Is Moore's Law dead at 40 or is this just a mid-life crisis?

Is Moore's Law dead at 40 or is this just a mid-life crisis?

Summary: Last week, Michael Kanellos published this FAQ on the 40th anniversary of Moore's law, which is famously known as the phenomenon that computer processing power will double every 18 months.? Actually, Gordon Moore only said that transistor count would double every 24 months and it was David House (a former executive of Intel) who extrapolated that performance would double every 18 months as a result of the increase in transistors.?

SHARE:
TOPICS: Processors
26

Last week, Michael Kanellos published this FAQ on the 40th anniversary of Moore's law, which is famously known as the phenomenon that computer processing power will double every 18 months.? Actually, Gordon Moore only said that transistor count would double every 24 months and it was David House (a former executive of Intel) who extrapolated that performance would double every 18 months as a result of the increase in transistors.? Ironically, it is House's unofficial reinterpretation of Moore's law that has become the popular definition of Moore's law.

Over two years ago, Tomshardware released this excellent article showing the historical progress of Intel and AMD CPUs from 100 MHz to 3000 MHz from year 1994 to 2003.? The results were astonishingly true to the 18-month performance doubling cycle, which reminded me when name-brand 33 MHz 486 computers were routinely sold for nearly $10,000 back in the early 1990s.? Because of this, I had been conditioned to the point that if someone had told me two years ago that we would still be stuck with 3 GHz class CPUs today, I would have told them that they were crazy.? It's a good thing I never made a wager on that theory since what has happened since February 2003 seems to signal?that Moore's law is dead at the age of 40, or at least going through a serious mid-life crisis.

Today, the Intel CPU is barely hitting 3.8 GHz,forcing the Intel marketing machine to give up its long rigid stance that the megahertz is king.? The constraints on current CPU fabrication technology means that processor manufacturers really can't squeeze anymore GHz out of a CPU without a ridiculous amount of power consumption, heat output, and exotic cooling techniques.? This has forced the processor manufacturers to abandon vertical scaling and embrace horizontal scaling by adding more processing cores in parallel.? Fortunately, the last two years hasn't been all doom and gloom, since there was progress made in other areas of processors that were not necessarily all about the thirst for more speed.

  • Low power processors - Transmeta originally started the concept of MIPS per watt, which resulted in a processor that could do as much computational work as possible with the minimal amount of electricity.? Because of the success of Transmeta, Intel responded by designing a?low-power processor called the "Pentium M" that is currently marketed as the processor component of "Centrino" for notebook computers. Note that the Transmeta and Pentium M haven't been?limited to notebooks; they were also incorporated in server technology for high-density blade servers. The Intel CPUs were clocked much lower than their desktop cousins (around 1.8 GHz) and used much less power, but could match the 3 GHz desktop CPUs in performance.
  • Hyperthreading - A technology from Intel that allows a single CPU to behave as two CPUs. Although you're really not getting two true CPUs, it?does deliver a nice little performance boost while running multiple processing threads. IBM later released a similar technology called multi-threading.
  • NX - This is a security feature generically called "No Execute" and probably has half a dozen other marketing names depending on which company it's coming from. AMD beat Intel to the punch a year ago with their high-end Sempron processors, Opteron, and Athlon 64 processors, and Intel gets to play catch-up in this arena.
  • 64 bit extensions - AMD added 64 bit extensions to their 32-bit x86 processor line over a year ago so that there may be a smooth transition from 32- to 64-bit. Since the popularity of these processors have grown, Intel was pressured to release their own compatible version of 64-bit extensions for their high-end Xeon MP (Pentium 4 cores with a lot more cache) processors. Initially, Intel did not want to release the 32/64-bit hybrid technology because of the negative affect it would have on the Intel Itanium processor, which is a true 64-bit processor with 32-bit emulation.? The market has since proven that the hybrid technology is what?customers ?want.? Just last week, Microsoft released a version of Windows that will support these new 32/64-bit hybrids.
  • Multi-core processors - AMD recently released their own multi-core 64-bit extensions processor and Intel will soon be releasing their own multi-core processor.? The big question is: What does this do for my performance numbers?? The answer is, most likely, not much for most existing applications since it's very difficult to design computer applications that can achieve "perfect scaling" where doubling the processors will double the performance.? Contrast this with the past when double the MHz almost always doubled the performance, regardless of multi-processor optimizations.? The other big question is: What will multi-core processors cost?? Will they charge double for a dual-core processor?? My guess is that they will most likely charge a premium.? Unfortunately, customers have?grown accustomed?to getting double the performance for either the same or less money every 18 months.? We most certainly will not pay double the money for something that won't even deliver double the performance.? The biggest problem of all for multi-core processors is the threat of paying double for software licensing.? No one will spend $2K on a new CPU if it will mean that they need to spend an extra $100K on software licensing.? Microsoft is the first major database vendor to not punish you for buying multi-core processors, but IBM and Oracle are sticking to their per-core licensing models.

