Intel: Fifty Years Under Moore's Law

[Guest]

Fifty years ago yesterday, Gordon Moore published a document in Electronics magazine titled "Cramming More Components Onto Integrated Circuits." The idea he put forth would become known as "Moore's Law."

Intel: Fifty Years Under Moore's Law : Read more

 

[CaedenV]

Moore's law is truly an astounding principal, but it died some 10 years ago and Intel keeps redefining it in a weird fettish of keeping it true at all costs. It use to be the number of transistors per dollar, but then we started having so many transistors that this could not be sustained. Then it was effective processing power per dollar, which pushed them through another few generations... until raw performance hit a brick wall. Now it is popularly defined as performance per watt per dollar, which works out well right now as effective performance is at a near stand-still, but they are achieving that performance at lower and lower TDP. But that is only going to work for a few more years until they are done with their die-shrink campaign. It will be interesting to see how it gets redefined again in the next 5-10 years so that Intel can continue to claim that Moore's Law is still somehow with us.

Don't get me wrong, the man was/is a visionary, and his 'Law' was totally unexpected but turned out to be very true for a very long time; truly a game changing paradigm in the industry. Intel just needs a new vision to follow... or perhaps just a little bit of competition from another chip maker in order to get them to make improved products again rather than sad incremental upgrades to existing products.

 

[aldaia]

A friend of mine, who works in the semiconductor industry, made another interesting prediction: "Intel will go bankrupt before recognizing that Moore's Law is over"

 

[sykozis]

Intel isn't going to admit that "Moore's Law" is over, until someone proves it publicly....

 

[anthony8989]

If you had read the first comment you would understand that Moore's Law ended a decade ago. They've been moving the goal post ever since.

 

[ykki]

Intel isn't going to admit that "Moore's Law" is over, until someone proves it publicly....

AMD already did that. http://www.tomshardware.com/news/amd-apu-efficient-goal-moore,27099.html

 

[right2myopinion]

Say what you want, but Intel CPUs are currently the best. If "Moore's Law" is what's driving them to achieve higher standards, then by all means continue because we (the consumers) greatly benefit from it. I don't care if it's Intel, AMD, etc., competition is great for business and innovation. At first it was clock speed, then multi core, can't wait to see what's next.

 

[bin1127]

Since moore's law ushered in the age of semiconductors, the law will probably only fade away when a new law predicting a completely genuine type of computer processing is possible. Maybe quantum computing will finally retire moore's law and a whole new era to innovate begins anew.

 

[beetlejuicegr]

Since moore's law ushered in the age of semiconductors, the law will probably only fade away when a new law predicting a completely genuine type of computer processing is possible. Maybe quantum computing will finally retire moore's law and a whole new era to innovate begins anew.

On the contrary, i believe that if a new technology of computing appear, the moore law will live again as the rise in performance every few years will skyrocket again.

Think about it, we say the moore law is now out of sync as newer chips are say declining in the performance acceleration of the last decades.

It is like when moore thought about his law. They were not making semiconductors as they were expensive right? And he did his research and foreseen that the new technology will rise etc.


So it's the same now. Semiconductors are going 3D etc etc etc but at some point things will change. Who knows, optical chips? Crazy stuff? crystal like chips? who knows, but when this happens , the performance will go acceleration mode again rising and rising till the next tech advancement.

 

[kinggremlin]

Moore's law is truly an astounding principal, but it died some 10 years ago and Intel keeps redefining it in a weird fettish of keeping it true at all costs. It use to be the number of transistors per dollar, but then we started having so many transistors that this could not be sustained. Then it was effective processing power per dollar, which pushed them through another few generations... until raw performance hit a brick wall. Now it is popularly defined as performance per watt per dollar, which works out well right now as effective performance is at a near stand-still, but they are achieving that performance at lower and lower TDP. But that is only going to work for a few more years until they are done with their die-shrink campaign. It will be interesting to see how it gets redefined again in the next 5-10 years so that Intel can continue to claim that Moore's Law is still somehow with us.

Don't get me wrong, the man was/is a visionary, and his 'Law' was totally unexpected but turned out to be very true for a very long time; truly a game changing paradigm in the industry. Intel just needs a new vision to follow... or perhaps just a little bit of competition from another chip maker in order to get them to make improved products again rather than sad incremental upgrades to existing products.

Not sure what you are talking about. Moore's law has NEVER been a performance measure. The actual observation that Moore made was that transistor count would double every 2 years in a dense integrated circuit. That's it. He made other observations in the same paper, but that's the only one that is labeled Moore's Law.

The only other interpretation of Moore's Law that I've heard is that performance will double every 18 months. This in fact was stated by another Intel exec, and was not a prediction made by Moore.

Having followed the industry for 30 years, I've never heard of any of the other interpretations you stated. TDP? What on earth, such a measure didn't even exist when Moore made his observation.

 

[CaedenV]

Not sure what you are talking about. Moore's law has NEVER been a performance measure. The actual observation that Moore made was that transistor count would double every 2 years in a dense integrated circuit. That's it. He made other observations in the same paper, but that's the only one that is labeled Moore's Law.

The only other interpretation of Moore's Law that I've heard is that performance will double every 18 months. This in fact was stated by another Intel exec, and was not a prediction made by Moore.

Having followed the industry for 30 years, I've never heard of any of the other interpretations you stated. TDP? What on earth, such a measure didn't even exist when Moore made his observation.

