AMD sees the era of Moore's law coming to a close

AMD sees the era of Moore's law coming to a close

Summary: Moore's law has remained relevant for over half a century, but evidence is mounting to suggest that it is coming to an end. What will the end of Moore's law mean to the tech industry?

SHARE:
TOPICS: Processors, Hardware
90

Shifting from 28-nanometer to the smaller 20-nanometer architecture has taken longer than AMD had hoped, and this could signal the beginning of the end for Moore's law.

(Image: AMD)

Back in 1965, Intel co-founder Gordon Moore noted that the total number of transistors within integrated circuits had doubled approximately every two years since 1958, and he predicted that this trend would continue "for at least 10 years". It continued a lot longer than that, up until 2010, in fact, when the International Technology Roadmap for Semiconductors finally started seeing evidence that the pace was slowing down at such a rate that by the end of 2013, it would take three years for the transistor count to double.

Now an industry insider is seeing the same thing. Speaking to The Inquirer, John Gustafson, chief graphics product architect at AMD, claimed that Moore's law is hitting the buffers because the law was always about the economics.

"You can see how Moore's law is slowing down," said Gustafson. "The original statement of Moore's law is the number of transistors that is more economical to produce will double every two years. It has become warped into all these other forms, but that is what he originally said."

And Gustafson should know what Moore said, because the phrase "Moore's law" was first coined by professor Carver Mead at Caltech, and Gustafson was a student of Mead's.

"We [AMD] want to also look for the sweet spot," said Gustafson, "because if you print too few transistors your chip will cost too much per transistor and if you put too many it will cost too much per transistor. We've been waiting for that transition from 28nm to 20nm to happen, and it's taking longer than Moore's law would have predicted.

"I'm saying you are seeing the beginning of the end of Moore's law."

Does the end of Moore's law mean that the sky is going to come crashing down on us? Well, yes and no. The PC industry is in terrible shape, and anything that comes along to put the brakes on progress is not good. A slower hardware upgrade cycle will mean that there will be less in the way of new CPUs and GPUs to tempt buyers.

The end of Moore's law is not good for the PC.

That said, CPU and GPU performance is already at the point where it offers more power than most users know what to do with. Silicon already comes with more cores and more threads than most applications can handle. What people want more than a faster CPU or GPU is a cheaper CPU or GPU.

Consumers are also turning their backs on traditional desktop and notebook PCs, and instead focusing on mobile devices such as smartphones and tablets. While there's a drive here to make silicon smaller and more compact, it is primarily driven by the desire to reduce battery consumption. While a breakdown of Moore's law might slow down progress here somewhat, users are far less likely to notice because the sale post-PC devices have rarely focused on the speed of the silicon.

Moore's law has served us well, and rather than be surprised that its era is drawing to a close, I for one am surprised that it reigned for as long as it did.

Topics: Processors, Hardware

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

Talkback

90 comments
Log in or register to join the discussion
  • Moore's law is more relevant than ever

    This is a classic AMD defense pitch. They are having issues with creating denser silicon but the users are demanding more, they hardly play in laptops and tablets, for a reason.

    Moore's law on desktops is a mute point, yes consumers can't consume all the performance these chip provide, but consumers want desktop performance in a tablet or touchscreen laptop. In this growing market processors can't keep up and so the demand is going to increase still further.
    dezeit
    • Funny how AMD sees it coming to an end quicker than intel does.

      Meanwhile intel had no trouble with 20nm and is not having any trouble with marching on to 14nm and 10nm either.
      Johnny Vegas
      • amd?

        just cause this poorly run semiconductor company says so doesn't make it so. Ask intel next time, you know, they company moore was at when he made this prediction
        dosmastr@...
        • Actually More Was Not at Intel

          Moore wrote the article in Electronics magazine in April 1965 while at Fairchild.

          Moore co-founded Intel with Noyce in 1968.
          Patrickgood1
      • Really

        Not sure where you get your story. Frequently Intel puts out FUD, so folks are not really aware. The equiptment for manufacture is made by 3rd parties as well, so AMD and Intel have access to the same process facilities.

        At some point even mighty Intel will have issues with scaling down. Eventually with all the creativity in the world, the laws of physics will prevent further scaling down.

        Actually - 10 years ago the doubling occurred every 18 months, it has been slowing ever since then.
        wiseoldbird
        • Wrong

          Moore's Law is still "The number of transistors incorporated in a chip will approximately double every 24 months."

          I has never "slowed" and is expected to continue for 10+ years.

          Intel has a new 22nm 3D Tri-Gate transistor that AMD does not have. May 2011 Intel announced they were rolling out this new transistor Gordon Moore (as in Moore's Law) said:

          "For years we have seen limits to how small transistors can get," said Gordon E. Moore. "This change in the basic structure is a truly revolutionary approach, and one that should allow Moore's Law, and the historic pace of innovation, to continue."

          Doubling occurred every year from 1965 to 1975. Moore revised it in 1975 to doubling every two years. It was David House that that said 18 months, not Moore.

          This 22nm process is currently used in Ivy Bridge processors. When its use trickles down to the Atom, the ARM will be history.
          Patrickgood1
        • That is not correct.