So the?question remains, is Moore's law dead?? As far as transistor count progress is concerned, Moore's law is alive and well.? Unfortunately, it just doesn't meet the popular perception that processing power is supposed to double every 18 months, and the public is feeling a bit short-changed by recent progress.? Ultimately, for the technology industry, it just means that?customers ?will simply hold on a little tighter to their wallets when it comes to buying a new computer because the one from two years ago isn't that much slower.? No amount of?re-education that GHz don't matter will?loosen our wallets?until the newer products start delivering a lot more performance and innovation?for even less money.

Topic: Processors

Kick off your day with ZDNet's daily email newsletter. It's the freshest tech news and opinion, served hot. Get it.

Talkback

26 comments
Log in or register to join the discussion
  • More Ghz is dead

    Two more generations to go - 65nm and 40nm - then it ends. 40nm process will be VERY expensive - and I bet only InHell will do it. They will chase the elusive 25nm process - but they will never make it work. This means that 5 years from now, the number of transistors per square centimeter will be a fixed number. At THIS point, the ONLY way to get a faster processor, is to make the most of the space you have. This means that you fill it up with cache and registers and remove anything that you can get away with. Viola! You now have an Itanium chip.
    Roger Ramjet
  • my new rule says we double ever 5 years

    you read it here first
    JasonL31
    • Jasons Law

      has a nice ring to it :)
      JasonL31
      • I like Jordy's Law better

        Power is directly perportional to fun :)

        actually he came up with it in terms of cars, but it clearly applies to computers as well proving the universality of it
        zijiang
  • Moore Law

    Frist, Moore Law is the doubling of the density not the performance.

    Second, The density will increase just fine for quite a while, so Moore law is fine.

    The performance is a trickier problem. Scaling laws have recently broken and that is what is hurting performance. It use to be that we could shrink everything and nothing will blow up. Now we can not shrink certain dimension - gate oxide without the power comsumption increasing. It is the end of scaling that needs to be fixed. New materials could fix it, but this industry is one of the most new material phobic industries I have ever seen.

    The system performance is really the problem, not just the chip. That is why no one is pushing the GHz of the processor anymore. It does not really make the system faster or the user experience any better.

    I buy my systems more on the cycle of the support logic chips than the CPU, they control the functionality of the system more. That is why Intel just reorganized.

    Moore law is fine, we have to learn to think in terms of systems. Dual chips is good first step, but will require all of the vendors to rewrite their applications before we see a big benefit. Hopefully, after the software is rewritten our machines will not pause for some stupid process that decides to run while I am working. Multiply threads and processors should combine into a better and snappier user experience and that is speed for the most of us, not the frame rate of some game.
    donf_z
    • I concluded the same thing

      First of all, I concluded the same thing as far as the true meaning of Moore's law is concerned. It's the popular unofficial version of Moore's law about performance that I said was suffering.

      Unlike you, I'm not going to buy a new computer just because of a snappy new chipset or some new marketing campaign. I better see some better transactions per second at work and some better frame rates at home before I toss my 3.4 GHz single core based computer. I won't ever buy first and hope for the optimizations later, I'll buy when they show me the pudding.
      george_ou
      • Now I see

        I went back and read your article better and as you said you did have it correct.

