Forget Moore’s Law—Chipmakers Are More Worried About Heat and Power Issues

GoodBoy

GoodBoy

2[H]4U
Joined
Nov 29, 2004
Messages
2,752
Good short read: https://spectrum.ieee.org/view-from...ower-problems-might-drive-chip-specialization

Article said:
John Hennessy—president emeritus of Stanford University, Google chairman, and MIPS Computer Systems founder—says Moore’s Law “was an ambition, a goal. It wasn’t a law; it was something to shoot for.”

“It is definitely slowing down,” he says, “but to say it’s dead is premature.”

That slowing, at this point, isn’t his biggest concern. The real problem, Hennessy says, is the failure of Dennard scaling, an observation that as transistors get smaller and circuits become faster, a chip’s power consumption stays the same.

“Who would have thought,” he says, “that microprocessors would have to slow down clock speeds or turn off cores to keep from burning up?” Hennessy spoke as part of a panel at a Churchill Club forum held Monday in Menlo Park.
Click to expand...
 
As an IT Professional I've seen Moore's law is going to die more times than I can shake a stick at. I mean seriously I've seen this come up as a news item AT LEAST once a year since 2000. (apparently my stick shaking limitations are a lot lower than I expected them to be.)
 
I think the first talk I ever heard of it dying was 2008. First signs were transistors getting so small that the quantum effects starting affecting their operation. And if you go by the letter of the "law" the 2 years doubling is in fact already "dead".
 
As a person who has watched industry over pro-longed time, i can say with certainty that Moore law is not dead.

Ones (in tech industry) stating that it is; Hope they are going to follow through on that - fall over and die.

and i quote actual leader in the field Jim Keller

"Moore's law is not dead but if you think so you're stupid"
He asserted that Intel can keep it going, and supply tech companies ever more computing power.

Keller said that using nanowires and stacking his team had mapped a path to packing transistors 50 times more densely than possible with Intel's 10 nanometer generation of technology. "That's basically already working," he said.


The Moore Law was observation rather than 'law' of how fast industry advanced.
Industry doesn't need to slow down at all, companies that see it too expensive or want to stay with silicon are limited. Intel, IBM, AMD, and few others have already mapped out alternatives that can scale and provide much greater powers than we have now.

btw. how does TR zen2 scale out

6ricxa.png
 
Last edited:
GoodBoy said:
I think the first talk I ever heard of it dying was 2008. First signs were transistors getting so small that the quantum effects starting affecting their operation. And if you go by the letter of the "law" the 2 years doubling is in fact already "dead".
Click to expand...

Watch the Keller video, he dates the sentiment back a long ways.
 
I've also heard repeatedly over my career that silicon itself was a dead tech. I heard it first at the 50 MHz mark. It popped up a couple times, went away, and there are more rumblings again. It will of course be true at some point, but various optimizations and changes (doped silicon? That's cheating!) keep it alive.
 
cyklondx said:
As a person who has watched industry over pro-longed time, i can say with certainty that Moore law is not dead.

Ones (in tech industry) stating that it is; Hope they are going to follow through on that - fall over and die.

and i quote actual leader in the field Jim Keller

"Moore's law is not dead but if you think so you're stupid"
He asserted that Intel can keep it going, and supply tech companies ever more computing power.

Keller said that using nanowires and stacking his team had mapped a path to packing transistors 50 times more densely than possible with Intel's 10 nanometer generation of technology. "That's basically already working," he said.


The Moore Law was observation rather than 'law' of how fast industry advanced.
Industry doesn't need to slow down at all, companies that see it too expensive or want to stay with silicon are limited. Intel, IBM, AMD, and few others have already mapped out alternatives that can scale and provide much greater powers than we have now.

btw. how does TR zen2 scale out

View attachment 192355
Click to expand...
Show the task manager window
 
If it has slowed, i.e. taking longer than 2 years to double the density, then it's dead. Moore's law no longer being realized does not mean progress stops. It means it is happening at a slower pace.

cyklondx said:
Keller said that using nanowires and stacking his team had mapped a path to packing transistors 50 times more densely than possible with Intel's 10 nanometer generation of technology. "That's basically already working," he said.
Click to expand...

I hope so, and that sounds interesting. Time will tell.
 
How is stacking making things ' denser' stacking is akin to fold something, not actually making it "denser". Don't get me wrong, in the end who cares how much silicon is needed to achive more and more performance.. its not an issue anymore, chips are small enough already you can keep upping the die sizes stacked or otherwise without issue...
It does bothers me the bull, of saying its not dead.. its evolved!... Yeah ok. Semantics I guess..
Don't see the need to do that bullshit, and I don't see a failure if we go to 5nm at most for the next 30 years, and just keep upping the effective die size (again stacked, chiplet or mono, doesn't matter) to increase performance.
 