          Intel uses its own process and it has been estimated that it is 3 years ahead of it next nearest competitor (not AMD BTW) in that process. Manufacturing equipment is custom built for Intel.
          DevGuy_z
      • Yes but it will impact Intel too.

        The rate of change was part of Moore's law and we are already stretching it now in the sense that we aren't currently doubling the transistors every two years anymore. The "approximately" part of the statement is being stretched quite a bit now. We may start seeing 3 years instead.
        DevGuy_z
    • Two things..

      It's 'moot' not 'mute'.

      Second, Moore's law isn't really relevant with ARM processors because they *started* with smaller transistor counts in order to save power. They can artificially follow Moore's Law for a few generations as they get beefier (and draw more power) and then they'll hit the same wall Intel and AMD processors have hit.
      TheWerewolf
      • Smaller transister counts mean less power consumption ...

        ... but they also mean less processing power. ARM-based operating systems are less powerful out of necessity. That's fine for consumers because they don't need all the extra power provided by preemptive multitasking operating systems but the power required to deliver to consumers what they want/need on a mobile device still has to come from somewhere ... usually a large-scale data center with racks and racks of high-performance processors.

        Say what you want about the power requirements but the x86 architecture has been remarkably successful for decades and will continue to be for some time to come. If not in personal devices, certainly in the data centers serving those devices.
        M Wagner
        • Battery Size and Power Consumption

          In mobile devices it is about the battery and power consumption. They can put powerful 8 or 16 core ARM processors in them but they might only run 2 hours on a charge. A typical user wants it to last a day at least, so they have some novel methods of shutting some of the cores down when the processing demand is low. That is where BB and QNX might get an advantage because it excels at this core usage threading.
          bigpicture
      • yes

        This is whats happening now, because ARM processors are getting bigger and bigger, they are not better designed, they are just low power. Desktop CPU's are already there. The only difference is they are built from the bottom up so they can do less power and less speed easier than the desktop CPU's which are built for max power.
        Jimster480
      • ARM does not have their own Moore's Law

        ARM has no foundry and produces no chips. They license the ARM architecture and Instruction Set to other companies like Intel, Apple, Samsung.

        Putting that aside Moore's Law referrers to the largest chip being manufactured by anybody in the Industry. There is no Artificial implementation.
        Patrickgood1
    • Exactly

      People are looking at the fastest core of a desktop and noting that it hasn't gotten much faster in the last three years or so.

      Nevermind that mobile chips have been making leaps and bounds. If you can do the same amount of work using half the power in half the size, then Moore's law is just taking a detour.

      Furthermore, software development hasn't caught up with multiple cores yet. When easy multi-threading tools are finally available all those extra cores that have been building up will suddenly be useful.

      While there may be limits to what we can do with known physics we aren't even close to reaching those yet. We are just close to the limits of what our existing designs can handle with known physics. And we've been there before.

      Not to mention that there is still a lot of unknowns in physics which can greatly affect processing power. Figure out how to utilize closed time-like loops and a Commodore 64 has effectively infinite processing power.
      SlithyTove
      • Desktops have Not Gotten Faster in the Past Few Years????

        Have you been living under a rock?
        Intel Nehalem, 45nm,
        Sandy Bridge 2011 32nm,
        Ivy Bridge Mobile i7 2012 22nm.

        Every year Intel makes significant improvement to the architecture as in execution units, branch prediction, cache management, micro instruction execution, Turbo-Boost, Hyper-Threading and etc.
        Patrickgood1
        • Sorry to spear a sacred cow of yours....

          But no, no rock. You will note that I said they have not gotten MUCH faster.

          Short of some VERY expensive and rare beasts my now "ancient" 1st gen Core i7 system is still only about 30% short of the fastest desktops on the market.

          That's well off the mark of the > 400% gain expected by Moore's law over that timeframe.

          Intel owns the top end of desktops right now and is rightly focusing on bringing down size and power usage to combat the creep of ARM.
          SlithyTove
    • no

      its just that everyone is doing independent research. What about when Intel's P4's sucked for a bunch of years? Thats what happens when companies don't share research. The thing is that AMD is making super efficient chips that stomp the crap out of Intels offerings at everything but the bleeding edge in both price and overall performance.
      Jimster480
      • 10 years ago

        That comment had some validity 10 years ago. Today I don't see AMD competing anywhere except basic game machines, the only places the APU wins. There are APUs getting Passmarks of 6000 that aren't much faster than the latest Atom for calcs. Vastly faster on video yes, but not if your are trying to run SQL or do a 30 page spreadsheet.
        mswift@...
        • He's right

          You're living under a rock. Gaming CPU under $100 -- all AMD (Nothing still REALLY competes with Phenom at that price range). The APUs are amazing. I've yet to try a game where I didn't achieve higher FPS than expected/reported.

          They're also making the cpu for the next gen. consoles.

          Like he said, other than the higher end machines... it's all AMD or older Intel processors.
          Kalan Petty
          • v

            ps. Did I mention how ridiculous the Price to Performance ratio is when comparing Intel to AMD?

            Just like it's been for a long time, AMD is more bang for your buck.
            Kalan Petty