That is exactly my point! But if you read the Moore's Law PR crap that comes out of Intel every few years you will find that they redefine it every few years to suit their whimsey. Moore's Law is dead, at least for transistor technology, and has been dead for a while, but they go through great lengths to redefine it every few years just so they can say it is alive and well. Sad thing is that Intel is the only ones who really care if Moore's Law is still around and they are starting to look a little silly.

 

[aldaia]

Moore’s Law is primarily an economic law. It states that density at MINIMUM COST per transsistor will grow exponentially. Moore examined the density of transistors at which cost is minimized, and observed that, as transistors were made smaller through advances in photolithography, this number would increase at "a rate of roughly a factor of two per year". As originally stated by Moore, more's law already stoped working. It will still be possible to double the number of transistors but cost per transistor is not scaling anymore.

Despite a popular misconception, Moore did not predict a doubling of performance. David House, an Intel colleague, had factored in the increasing performance of transistors to conclude that integrated circuits would double in performance every 18 months. This "House Law" stopped working more than 10 years ago.

Annother misconception relates moore's law with performance per watt growing exponentially. This is actually know as Dennard scaling, after Robert Dennard (inventor of DRAM memory at IBM). Since around 2005–2007, Dennard scaling appears to have broken down.

Finally Rock's law states that the cost of a semiconductor chip fabrication plant doubles every four years. Rock's law is the reason why AMD had to spin-of Global Foundries. Doubling of transistors will eventually stop much before the physical and technologycal limits are reached, just because it would not be economycally feasible.

 

[Entangled]

The statement that a modern CPU has "more than 4,000 times the performance" is a gross understatement. You get that sort of number if you divide 3GHz by the frequency of the 4004 CPU which is 740kHz. However, the instruction set was MUCH more limited: The Wikipedia entry states that two 32 bit numbers could be added in 850 micro seconds, and multiplication would be much slower of course. Using 4 cores and AVX2 instructions, I suspect that 32 bit addition alone would be at least 100,000 times faster, while a general modern workload (using 64 bit numbers, multiplication and division) would be between 100,000 and 1,000,000 times faster.

 

[gmuser]

Moore's law is truly an astounding principal, but it died some 10 years ago and Intel keeps redefining it in a weird fettish of keeping it true at all costs. It use to be the number of transistors per dollar, but then we started having so many transistors that this could not be sustained. Then it was effective processing power per dollar, which pushed them through another few generations... until raw performance hit a brick wall. Now it is popularly defined as performance per watt per dollar, which works out well right now as effective performance is at a near stand-still, but they are achieving that performance at lower and lower TDP. But that is only going to work for a few more years until they are done with their die-shrink campaign. It will be interesting to see how it gets redefined again in the next 5-10 years so that Intel can continue to claim that Moore's Law is still somehow with us.

Don't get me wrong, the man was/is a visionary, and his 'Law' was totally unexpected but turned out to be very true for a very long time; truly a game changing paradigm in the industry. Intel just needs a new vision to follow... or perhaps just a little bit of competition from another chip maker in order to get them to make improved products again rather than sad incremental upgrades to existing products.

You probably misunderstood what Moore's law really says: transistor count in chip/CPU will double every two years or so.

You say that law died 10 years ago, and yet if we look at best mainstream CPU from 2004 (Pentium 4 Prescott with 112M transistors) and compare it to best mainstream CPU from 2014 (Core i7 Haswell with 2600M transistors), it shows that number of transistors increased about x1.9 every two years, meaning Moore's law is still alive.

Even derived (non-Moore's) law about performance doubling every 1.5 years is still alive, since in addition to number of transistors we still have performance increases due to shifts from 20nm to 14nm etc.

BTW, if you look at non-Intel chips, for example NVidia GPUs, you can also see that Moore's law was still valid in last 10 years: G70 from 2005 had 303M transistors, compared to GM200 Maxwell from 2015 with 8100M transistors. That amounts to x1.93 increase every two years.

Even looking at performance ot those NVidia cards, where later one has 6140 GFLOPS vs 160 GFLOPS in one from year 2005, making that x2.08 every two years in performance (or x1.7 every 18 months).

 

[sc14s]

A friend of mine, who works in the semiconductor industry, made another interesting prediction: "Intel will go bankrupt before recognizing that Moore's Law is over"

I sure hope not. We need all the competition we can get in the chip market.

 

[belardo]

There isn't any real MOORE's "LAW" anyway... because it can (and does) change depending on the skills and ability of engineers, manufacturing and technology.

There is the LAW of gravity. Because it works based on facts (and yes, there are a few holes in it) - but no matter, the laws of gravity is known and we can calculate rocket launches, etc because of it.

And of course we have laws on Hydrodynamics, Electromagnetism, Thermodynamics so on and so forth.

Both AMD and Intel have hit their prospective walls and the market isn't really dictating any true performance enhancements in the technology. Just portability... in a few years, the power of an i7-4770 should be able to fit in a cell phone.

But for desktop CPUs... The i5-2500 / 3570 / 4670 are all the same in general performance. Maybe a 2~3% difference between the 2500 and the 5670. The cost of the CPUs to consumers have remained the same ($200) - while die shrinks and reduced power means more profit for intel, less power requirements on our end (better battery, etc).

Operating systems haven't changed demands since Win7... So... unless we're doing homodecks or True VR... nothing is going to require more horse power for the consumer.

 

[ivyanev]

The law may be for transistors, but the spirit is for performance. Doubling the transistors won't guarantee doubling of the performance but usually it does.And while indeed the CPU performance is not climbing as fast as we want, the fact is that intel has managed to shift crap integrated graphics(on motheroard) to the CPU and even make it good enough(in the last 2 generations).