        65 nm is bringing some better design, but nothing to really fix the power issue. We can not scale gate oxide thickness and that means no huge performance improvements. Even 45 nm is looking like it will not fix this issue, the choices being made right now do not look like the companies will try to fix the gate oxide problem. Intel and others are trying to design the chips such that sections can really turn themselves off when not needed. They hope that will save power(?), but until we can get back onto the scaling law curve we will see the huge increase in performance every node. That will require a new material to be used at the gate level. So far no one has decided to do this.

        Our only hope in the short term is dual processors on a die or multichip modules. Intel is doing both in the short term. Now our operating systems and applications just need to be rewritten to better balance the work load across multiply processors.

        The Cell processor I do not see helping any except is very special applications. It is way to asymetrical and it is a very big chip.
        donf_z
        • Are they going to sell multi-cores at the same price as single cores?

          I'm ok with doubling the cores every 2 years to increase processor performance. Even assuming that software will rise to the challenge of dealing with SMP more efficiently, are the processor makers going to sell me a 3.4 GHz dual-core at the same price of a 3.4 GHz single-core processor from 2 years ago? Somehow I doubt it. Unfortunately, that is the only thing that gets me to go out and buy a new computer every 2 years. If they don't maintain their rate of increase in the bang/dollar, I can't maintain my rate of hardware purchases. This doesn?t even begin to address the licensing issues.
          george_ou
          • Same price

            Yes, they will have to that is Moore's law. If speeding up the pipeline and deeping the pipeline does not improve performance than doubling the processor units will have to be used to drive sales. Double the processor for the same money. This really is Moore's law doubling the density for the same money.
            donf_z
          • I hope you're right

            But the first 3.6 GHz dual-core CPUs they sell will probably be very expensive somewhere in the range of $600 on up. By the time it is released sometimes next year, it will be double the price of a 3.4 GHz processor was two years earlier. In that sense, they are violating Moore's law. Just speculating here, but I'm concerned.
            george_ou
          • first dual cores

            The first batch of dual cores desktop CPUs are not really worth considering, and they will be expensive. They are trying to use a cheap multichip package without a lot of chip to chip connections. Nor is the memory optimzied to really allow high speed swaping from die to die. Even with these lack, it will cost alot, since Intel and not get a cheap packaging house to do the work. Wait till the dual Pentium M chips core on one die are availible, those are the ones I want to see.
            donf_z
  • A Better Question?

    Perhaps a better question might be, Is the death of Moore's Law a bad thing?

    Over the past 25 years Intel has giveth and Microsoft has taken away. Perhaps this slowing of Moore's law will force the creation of better code. For a variety of reasons Apple got stuck with sub standard porcessors for several years. Remember when the G4 was stuck at 400 mhz for over a year, while intel zoomed past 1 ghz?

    While at first glance this would seem to have been a disaster for apple, I believe it made them write better softwared. While MS continued to require faster and faster hardware just to run the their latest versions of XP Apple has forced their software to work on slower and slower machines. The result, each new version of osX is faster than the previous on the same hardware.

    In the end this isn't a question of MS bashing, it may just mean that all writers of software, not just MS need to focus on the performance of their software on machines that are approaching five years old.

    I for one am tired of chasing the performance curve and now keep machines as long as possible. With a wife and two kids I no longer have the luxury of replacing machines every 30 months like I did in the early 90s.

    regards

    Joe
    joedokes
    • You got it all backwards

      1. Moore's unofficial performance law is a great thing. It is the reason why we now have $400 3000 MHz computers with 256 MBs of RAM and a 120 GB Hard drive on sale today. In 1991, a 33 MHz 486 with 4 MBs of RAM and a 0.04 GB Hard drive for $10,000. If you really think this is a bad thing, call me because I have a 486 I'd like to sell you at half price.

