to 24 months."

Is he claiming that the cost of doing lithography on wafers doesn't increase? That's crazy talk! The cost of building and running fabs is in fact also growing exponentially. According to Rock's Law, the cost of building a chip-making plant doubles every four years, and is already into the multi-billion dollar range.

In fact there's alot of debate whether Moore's Law will break-down due to fundamental barriers in the physics, or whether we will first hit an economic wall: no bank will be willing (or able?) to fund the fantastically expensive construction of the new technologies.

Re:Cost of fabs...

[Tablizer]

The cost of building and running fabs is in fact also growing exponentially. According to Rock's Law ...

Rock's Law??? Tablizer's Law: The number of tech "laws" doubles every 2 years.

Here come the pedants

[Dara Hazeghi]

Cue all the pedantic asshats who absolutely have to point out that Moore's Law really isn't a Law... it's an observation.

Moore's Law is a crappy measurement

[Ant P.]

...Like GHz or lines of code.

Take the Itanic for example, or the P4, or WindowsME/Vista.

Why we need faster computers

[Animats]

• Windows. The speed of Windows halves every two years.
• Spyware, adware, and malware. Extra CPU power is needed to run all those spam engines in the background.
• Spam filtering. Running Bayesian filters on all the incoming mail isn't cheap.
• Virus detection. That gets harder all the time, since signature based detection stopped working.
• Games. Gamers expect an ever-larger number of characters on-screen, all moving well. That really uses resources.

Despite this, there have been complaints from the PC industry that Vista isn't enough of a resource hog to force people to buy new hardware.

Computers have become cheaper. I once paid $6000 for a high-end PC to run Softimage|3D. The machine after that was $2000. The machine after that was $600.

Simple answers from an old Guru

[Applekid]

What we do with the transistors? Run software of course. Enter Wirth's Law:

"Software is decelerating faster than hardware is accelerating."

Instruction set != architecture

[Criffer]

If only because it keeps us tied to the x86 instruction set. If we didn't have the luxury of increasing the transistor count by an order of magnitude every few years, we'd have to rely on better processor design.

I'm just going to refer you to my comment made earlier today when discussing a "new, better" processor architecture. Because there's always someone who thinks we are somehow "hindered" by the fact that we can still run 30-year old software unmodified on new hardware.

See here.

Re:Instruction set != architecture

[$RANDOMLUSER]

The miniscule number of registers everyone complains about is irrelevant

Were it not for the opcode fetches to register-dance (because only certain registers can do certain things), or having to use memory to store intermediate results (because there aren't enough registers), or stack-based parameter passing, (not enough registers) or, again, the single accumulator (more opcode fetches and more register dancing) you might have a point. But what you're suggesting (in the rest of your post) is that having 1000 horsepower on bicycle tires is the same as having 500 horsepower on real tires - and I can't agree.

...30-year old software unmodified...

Can you name any 30-year old software that is worth running unmodified? Hell, I'll give you a break. Can you name any 10-year old software that is worth running unmodified?

Re:Instruction set != architecture

[Doctor Memory]

there's always someone who thinks we are somehow "hindered" by the fact that we can still run 30-year old software unmodified on new hardware. We are, because it's just a new implementation of a crappy architecture. Apple showed that it's quite feasible to run old software on new hardware, even new hardware that had almost nothing in common with the old hardware. Intel provides x86 compatibility on Itanium, there's no reason why we can't all move to a new processor and take our old software with us. It's just that nobody's coming out with any new processors for PC-class machines.

I'd say the ability to run 30-year-old software unmodified on a modern processor shows just how little progress we've actually made...

Re:Bad code hinders the PC industry.

[ex-geek]

Windows vista on modern hardware boots no faster then 98 did on hardware that was modern for its era.

Boot time is constrained by harddrive seek times, not CPU throughput. Today's harddrives have only marginally better seek times than harddrives from 1998. PCs didn't improve much in terms of latency at all.

But few developers seem to be aware of this, which is probably one of the reasons for many types of apps starting even slower than they used to. Many apps abuse the filesystem as a database. My system has currently >600.000 files on it. In 98 I would have had maybe 2000 and back than, most of these files were my user files, rather than files for apps, configs and caches.

But can we?

[Rix]

30 year old software won't run on new operating systems, and 30 year old software won't support modern hardware. You won't be getting USB support, and parallel/serial ports are quickly disappearing. Where would you find a modem with drivers for your old OS? Where would you find a dial up ISP, let alone one that would support 1200 baud or whatever you'd be limited to.

You're going to be *far* better off running 30 year old software under emulation, where these things can be faked.