Moore's Law is misunderstood?

DeLaTroY

Moore's law is always one of those things that's bothered me over the years because its seems on the ground that "power" isn't in fact doubling every 18 months or so. I've even heard on occasion that it's accelerating from 18 down to 12 months or whatever however even on a 24 or 48 month timescale, I don't see how it's even anywhere near a doubling..

I know there's a ton of different technologies that it applies to however I would have thought that desktop CPUs would have been a reliable cornerstone to test Moore's Law over time.

I have the Intel i7 980x (http://ark.intel.com/products/47932) which was released in March 2010. It's a 6 core CPU that I use mainly for video encoding.

Cinebench is the standard for benchmarking multicore intensive tasks like encoding. The 980x scores around 9.0 at the 3.33ghz stock clock speed.

November 2011 or 19 months after the 980x was released, the 3960x - the natural successor ("extreme edition") to the 980x was released which also sported 6 cores. It scores around 11.3 at stock.

Roll on September 2013 and the 4960x was released which again was an 6 core extreme CPU. It scores just 0.7 more over the 3960x in Cinebench or 12.0 at stock speeds.

So basically what we're looking at is a mere 33% increase in performance (9 -> 12) in Cinebench over the space of a massive 42 months. Where did the 100% in performance in 18 months that Moore promised me go? According to what's generally assumed when thought about Moore's Law, the 4960x should be scoring 20.7 by now.

My question is where is Moore's Law in practice here?

Looking up Moore's Law in Wikipedia we find: ".. that the number of transistors on integrated circuits doubles approximately every two years."

Further: The period is often quoted as 18 months because of Intel executive David House, who predicted that chip performance would double every 18 months (being a combination of the effect of more transistors and their being faster).

So Moore is saying transistors doubling in 24 and House is saying performance doubling every 18.

Number of transistors:
980x: 1.17bn @ 32nm
3960x: 2.27bn @ 32nm
4960x: 1.86bn @ 22nm

So it's true that we've seen almost a doubling of transistors in 19 months but nothing like that in performance so wtf? My point is that Moore's Law surely only matters and the endlessly touted utility it some day will bring to all ("the singularity", amazing price / performance, etc) if performance is physically realised!

The number of transistors has little bearing because most people when talking about Moore's Law are referring, or more accurately I suggest, are wrongly assuming a doubling in physical performance.

What am I missing in all this Moore's law, singularity malarchy? Maybe people are confusing a doubling of transistors to mean a doubling in performance? Or is there a real doubling in performance (price, size, power consumption, etc) occurring elsewhere beyond the realm of a Cinebench score that matters?

Turtwig

Moore's law isn't an actual physical law. It's more an observed trend and set of desirable goals for the Semiconductor industry. If you take it literally then you're going to end up with a situation where the average power density of an processor is greater than the photosphere of the sun! Now, obviously that cannot be allowed to happened as consumers would be rather pissed! So engineers have to take steps to ensure this doesn't turn out to be the case.

Moore's law has a limit to it. It won't be true forever. When we'll reach that limit though is still far from clear. Suffice to say for now it's more used to guide research and development into steering clear of potential problems when it comes to fabrication. While also setting performance goals to try to attain. Such improvements in performance though aren't necessarily ever going to be guaranteed. But they are goals the industry believes it should be striving for.

tl;dr Moore's law is pretty much the core of Semiconductor industry's roadmap. It is not however an actual law stating that the future will follow an exact trend.

TheChizler

Exactly, it was originally an observation of a trend when using a certain set of available technologies. It continues if you include power consumption halving every 18 months, transistor frequency doubling, chip area doubling etc. Microelectronic designers strive towards it as a goal, it forms part of the general industry roadmap.

Edit: The ITRS website is worth a look.

Simi

Well for most of the 90's/00's there was quite a large performance difference between each CPU generation right up until the core 2 duo. After that CPU performance enhancement sort of tapered off.

I've no doubt that if they really wanted to Intel could double CPU performance every 18mths, but the resulting chips would require ever increasing amounts of power.

Fact is the CPU is no longer the bottleneck in the majority of consumer level PC's. For gamers, a better graphics card will be of far more benefit to them than a better processor. Upgrading a mechanical hard drive to an ssd, will give a much greater performance increase than a CPU upgrade etc.

I also think mobile processor/graphical capability is currently a better reflection of the pace of technological improvement. The rate of improvement each generation is much closer to Moore's law. Of course that's not really new development, more trying to figure out how to downsize what already exists so that it fits in a tiny device and consumes a fraction of the power.

Still I think the improvement in mobile capabilities is in keeping with the spirit of the law, so it's not dead just yet. But the technological cliff is looming.

DeLaTroY

I know Moore's Law isn't a physical law.. lol My point is how it's misunderstood as a doubling in performance instead of transistors.

Simi wrote: »

Still I think the improvement in mobile capabilities is in keeping with the spirit of the law, so it's not dead just yet. But the technological cliff is looming.

In the silicone paradigm yeah where the scales get down to levels where they become too small for electrons to operate on but there's many replacements that have been put forward like carbon.

Approaching this size barrier and clock speed limitations (physical thermal limits) is why we've been going down more and more of the multicore path for some time however this introduces all sorts of new problems in leveraging all the theoretical power available.

Last I heard, we're a long way off the physical maximum amount of processing potential - like 200 years of continued exponential growth but I can't remember the details of how. Not sure how quantum computing fits into the Moore's law trend either though it could only be used for a narrow set of tasks anyway.

SteelyDanJalapeno

I think you need to take Moore's Law with a pinch of Salt......

longhalloween

DeLaTroY wrote: »

Moore's law is always one of those things that's bothered me over the years because its seems on the ground that "power" isn't in fact doubling every 18 months or so. I've even heard on occasion that it's accelerating from 18 down to 12 months or whatever however even on a 24 or 48 month timescale, I don't see how it's even anywhere near a doubling..

Was it to do with power?

I though it was the number of transistors on a given space would double, or the space would half?

yawhat!

We are the peak of processing power. I don't see any 8GHZ processors coming out anytime soon or anything too extravagant in the next two years!

ED E

Silicon is coming to its end, going down into the single digit nm die process is going to very tricky. Thats why graphene is being researched, if that becomes the next generation of chip fabrication then Moore's law will no doubt get a new lease of life.