      2. MS software is relatively trim in most of the last decade. Take Office 2003 applications for example which loads up on my computer in under 3 seconds (with no pre-load cheats in the startup). Most of the office applications hover around 10 megabytes of system memory. The arrival of truly grotesque bloatware correlates with the arrival of the Internet. Applications started showing up that took 30 seconds to load. Versions of Netscape communicator began to approach the time it takes to boot the base operating system. More recent versions of Netscape took a massive 50 megabytes of ram to operate just to have the email application open. Java proudly announced that its sole purpose was to eat up as much processing and memory resources as Moore?s law would feed it for the sake of write-once-run-anywhere. Oracle 11i has a nasty habit of taking 30 seconds to load and swallow up 40 megabytes of ram for every window open. XML has a really nasty habit of using more than 1000% more memory and storage than their binary equivalents due to grossly excessive and useless metadata. With this in context, I long for the days of fast and lightweight Win32 applications.

      3. Apple stayed with the slow G4 processors because their suppliers were failing them. Apple does write good software and does many things well, but being friendly to old hardware is not one of them. Apple made two major evolutionary changes in its life time. First when abandoned all CISC based 68000 architecture, and second when it moved to the BSD based OS X. Both times, they abandoned all of the old hardware and forced entire populations in to obsolescence. Let?s not propagate revisionist history here.

      4. It looks like you may have bought in to the ?future proof? myth as far as your personal computers are concerned. I don?t ?chase? the performance curve; I ride it by buying 1 or 2 notches below state of the art every 18 months. I spend no more than $400 for an office computer and no more than $1000 for a powerful gaming computer. I recently built my daughter a nice $250 Sempron 2200+ based computer where the CPU and Motherboard cost me $49 combined. I?ve never bought in to the notion that you buy a $4000 computer and keep it as long as possible and fall 2 years behind the curve. I love Moore?s law and I?m sad to see it stall.
      george_ou
    • Performance comparison

      Microsoft and Intel/AMD must be doing okay on perforance, since I seen many head to head by CPU Mag and Maximum PC where the PC mostly wins.

      The dual G5 systems almost always loses to the Athlon system that cost the same as the dual G5, even with Mac optimized softare such as Photoshop.
      donf_z
      • Vendors fudge the numbers, but Apple is worst

        Although many vendors like to fudge the numbers, Apple is notorious for rigging bench numbers. They'll pull stunts like disable Hyperthreading on the P4 or they'll only test an extremely limited range of processing tasks that favor the Mac. In the past, they use to compare dual G4s to single chipped Pentiums.

        Both Intel and AMD has done some fudging of their own but not nearly as blatant.
        george_ou
        • Worst in whose book? Yours?

          <Although many vendors like to fudge the numbers, Apple is notorious for rigging bench numbers.>

          Sounds like you have an "ax to grind". No more so than InHell -- who's been claiming the prize in the Ghz wars for years even though AMD and PowerPC processors at lower speeds outperform them?

          <They'll pull stunts like disable Hyperthreading on the P4 or they'll only test an extremely limited range of processing tasks that favor the Mac. In the past, they use to compare dual G4s to single chipped Pentiums.>

          Practicing a little revisionist history yourself now, aren't you?? Apple DID publish test results showing PPC vs. P4 with and without Hyperthreading enabled. But, the reason was that PhotoShop had not yet been updated to use the Velocity Engine or even optimized for the multi-threading capability of the G4. So, Apple's intent was stated clearly: To level the playing field to account for software differences.

          When they compared dual-processor Macs to single-processor P4's, again the reason was to level the playing field; a dual 1.4 Ghz G4 could beat a 2.8 Ghz P4 in most tasks for about the same price.

          Apple was always HONEST about stating the facts -- and the G4 STILL beat the P4 -- even with Hyperthreading enabled on the P4. Let me repeat one point -- Apple clearly stated the test environment and included unfavorable results as well as results in their favor. I don't remember InHell ever doing that.

          All we've heard from InHell for the last ten years is that more Mhz is always better -- right up to the time they actually released the P4M which, by your own words, outperforms the P4 at half the clock speed.

          Where is InHell taking us now?? Multi-core and multi-processor machines. Where have we heard that before?? It would be funny if it weren't so ironic.
          macuser_z
          • also mine

            I also have been bothered by Apple's comparisions and have typcially ignored them and looked to the tech magazines to tell a more honest story. Granted I have not read every magazine, but I try to look at the Mac based ones in B&N while I drink a coffee if I see a good article.

