14nm heat?

[Guest]

i was reading this cnet article about Intel's new Core-m broadwell lineup  (not endorsing, i think cnet is the worst) and this comment

 

 

reminded me of the whole process shrink vs. heat debate. does heat reduce with the process size or the opposite?

[Tea1337]

The temps will go down, but to get the same amount of performance you need more power going in, but with a die so small adding more power, it will likely run cooler than Haswell.

[JAKEBAB]

The temps will go down, but to get the same amount of performance you need more power going in, but with a die so small adding more power, it will likely run cooler than Haswell.

I thought it was suppose to consume less power dafuq?

[helping]

I think heat might go down, but thermal density goes up (ergo hotter) but who knows what sorcery Intel does. 

 

You're forcing less power into it, but at the same time into a smaller space. 

Consider that if the heat from the flame of a candle were evenly spread across a 3m square area, it might be tolerable. If the heat from the flame of a single candle were evenly spread across a 3mm square area, it would probably hurt. 

 

disclaimer: this is uneducated babbling 

[Tea1337]

I thought it was suppose to consume less power dafuq?

It will, I didn't mean more power than haswell

[JAKEBAB]

It will, I didn't mean more power than haswell

ok now im really confused so what is it going to use more power than?

[steffen_anywhere]

To clarify:

The overall power requirement i.e. Heat output for the same processing is lower. This means more efficient.

However, the thermal density will increase and it will be just as hard to cool really (actually depends on several factors and cannot be said as easy as that).

[D3MON]

I think heat might go down, but thermal density goes up (ergo hotter) but who knows what sorcery Intel does. 

 

You're forcing less power into it, but at the same time into a smaller space. 

Consider that if the heat from the flame of a candle were evenly spread across a 3m square area, it might be tolerable. If the heat from the flame of a single candle were evenly spread across a 3mm square area, it would probably hurt. 

 

disclaimer: this is uneducated babbling 

this^ pretty much hit the nail on the head.

if you halved the die size but kept the same number of transistors it should consume about the same amount of power and produce the same amount of heat except due to the smaller nodes less voltage is required so power consumption drops however theres less area to pass the heat to a heat spreader so temps often go up and you end up with 80*c haswells and likely 85*c broadwells, things like piledriver 8 cores often run at a much cooler 65*c but they have a relatively large surface so thier thermal output it higher (not nessesarily a bad thing as far as longevity goes).

[airdeano]

a majority of the heat increase (excluding the TIM fiasco) was the

FIVR (Full Integrated Voltage Regulator) included on the Haswell base. note

the refresh and devils canyon, the possibility of putting the VR back onto the

motherboard should bring back normalized thermal processing.

[Tea1337]

ok now im really confused so what is it going to use more power than?

OKay, what I mean is in order to shove something into something small, you're going to add and use more power to do so. Relative to haswell, broadwell will use lets say 1/4th the power in the 5770k chip, than in the 4790k chip. But because of the smaller die it's going to use a but more power than that original 1/4th, so it may need to use another 1/8th of that power to get the same performance, while still using less power than the haswell chip.

[JAKEBAB]

OKay, what I mean is in order to shove something into something small, you're going to add and use more power to do so. Relative to haswell, broadwell will use lets say 1/4th the power in the 5770k chip, than in the 4790k chip. But because of the smaller die it's going to use a but more power than that original 1/4th, so it may need to use another 1/8th of that power to get the same performance, while still using less power than the haswell chip.

oh ok gotcha now 

[harrynowl]

Temps technically go down but smaller die size = less surface area to transfer heat.

 

FX chips run hotter than Haswell, but they can be kept under 60c with basic coolers like the 212, you'll struggle to do that with Haswell because it transfers heat from the cores so poorly.

 

Proof is a Haswell running at 80c won't really warm up your watercooling radiator, but FX8350 running at around 60c will cause it to belch fire.

[Socratic]

48pm x 10^(-3) x 14nm = 672 oxygen atoms wide?

[LukaP]

The heat goes down, as well as power consumption, but that is in case you have the same amount of transistors. (that is due to less voltage needed, and less leakage with FinFET)

 

But what does happen is that the heat density goes up, and since a die shrink means more transistors per die, you get more heat as a result.

[Socratic]

The heat goes down,

This mean heat dissipation goes down?

 

I don't understand die architecture well enough to begin to make a guess on the heat. The wires in the silicon are smaller, will there be more of them or are they just shrinking the size and restructuring the engineering? If they increase the amount of wires would they have to have more transistors, and would this make for worse overclocking because you might be more likely to hit transmission delay? Or can they possibly engineer to improve delays?

[LukaP]

This mean heat dissipation goes down?

I don't understand die architecture well enough to begin to make a guess on the heat. The wires in the silicon are smaller, will there be more of them or are they just shrinking the size and restructuring the engineering? If they increase the amount of wires would they have to have more transistors, and would this make for worse overclocking because you might be more likely to hit transmission delay? Or can they possibly engineer to improve delays?

There are very little wires in the chip itself, the transistors are made of silicon. And yeah what goes down is the heat produced when one transistor switched its state. This heat is very low, but you have billions of transistors so it adds up, and since they are smaller there is more heat per unit of area (heat density)

[Socratic]

There are very little wires in the chip itself, the transistors are made of silicon. And yeah what goes down is the heat produced when one transistor switched its state. This heat is very low, but you have billions of transistors so it adds up, and since they are smaller there is more heat per unit of area (heat density)

Isn't there wires connecting the transistors? 

[juretrn]

There are very little wires in the chip itself

Not really, there are millions of copper interconnects between transistors.

[LukaP]

Isn't there wires connecting the transistors? 

 

 

Not really, there are millions of copper interconnects between transistors.

yeah basically really short wires, that have almost no resistance  (and hence heat) due to being much shorter than they are wide, and so the formula for resistance gives us a looow number

 

and since l is much smaller than A, and p for copper is low too, we get a number low enough to discard it

[Socratic]

yeah basically really short wires, that have almost no resistance  (and hence heat) due to being much shorter than they are wide, and so the formula for resistance gives us a looow number

 

and since l is much smaller than A, and p for copper is low too, we get a number low enough to discard it

Well I think I've learnt a few things from you. You've opened my eyes a bit. Thanks.

[juretrn]

yeah basically really short wires, that have almost no resistance  (and hence heat) due to being much shorter than they are wide, and so the formula for resistance gives us a looow number

 

and since l is much smaller than A, and p for copper is low too, we get a number low enough to discard it

Yes and no. The length actually adds up to a lot.And if copper wasn't a problem, then there would be no research going into using exotic materials (nanotubes) for interconnects... but it is.

Spent enough time on FE UNI to know a thing or two about electronics (shameless brag  )

[LukaP]

Well I think I've learnt a few things from you. You've opened my eyes a bit. Thanks.

Glad to have helped

[LukaP]

Yes and no. The length actually adds up to a lot.And if copper wasn't a problem, then there would be no research going into using exotic materials (nanotubes) for interconnects... but it is.

Spent enough time on FE UNI to know a thing or two about electronics (shameless brag )

Yeah id does add up to a noticable number, and they are researching new tech due to that, but compared to transistors, its still relatively low. The ratio is bigger with FinFETs since they dump less heat tho.

And yeah im at FMF Physics so i got and idea about this too (shameless bragg) just that as a physicists i neglect results that are (comparably) unnoticable haha