no, it really isn't

it's a conjecture, really

Isn't Moore's law about the number of transistors and not their size? What I mean is with 3D lithography they could increase the number of transistors even if they remain the same in size, no?

@TildalWave possibly, yes, but that would increase the "volume" of your CPU, instead of increasing its die size -- either way, the result is the same: if you kept transistor size fixed and wanted to keep doubling the number of transistors every so often, your TDP and heat output would go through the roof.

it doesn't matter where you put them or whether you separate it out into separate CPU packages or what: the main thing that's been driving the digital revolution is the shrinkage, not the number of transistors. the number of transistors becomes unmanageable if you leave size constant.

you'd have to have a CPU about a kilometer long on both sides (width and height) to have as many transistors as a Core i7 3770 using fab technology from the Intel 486

that's what shrinkage has done: it's prevented us from having to do that, and kept heat dissipation and energy usage manageable

if we stop shrinking, then if we keep doubling, within 20 years our CPUs will be a mile long

and use megawatts of electricity

@allquixotic that's assuming no advancements in materials used and that indeed everything but the number of transistors remains unchanged. but we're seeing constant advances in all areas of science touching the lithography process

@allquixotic heh, did you just do that math?

@TildalWave true, but no field of research has been able to incur exponential improvement of any variable factor that works in our favor except for fab size... that is the ONLY thing that has consistently been improved exponentially

also, as we're rather rapidly approaching physical limits of how much we can shrink a 'traditional' transistor, we're also at the same time approaching size where quantum physics becomes cheap and feasible... that will change pretty much everything about how we look at transistors

Let's say we put a bump in the road by reducing the amount of energy and heat dissipation required for a transistor, while maintaining its fab size, by a factor of three. Can you do that again and again every so many years and create an extension of Moore's Law based on a different variable (this time energy consumption and heat dissipation)?

I just haven't seen any evidence that anything in the history of man, not even anything in the early experimental stages, has been able to maintain an exponential improvement curve for any significant length of time, except for transistor fab size... aside from, well, population growth, but that's a bad thing

i meant quantum computing of course, quantum physics is obviously already here LOL

gotta go -- but I don't think anyone has a firm grasp on what we can do in terms of iteratively improving quantum computers, since we can barely create them and control them to begin with

@allquixotic cheers have a great one ;)

I'm not saying we won't ever find another gravy train where we can do mild R&D (mostly engineering, with some basic science) to exponentially improve some attribute of our computing devices for around 30 years; just that whatever it is aside from fab size isn't yet here

and the fab size gravy train is coming to a halt, soon

@AviD Avi isn't Intel building a 10nm fab in Israel?

@TildalWave probably, we get all the best fabs first.

It's only fair, really - most of Intel's best tech was developed in Israel.

@AviD Well if it wasn't for you guys they'd be still stuck with those netburst chimneys LOL

@TildalWave I think netburst was also developed here, back when it was considered a good thing.

seriously, though - name an Intel tech, odds are it was from here. Intel really should be an Israeli company.

of course, if it was, they would have no clue on how to market or sell the damn things.

@AviD It's still viable for future tech actually, materials permitting. It was capable of reaching far greater frequencies than current architectures

yeah, but at much less effective frequency.

although that's true for most lengthy pipeline architectures

shorter stages means greater frequency

but ends up having a lot more of them and harder to optimize... which doesn't help with branch prediction or stuffing a lot of simpler instruction parallelism like FPU

@AviD yes nowadays it's all about how quickly you can get in and out, not how much you can stuff in it at the same time :)

@TildalWave kinda the opposite of "what she said". Damn you.

@AviD If I had a female CPU I'd be still waiting for your response....

...you know, selective hearing ;)

Geez man you're old!!! What month were you born on? :)))

ahh, screw you, @RoryAlsop is the old man of the room

:)

@AviD Pfffffffffffffffffffffft