Uvaman2 said:
How is stacking making things ' denser' stacking is akin to fold something, not actually making it "denser". Don't get me wrong, in the end who cares how much silicon is needed to achive more and more performance.. its not an issue anymore, chips are small enough already you can keep upping the die sizes stacked or otherwise without issue...
It does bothers me the bull, of saying its not dead.. its evolved!... Yeah ok. Semantics I guess..
Don't see the need to do that bullshit, and I don't see a failure if we go to 5nm at most for the next 30 years, and just keep upping the effective die size (again stacked, chiplet or mono, doesn't matter) to increase performance.
Click to expand...


Moore law states that the number of transistors on a microchip doubles every two years.
A microchip refers to integrated circuit / a set of electronic components on a single unit. Thus stacking, 3d transistors, chiplets or size still apply without any issue. No changes to semantics.

Silicon might appear to be nearing its limits for the moment, (even so TSMC and Samsung completed their 5nm EUV Sillicon, and 3nm is in dev (still silicon) - but other technologies other materials are there capable of much higher frequencies, lower power usage, lower heat output.
Those would break the moore law, as for indium antimonide you could have have a chip with 140nm lithography with P4 class amount of transistors outperforming current chips couple times - simply because they would be clocked at 200+GHz under 0.5v. That would make little sense.

How is moore law dead or dying?

Is moore law referencing to actual performance? Yes, maybe, not always. IBM Cell processor platform is a proof of that.
We can see transistors steadily rising, if we were to follow moore law of every 2 years doubling of transistors (since 1971) we would be at, and real world microchips would be ahead or behind their time by moore law:

Year, Transistor Count (2x rise per moore law), % of Actual Transistor Count by that year (we are = + ahead of time, - behind)
1971, 2250, 0%
1973, 4500 +144%
1975, 9000 +18%
1977, 18000 -63%
1979, 36000 +47%
1981, 72000 -5%
1983, 144000 -7%
1985, 288000 -4%
1987, 576000 -4%
1989, 1152000 +2%
1991, 2304000 -70%
1993, 4608000 -48%
1995, 9216000 -67%
1997, 18432000 -84%
1999, 36864000 -34%
2001, 73728000 +61%
2003, 147456000 +22%
2005, 294912000 +50%
2007, 589824000 +65%
2009, 1179648000 +61%
2011, 2359296000 -3.7%
2013, 4718592000 +5.9%
2015, 9437184000 +5.9%
2017, 18874368000 +1.7%
2019, 37748736000 -15%

So are we in a bad spot? No.
2019 is a slow year so far.

built using transistor count data from
https://en.wikipedia.org/wiki/Transistor_count
 
“It was scary stuff, but radically advanced. I mean, it was smashed, it didn't work, but...it gave us ideas, took us in new directions. I mean, things we would have never…All my work was based on it.” — Terminator 2
 
The problem I see with 3d stacking is that it doesn't really solve the issue of heat / hot-spots and may actually compound it, this is me just thinking out loud for cpu/gpu which are the two types of chips currently that consume the most power. For most everything else, sure it should be good, we have seen it on hbm & hbm2 already, maybe you will have to active cool more parts of your system in the future, but in theory everything should attain a smaller footprint.

^^this semi rant is about the option that Jim Keller said that Intel had to pack transistors up to 50x more densely than in their 10nm plans.

So I feel that indeed heat and power will still be a big issue.
 
Revdarian said:
The problem I see with 3d stacking is that it doesn't really solve the issue of heat / hot-spots and may actually compound it, this is me just thinking out loud for cpu/gpu which are the two types of chips currently that consume the most power. For most everything else, sure it should be good, we have seen it on hbm & hbm2 already, maybe you will have to active cool more parts of your system in the future, but in theory everything should attain a smaller footprint.

^^this semi rant is about the option that Jim Keller said that Intel had to pack transistors up to 50x more densely than in their 10nm plans.

So I feel that indeed heat and power will still be a big issue.
Click to expand...
I personally feel that going for nanowires or fiber etc is just a marvel - but wouldn't make a good cpu/gpu with our current tech. Jim is taking us for a ride with nanowires.

We'd be better to look at
indium antimonide - we have exp printing transistors with it.
gallium arsenide - ~100GHz realistic
indium gallium arsenide - realistic more expensive needs more research
gallium antimonide - THz realistic expensive needs more research
zirconium dioxide - THz realistic expensive needs research
 
cyklondx said:
I personally feel that going for nanowires or fiber etc is just a marvel - but wouldn't make a good cpu/gpu with our current tech. Jim is taking us for a ride with nanowires.

We'd be better to look at
indium antimonide - we have exp printing transistors with it.
gallium arsenide - ~100GHz realistic
indium gallium arsenide - realistic more expensive needs more research
gallium antimonide - THz realistic expensive needs more research
zirconium dioxide - THz realistic expensive needs research
Click to expand...
#2 and #3 are used in laser and led diode sources. Industrial cpu development based there would make a lot of sense.
 
cyklondx said:
I personally feel that going for nanowires or fiber etc is just a marvel - but wouldn't make a good cpu/gpu with our current tech. Jim is taking us for a ride with nanowires.

We'd be better to look at
indium antimonide - we have exp printing transistors with it.
gallium arsenide - ~100GHz realistic
indium gallium arsenide - realistic more expensive needs more research
gallium antimonide - THz realistic expensive needs more research
zirconium dioxide - THz realistic expensive needs research
Click to expand...

