Doesn't lower temperature increase resistance?

[Haswellx86]

At first I didn't used to get the idea of sub zero cooling for extreme overclocking. We are way out of range of thermal throttling, then why cool at such low temperatures?

 

Then somehow I picked up that the cooler the chip is, the lower the resistance is, which results in lower voltage requirement. I believed in this theory for a while.

 

But then I just learnt that the cooler the temperature is, the more the resistance goes up in semiconductors. More heat gives the electrons more energy to jump from the valence band to the conduction band. So if lower temperature increases resistance, then why do we do sub zero cooling for extreme overclocking?

 

Why even cool the chip in the first place, until the threat of melting? AI says that it is because of increased thermal noise and instability, which doesn't seem like a clear answer.

 

If resistance does increase in lower temperatures, then more voltage is required, hence more power. But I have seen that sub zero cooling brings the power usage quite low. How does all this connect?

[monopoly_hater]

16 minutes ago, Haswellx86 said:

Then somehow I picked up that the cooler the chip is, the lower the resistance is, which results in lower voltage requirement.

This is completely incorrect.

 

17 minutes ago, Haswellx86 said:

But then I just learnt that the cooler the temperature is, the more the resistance goes up in semiconductors. More heat gives the electrons more energy to jump from the valence band to the conduction band.

Yes, this is correct. It's known as a NTC (Negative temperature coefficient).

 

 

20 minutes ago, Haswellx86 said:

So if lower temperature increases resistance, then why do we do sub zero cooling for extreme overclocking?

The thermal noise AI mentioned is correct. It's random fluctuations in voltage that makes the chip more unstable. Now that there are less fluctuations, you can run your chip at lower voltage for given frequency. One more thing I came across in my quick search is that apparently it also leads to less collisions. That balances out the higher resistivity. 

 

24 minutes ago, Haswellx86 said:

If resistance does increase in lower temperatures, then more voltage is required, hence more power. But I have seen that sub zero cooling brings the power usage quite low. How does all this connect?

As I said, overall reducing the temps lead to lower voltage. Which OF COURSE leads to lower power consumption.

[porina]

I'm outside of my comfort zone in what I'm about to say so anyone who knows better feel free to correct, but for high speed digital circuits it isn't exactly a resistive problem more capacitive. Not to say resistance isn't a factor, but it is more complicated than that.

[Haswellx86]

36 minutes ago, anirudthelinuxwIzard said:

Now that there are less fluctuations, you can run your chip at lower voltage for given frequency.

But doesn't more resistance mean more voltage is required? And how does having less thermal noise relate to being able to run at a lower voltage?

[podkall]

1 hour ago, Haswellx86 said:

Why even cool the chip in the first place, until the threat of melting? AI says that it is because of increased thermal noise and instability, which doesn't seem like a clear answer.

Just because sensor picks up X temp, doesn't mean the temp is X and not Y.

 

If a CPU runs very fast, it gets hotter faster, that's pretty normal right.

 

But how do you know how hot it gets reliably? Sensors in chip? Thermal camera?

 

Sensors don't have to know exactly the CPU is hot. Or maybe they do, but not all of the CPU.

 

The hottest parts of chip become hotter, the usually colder parts of chip suddenly are warmer, there may be regions on chip that aren't accounted for thermally because under normal circumstances they don't get warm enough to cause issues/failure.

 

But extreme speeds is something else.

 

Thermal camera doesn't work, because you don't read immediate temps of the CPU, you read the temps of the surface, which first needs to get transferred through metal/heatsinks.

 

This transfer isn't instant, if a CPU is 95C on surface, it could already be 105C on certain areas.

 

 

A very simplified reasoning of why people use such low temps like sub zero cooling, think about oldschool light bulb.

 

You break the glass, and the wire melts, put the wire into an atmosphere, and it doesn't melt but it still glows.

 

Put the wire from litghtbulb into liquid nitrogen, it doesn't melt either there.

 

 

And as for resistance, things like material and resistors do increase in resistance as they heat up, so why should transistors and other parts of CPU an exception? Did they get some kind of thermodynamic/physics pass?

 

[Haswellx86]

6 hours ago, podkall said:

 

If a CPU runs very fast, it gets hotter faster, that's pretty normal right.

Yeah but if the sensors say -230 C, then how far would the hotspot temperature be? Not by a lot.