            First Apple does not offer a multicore chip. Their dual processor system is not even a multichip module, but sepearate chips that have to use the mother board to communicate. That is slow. Granted the first version of the desktop dual chip from Intel is not the best implementation, but it is better than going thru the Mother board. YOu can buy dual chips sytems including chips from Intel and AMD, this is not what we are talking about, again old tech.

            Plus, everyone knows and talks about the P4's deep pipeline and why that allows the chips to run fast, but often not very efficently. That is old news and why AMD can smoke them.

            The G5 is based on IBM's Power 4 chips, with a Power PC hardware emulation built on to it. That is not the best technology. AMD's core is much better and give much better performance. When will IBM finaly get around to make something good for Apple? Next, look at the yield issues that IBM has had for the last few years and I would not expect them to be able to speed Apple chips up much. When is the G6 comming out? AMD is a partner with IBM, so they can get their good tech. AMD also knows how to yield chips, which IBM has never been very good at. I doubt IBM will ever win in this race.

            I have never heard of any multi core chips from apple, they must wait for IBM to get around to it.
            donf_z
  • At least it's gotten cheaper

    Part of what has driven Moore's Law is customer need. There has always been an appetite for power and throughout the lifetime of Moore's Law, a pronounced deficit between the needs of even the most basic users and what the processors could deliver. As demands in the most basic arena of word processing progressed from typewriter replacement to needing spell-checkers to special fonts to table formatting to integrated graphics, processor power progressed to fill each need. The GUI required a huge jump in processor/graphics card capabilities that really wasn't filled until the mid 90s. We saw another jump as we made our PCs a multimedia center. But with each power jump, we satisfied the needs of another segment of consumers such that 90 percent of users today don't require more than 3GHz. A few gamers, CAD users, modelers, and video editing fans are in the 10 percent still looking for more.

    But as we've hit the plateau, we can now replace those 3- or 4-year-old computers with shiny new 3GHz machines with plenty of memory and huge hard drives for cheap--sometimes less than $400. Many of us used to suffer for 5 years or more with slow, obsolete machines because computers were just too darn expensive to replace.
    Rodney Davis
    • No arguments there

      You're right Rodney. The prices have gotten cheaper and performance better for all these years. It's just hit an alarming plateau.

      You're right about the cheap 3GHz systems and I would always recommend replacing sub GHz machines in a heart beat for a new cheap 3GHz machine. The problem is that I can't recommend replacing the 3 year old 2.4 GHz machine and at some point, the performance better start going up in a hurry or else the buying will stop. You can re-label the processors with fancier names and larger model numbers all you want, but I can?t imagine that people are that stupid. Intel would like to get away from the GHz labels ASAP, but vendors are still putting up the GHz numbers because the public demands it.
      george_ou
      • Can Moore's law create new sales today.

        I can think of a few things that will get people to buy new machines, I do not know how improved performance can drive more sales today.

        I think latency is a bigger issue today. Why does it take so long to turn on or off a machine. Why does word & outlook take forever to open. Why does take so long to process video and to process my photos.

        Fixing these problems will drive sales.

        can others add more.

        1. New FPS Games will cause a small group to buy new machines.

        2. Video editing - This group is just starting to grow, I think and currently quite small as a real driver. Soon this will be a very important group. PLus, huge drive to improve the processing time and burn speed of video to disk. Intel and AMD should really be spending money here.

        3. Photography - This is a rapidly increaseing group. Could be a driver right now, but will they upgrade to improve thier experience and what might drive them to do that. I think low cost easy DVD burning might be a big selling point, you can burn your slide shows to disk and send them to grand parents and others. Most new machines today can do this, but those 2 years or older probably will not. A system really designed for viewing, saving, printing and improving photos might drive new sales.

        3. I would forget companies, they will upgrade when it cost more to fix the old mahcines than it does to lease new machines. So high service contracts will drive them upgrade. Other wise all they want employees to do is: email, create documents and presentations. Nothing new nor will it drive investment.

        3. The home market and Small Business especially professoraly home market are the real area that can be influenced today.
        donf_z