I agree wholeheartedly that the only real solution to the power consumption/dissipation problem is the jump to different materials, the issue is finding them economically viable for mass production, the cheapest options are the two louder ones of graphene + phosphorene, but they have massive issues and their performance is much smaller than those that you mention (while still being an order of magnitude better than our current solutions).
 
Same. It will likely take different materials to make the next big jump.

Different materials going faster, means the industry has a bright future.

Moore's law would still be dead though. The law is transistor count.

Chances are using a different material like Gallium Arsenide (if all of the technical issues could be surmounted) would result in faster but also larger transistors... so the transistor count could very well progress backwards, while effective computational speed moves ahead.
 
Moores law is not a law. It's not a scientific principle based on scientific method.

It's just some dude at Intel that said blah blah double every other gener.. blag blah

And since only 2 chip companies that matter exist for the sake of the discussion, there are nothing or none other to challenge the establishment of AMD and Intels adherence to nothing more than a market prediction for chip development.

Newton's observations are law.
Einsteins observations are law.
We have associative, commutative, and distributive laws in math and countless others
Avagadros observations are law.
Hookes observations are law.
Coulomb
Pascal
Bernoulli
Law of angular momentum
Law of inertia
And on and on and so forth

But Moore is not a law by anything more than name and market predictions about a product segment.

If the power grid died and all computers ceased to function all the above laws would still apply to the world, but Moores so called law died with the computers.

And I wrote this reply because some people actually think that Moores law is an established scientific law because it's been pushed by so many as some universal galactic truth that applies to every atom in existence because they themselves have failed to learn the basic sciences so they dont know what a real law looks like.

Moores law is dead because it was never a law to begin with. So argue all you want but dead is dead.
 
Last edited:
lol...

I suppose someone could think it is a law in that respect. But it being pronounced "dead" should have clued them in that it is not the same as natural laws...
 
Moore's law is about cost of manufacturing, not density.
 
I am not. Further, Moore's Law only set out to predict the market from 1965 to 1975.
 
Source?

The number of transistors in an IC was doubling yearly from 65 to 75, then slowed to every 2 years and he revised it:

Moore's law is the observation that the number of transistors in a dense integrated circuit doubles about every two years. The observation is named after Gordon Moore, the co-founder of Fairchild Semiconductor and CEO of Intel, whose 1965 paper described a doubling every year in the number of components per integrated circuit,[2] and projected this rate of growth would continue for at least another decade.[3] In 1975,[4] looking forward to the next decade,[5] he revised the forecast to doubling every two years, a compound annual growth rate (CAGR) of 41.4%.[6][7][8]
Click to expand...
 
If you look at per-core performance, it's not exactly been looking good for a while. A big reason why everything shifted to more cores, because we have really been beyond the low hanging fruit with regard to single core performance for a while.

Now, Moore's Law doesn't explicitly describe performance levels (Moore's Second Law deals with cost of production, btw), just transistor density. But historically, that correlation had always existed... and continued to up until around the early 00's. Moore's law continued (transistor density does continue to increase), but the per-core performance and price/performance no longer followed the same trend... largely due to the fact that advances in parallelism doesn't track the same as linear core performance.
 
Another sign we've all seen is that the chipmakers' are running them closer and closer to the edges of their performance in both speed and temp (see AMD Navi, Ryzen).

Axman I'll take that to mean you found you were wrong.
 
You can believe any magical thinking you want.
 
Axman said:
You can believe any magical thinking you want.
Click to expand...


You're only half correct. Which also means that you are half wrong.

If you're stating that is explicitly only about cost of manufacturing, then you are entirely wrong.
 
It's the relationship between manufacturing cost per component compared to the number of components per IC.

Not density.
 
Moore's law is not dead, but it looks like it is for two reasons:

1. Costs per-wafer are going up for more advanced manufacturing nodes, while the density and power improvements are going down. This means either larger die sizes, or more expensive process nodes to make things work, so chip prices will rise slowly.

2. We're hitting the limits of wafer size with 300mm (we were promised 450mm, but that's dead). So there's no real way remaining to reduce the costs of these complex process nodes which means higher costs going forward.

And although you can go 3D like NAND, it's more expensive for complex logic circuits, so it's a last resort.

Moores law if almost at an end for CPUs, but we have some tricks up our sleeves, like AMD's Chiplets. But how far that design can scale is anyone's guess..

Intel can only shove so many cores into a 14nm die while still making money, so even they are stuck with multi-core for the high-end.

Moores law is almost at an end for high end CPUs. You can only make die sizes so much larger, and the logic scaling is getting worse with every manufacturing node. I'd say we've got another 10 years before you hit a real wall on Doped Silicon, and multi-chip hits it's scaling/price limits.
 
Last edited:
We have 64 core processors now that could run over a hundred virtual computers. An iPhone can emulate an entire Windows 95 computer. Your modern PC can emulate an Xbox 360. Who cares what Moore's Law says. CPU advances have been insane.
 
You must log in or register to reply here.
Back
Top