Moore's Law ... dead in the water now

[Reynod]

http://www.computerworld.com.au/article/423378/physicist_says_moore_law_collapsing_/?fp=4&fpid=18

He says 10 years and Moore's Law will have collapsed.

http://en.wikipedia.org/wiki/Moore's_law

Even Wiki is worried ...

Moore's law is a rule of thumb in the history of computing hardware whereby the number of transistors that can be placed inexpensively on an integrated circuit doubles approximately every two years. The period often quoted as "18 months" is due to Intel executive David House, who predicted that period for a doubling in chip performance (being a combination of the effect of more transistors and their being faster).[1]

The capabilities of many digital electronic devices are strongly linked to Moore's law: processing speed, memory capacity, sensors and even the number and size of pixels in digital cameras.[2] All of these are improving at (roughly) exponential rates as well (see Other formulations and similar laws). This exponential improvement has dramatically enhanced the impact of digital electronics in nearly every segment of the world economy.[3] Moore's law describes a driving force of technological and social change in the late 20th and early 21st centuries.[4][5]

The law is named after Intel co-founder Gordon E. Moore, who described the trend in his 1965 paper.[6][7][8] The paper noted that the number of components in integrated circuits had doubled every year from the invention of the integrated circuit in 1958 until 1965 and predicted that the trend would continue "for at least ten years".[9] His prediction has proved to be uncannily accurate, in part because the law is now used in the semiconductor industry to guide long-term planning and to set targets for research and development.[10]

This trend has continued for more than half a century. 2005 sources expected it to continue until at least 2015 or 2020.[note 1][12] However, the 2010 update to the International Technology Roadmap for Semiconductors has growth slowing at the end of 2013,[13] after which time transistor counts and densities are to double only every 3 years.

I say it is dead in the water now with Ivy Bridge ...

Its 5% faster ... its smaller ... sure ... they got more transisters on the beast.

But there is no way they have doubled the performance.

It cant run any faster because we can't drag the heat away from the chip ... why ... its physically getting too small to anchor a heatsink in the traditional sense.

How can we get around the problem?

How about heatsinks that wrap around the four sides and the top of the chip?

How about building a chip with a physical hole through the middle?

How about cooling both sides (top and bottom) and having all of the circuit pins around the sides?

Dealing with the thermals has suddenly become a major issue with the first of the 22nm CPU's ... it heralds the need to look at this issue in more depth.

Ivy Bridge really needed 12 cores without an APU clocked at the same rate as a SB CPU to fully comply with Moore's Law ... I have not seen that ... have you? That would have doubled the performance of the current mainstream CPU's.


We need to get Gordon out of retirement to do some more work ... instead of spending his 4Billion.

 

[Ubrales]

^ Thanks! Good reading!

 

[welshmousepk]

you are thinking far too mechanically.

cooling, as with everything else, also progresses. simple heatsink design is only the afterthought.

the real progress comes with the eventual mass production of graphene or silicine as a buffer for the transistors, or nanocarbon for making heat transfer much more efficient in small spaces.

in such small scales, its the small scal cooling ideas that will help.

and also, heralding the end of Moores law because a single company released an average processor is both narrow minded and precursory. i'm sure Haswell and all the other super secret chips in production will have more transistors than you care to count. at least i hope so.

 

[stingstang]

Transistors only just hit short of 50% more than sandybridge, and that is a year fron SB release...ish. You know what I mean. Moore's law has been slowing down sice 2008, but now with the intoduction of more sophisticated on-chip graphics, the visibility of this performance increase as a whole is harder to see.

 

[welshmousepk]

Transistors only just hit short of 50% more than sandybridge, and that is a year fron SB release...ish. You know what I mean. Moore's law has been slowing down sice 2008, but now with the intoduction of more sophisticated on-chip graphics, the visibility of this performance increase as a whole is harder to see.

exactly. even the 'number of transistors' approach becomes much harder to define with IB. technically, would tri-gate transistors not equate to a 33 percent increase in transistor area, even with the same number of physical transistors?

efficiency is much more important than quantity. and soon, quantity will count for very little. as long as PERFORMANCE increases at the same rate, Moore's law will never die. and you can't just look at a single release. like every law, there will be peaks and troughs, but the progression will always be fairly linear when looked at over a larger timescale.

 

[gamerk316]

Stop thinking about CPU's in the tradition sense; GPU's will keep Moore's "law" alive for a while yet, and as more workload moves more toward GPU's, CPU performance scaling will matter less and less.

From a total system computation standpoint, Moore's law is alive and well.

 

[popatim]

How soon befopre we dont even have a cpu anymore, just a gpu running it all?

 

[gamerk316]

^^ Never. GPU's are HORRID at tasks that do not scale, just like CPU's struggle with tasks that could be spread over many more execution units.

CPU's are optimised for doing a handful of things REALLY quickly. GPU's are optimised for doing a lot of things, but at a slower pace for each. One or the other is optimal, depending on workload.

 

[Reynod]

I really wanted to point out the issue with the size of the silicon getting small to the point where it becomes a major issue.