 

Oh, is it because the higher the temperature difference is, the faster it can normalize? So does that mean that the chip would actually be getting so hot so fast, that a higher temperature difference (more cooling) would take out heat faster? Is it really that critical that a temperature close to absolute zero is even beneficial? I mean a not-so-extreme cool temperature should still be able to remove heat so fast that it doesn't thermal throttle, right?

 

Or I guess where I am wrong is that thermal throttling (~100 C) is not the thing that we need to worry about, but rather the aim of keeping the chip as cool as possible, because I guess that's what we are doing.

 

But then if colder temperatures increase resistance, then why cool it? Because of thermal noise and voltage fluctuation? And then how does lower temperature result in lower power, when actually more voltage is required?

 

6 hours ago, podkall said:

And as for resistance, things like material and resistors do increase in resistance as they heat up, so why should transistors and other parts of CPU an exception? Did they get some kind of thermodynamic/physics pass?

Idk man, you tell me. "resistance generally increases with temperature" and "on increasing temperature, the resistance of semiconductors decreases" are two very contradictory statements.

[Likwid]

There is correlation between power leakage, higher temps, higher leakage and power leakage increses again in endless loop if you can't cool chip and there are no power limit (same as 7800x3d chip failure).

https://arxiv.org/pdf/1809.03147

 

[podkall]

2 hours ago, Haswellx86 said:

Oh, is it because the higher the temperature difference is, the faster it can normalize? So does that mean that the chip would actually be getting so hot so fast, that a higher temperature difference (more cooling) would take out heat faster? Is it really that critical that a temperature close to absolute zero is even beneficial? I mean a not-so-extreme cool temperature should still be able to remove heat so fast that it doesn't thermal throttle, right?

If not-so-extreme temps were enough, would you think they'd stop doing extreme low temps?

 

3 hours ago, Haswellx86 said:

But then if colder temperatures increase resistance, then why cool it? Because of thermal noise and voltage fluctuation? And then how does lower temperature result in lower power, when actually more voltage is required?

You seem to be confusing increase with decrease, low temps decrease resistance:

 

 

 

3 hours ago, Haswellx86 said:

Idk man, you tell me. "resistance generally increases with temperature" and "on increasing temperature, the resistance of semiconductors decreases" are two very contradictory statements.

I asked ChatGPT, here's what he had to say about this:

 

 

 

Conductivity isn't everything.

 

[Haswellx86]

On 10/29/2024 at 2:05 AM, podkall said:

You seem to be confusing increase with decrease, low temps decrease resistance:

Search "do cold temperatures increase resistance" and "do cold temperatures increase resistance in semiconductors?". The answers you will get are very contradictory. I need explanation.

 

On 10/29/2024 at 2:05 AM, podkall said:

I asked ChatGPT, here's what he had to say about this:

No clear answer. Maybe thermal stability is valid, and about leakage current, isn't it supposed to be really low in effect? And does that even contribute in instability?

[podkall]

1 hour ago, Haswellx86 said:

Search "do cold temperatures increase resistance" and "do cold temperatures increase resistance in semiconductors?". The answers you will get are very contradictory. I need explanation.

 

No clear answer. Maybe thermal stability is valid, and about leakage current, isn't it supposed to be really low in effect? And does that even contribute in instability?

What do you mean "no clear answer"? Have you even read my reply?

 

And no, the answers are not contradictory, semiconductors increase and resistors and certain metals decrease in resistance when they heat up.

 

CPU is not just conductors, CPU is all kinds of stuff, CPU holds.

 

Heat can damage CPU or cause it to throttle, those are 2 reasons why CPU needs to be cooled during extreme overclocking.

 

Wrong microcode that put too much voltage into Intel CPUs was slowly cooking them -> they used more voltage -> they got cooked slowly overtime -> what do you do when overclocking? you increase voltage, especially during extreme overclocking.

 

 

 

And you can't just let it rip during extreme overclocking, the CPU will throttle, the CPU will wear significantly, the CPU may completely shutdown before it can get benchmarked, etc.

 

CPU is a delicate contraption, it's not comparable to something like car's engine 1:1, it's way more sophisticated, and very small.

 

You are dealing with super small size of many things, and pump 200-400-600W+ of power into it, during extreme OC. (especially with Intel those things turbo at 200W+ TDP out of factory)

 

 

 

You want real life comparison?

 

Go run marathon in summer heat.

 

Now run faster! Now run faster while not drinking more water!

 

Run way faster, but the water you get is now 2° warmer, sorry!

 

That's what extreme OC would be without proper cooling.