Perhaps at the next node down ... where doubling the transitors and trying to even maintain the current speed becomes impossible ... due to the size of the die ... physically how do you transfer the heat away from it?

http://en.wikipedia.org/wiki/Thermal_conductivity
http://en.wikipedia.org/wiki/Thermal_resistance_in_electronics


Derived from Fourier's Law for heat conductionFrom Fourier's Law for heat conduction, the following equation can be derived, and is valid as long as all of the parameters (x and k) are constant throughout the sample.

Ro = x / ak

where:

is the absolute thermal resistance (across the length of the material) (K/W)
x is the length of the material (measured on a path parallel to the heat flow) (m)
k is the thermal conductivity of the material ( W/(K·m) )
A is the cross-sectional area (perpendicular to the path of heat flow) (m^2)

 

[welshmousepk]

Now thermodynamics is certainly not my specialty, but is value K not something that can be (relatively) easily changed?

 

[Reynod]

http://www.extremetech.com/computing/129300-physics-ivy-bridge-and-the-slow-death-of-overclocking

 

[welshmousepk]

http://www.extremetech.com/computing/129300-physics-ivy-bridge-and-the-slow-death-of-overclocking

a very interesting read.

Although I am reminded of a quote i heard from a once famous scientist. He predicted that at the number of people and overall wealth in the town of london grew, within 20 years the horse manure would be 2 feet high on the streets.

The point is, there is always some problem that people foresee. And almost always it is sidestepped by some other innovation. We can't see that innovation now, but I am certain that before long we will see a new material to replace silicone (silicene graphene or something else i haven't even heard of yet) that will make the density and thermal capacity of silicone transistors redundant.

Hell, its possible to buy a working quantum computer right now. if you told people 5 years ago they'd have laughed at you.

 

[InvalidError]

due to the size of the die ... physically how do you transfer the heat away from it?

Simple: you graft more low-power circuitry like SRAM without increasing power-hungry Tag-RAM (which means wider cache lines) onto the die to act as thermal buffer / heat-spreader at the silicon level. If they run out of stuff to add to keep die size large enough to get the heat out, including blank/MLCC space is also an option.

 

[willzzz]

You know Moore's Law doesn't say doubling performance, only doubling transistor count. Not the same thing.

 

[Ubrales]

Correlated. High 'R-squared' value.

 

[InvalidError]

Correlated. High 'R-squared' value.

The correlation has been weakening at least as far as conventional CPUs are concerned ever since they plateau'd at 3ish GHz since the combination of clock speed and increased transistor count was needed to keep up with that performance curve.

But Intel and AMD's focus for CPUs is NOT performance anymore, it is performance per watt. Both companies are stepping back from the performance-doubling curve to favor efficiency.

 

[FinneousPJ]

It's not dead yet - I haven't done any calculations but I'm quite sure the transistor density is higher in IB than in SB, which is what Moore's law is about.

It is however true that at some point there's a physical limit to how tiny you can go. But we're not there yet 😉

 

[-Fran-]

Stop thinking about CPU's in the tradition sense; GPU's will keep Moore's "law" alive for a while yet, and as more workload moves more toward GPU's, CPU performance scaling will matter less and less.

From a total system computation standpoint, Moore's law is alive and well.

The correlation has been weakening at least as far as conventional CPUs are concerned ever since they plateau'd at 3ish GHz since the combination of clock speed and increased transistor count was needed to keep up with that performance curve.

But Intel and AMD's focus for CPUs is NOT performance anymore, it is performance per watt. Both companies are stepping back from the performance-doubling curve to favor efficiency.

Those two points.

When Moore's law came out, the paradigm was getting more transistors into a certain area surface to increase performance without thinking about the power in the context of CPUs, since it didn't matter at the time to look at other variables. The power issues that came on every new process made Intel and probably every manufacturer out there re-think the strategy in the long term. More over, a big shift in paradigm went with mobility a few years ago. Like IE (InvalidError) says, I'm sure ARM introduced another paradigm when their designs started to get serious performance and adepts in phones and tablets with no additional power requirements, and software development started to stagnate enough for low power, low processing alternatives to be successful. People now don't want to waste money on energy to have the best performance, I'm sure. Most of us are conscious of the perf/watt metric nowadays, in one way or the other (mobility is a given, but for enthusiasts, the lower power chip has the better OC potential, for example). AMD and nVidia introduced the second big shift or paradigm change with the GPGPU alternatives. The way CPUs work and use power is not always efficient for every task out there and Intel recognized this a few years ago. That will keep Moore's law alive for a while more and these changes in paradigm for development as well.

In another note, going smaller is not the only alternative for getting better and faster results. Software optimization (good arch of solutions, in particular) play a MASSIVE role in all this as well. In particular regarding Moore's law, current software doesn't push hardware hard enough to justify the desperate race in new process tech, but the perf/watt metric since everything is going mobile for the common user.

Cheers!

 

[Ubrales]

The correlation has been weakening at least as far as conventional CPUs are concerned ever since they plateau'd at 3ish GHz since the combination of clock speed and increased transistor count was needed to keep up with that performance curve.

But Intel and AMD's focus for CPUs is NOT performance anymore, it is performance per watt. Both companies are stepping back from the performance-doubling curve to favor efficiency.

Yes. Sooner or later all performance curves will have diminishing returns. That's ideally the time to adopt a radically different strategy and technology.