Why Intel CPU's run at 95°C and why AMD's should, also

[Grog6]

I love you, in the sense of physics.

The whole thermal problem is realizing the problem has not been solved to an exact number by anyone.

Everyone has tried; all the names have taken their hack at it.

To quote a movie I like, "It remains"

"Easy" is not a word that ever comes out in physics, if you are sharp.

Let the dweebs say easy.

The only real solution you will ever achieve, even with massive instrumentation, is an approximation.

If you ever reduce it to an equation, I will worship you.

I will not discourage you; do this, and get a Nobel Prize.

 

[Deleted member 185158]

I also encourage you to run your chips 90c!!



Temp gradient.
Can passively cooler here.

Pretty straight forward numbers.

 

[mtcn77]

K, so as it happens we have the necessary data inbound.
The OEM BoydCorporation is also measuring in accordance with the thermal resistance formula. 130W with a vapor chamber vs. a copper core makes for the same gradient difference. If we go by their, same as above, conclusion we should reach their gradient numbers.

Spoiler: BoydCorporation:

The spring-loaded thermocouple in the base minus the ambient temperature divided by the over power resulted in the resistance reported.

This is just the airflow bench. Not the normal conductivity bench.

The Thermabase solution shows a 21% increase in performance over the copper base of the same thickness when the CPU is centered on the heat sink. The gain increases to 27% when the CPU is offset from the center of the heat source by 20 mm (in the direction perpendicular to the airflow).

They seem to be uptick about them having missed experimentally by 10%. I know what it is, they calculated the resistance right from conductance, there is also the convection in some formulas.
My take on it is, they have a weird rounding factor. When you factor, you don't subtract, you reciprocate and when it comes up as 1.279: it is "28%", not 27%.
Gradient numbers are similar to what our original test references outline.
Vapor chamber puts out a 10.4°C temperature gradient next to the copper core. 40.3°C is right on target. What's more we have air convection data.

I still want to test this between liquid metal vs. graphenegraphite to see how high flux densities affect graphenegraphite pads. They certainly be getting some flak, however what they are good at is acting like a flat vapor chamber. Their vertical heat conductance is 28 W/mK, however laterally it is 400 W/mK.

So, there is also the issue of depth affecting conductance.

 

[mtcn77]

The heatsink constitutes a punitive amount in the total convective heat transfer formula. It is just 45% of the total resistance while the plate distribution is 30% and the thermal paste is 10% - almost equal in total.
Slipping in a vapor chamber underneath the heatsink and filling both surfaces with liquid metal has some backing in the electronics cooling association.

 

[eidairaman1]

Overclocking in any way was never guaranteed, it was a bonus, people are not greatful for any bonus they get.

 

[mtcn77]

I think temperature throttle limit, pttl, does not work when pb2-pbo is off, in the manual setting. Some user is reporting higher than 96°C even in a stock 3600.
Granted, there is also another predicament. Vapor chamber shims work worse than standard copper coldplates when the gravity vector is pointing upside down.
It will be interesting to point out that only extreme overclockers use their motherboards horizontally in open benchtops.

 

[EarthDog]

It will be interesting to point out that only extreme overclockers use their motherboards horizontally in open benchtops.

Oh? There are plenty of cases where the Mobo is laying flat.

 

[JackCarver]

Flat mobo will be needed by any extrreme liquid cooling (nitrogen/helium)

 

[mtcn77]

Oh? There are plenty of cases where the Mobo is laying flat.

If it works to increase ventilated temperature, I don't think we can market the accompanied worse temperatures of this idea in restricted environments to htpc crowds. The airflow is a limited resource & exit temps of 45°C vs. 65°C is all the rage.

 

[EarthDog]

Not going down your rabbit hole (doesn't matter what is on top if using ambient cooling when the heat can't get out of the die............. do all the math you want!)... just saying there are plenty of chassis which have the motherboard laying flat.

 

[mtcn77]

I still think the market is going to be huge for advanced contact thermal interface materials in the near future.

 

[JackCarver]

there are plenty of chassis which have the motherboard laying flat

Mostly HTPC cases, wouldn't buy such a case for another use case. It's for the Hifi-Rack living room ambience.

 

[EarthDog]

Mostly HTPC cases, wouldn't buy such a case for another use case. It's for the Hifi-Rack living room ambience.

There are plenty of others as well, Jack......

I still think the market is going to be huge for advanced contact thermal interface materials in the near future.

Sure, if the real issue is resolved... otherwise, this is simply a fun math exercise. Again, it isn't the TIM, nor the IHS, nor the cooler on top. It is the silicon's ability to get the heat out through its density and lack of surface area.

 

[Zach_01]

Again, it isn't the TIM, nor the IHS, nor the cooler on top. It is the silicon's ability to get the heat out through its density and lack of surface area.

Sure its very restrictive, such a small die surface. But creating colder IHS or applying faster heat transfer TIM is improving silicon heat dissipation, but not as much as one would expect.

 

[JackCarver]

Again, it isn't the TIM, nor the IHS, nor the cooler on top. It is the silicon's ability to get the heat out through its density and lack of surface area.

Why is it possible to improve heat dissipation only by delidding a CPU and replacing the Intel TIM by liquid metal? I achieved 15 degrees lower temps. And with my waterblock on top another 5 degrees compared to the Intel TIM with air cooling. So in my opinion you can achieve a lot here with better materials.

 

[EarthDog]

Why is it possible to improve heat dissipation only by delidding a CPU and replacing the Intel TIM by liquid metal? I achieved 15 degrees lower temps. And with my waterblock on top another 5 degrees compared to the Intel TIM with air cooling. So in my opinion you can achieve a lot here with better materials.

Think about it.

Like zach said, were at the point of diminishing returns already. All this change will end up netting little over the already over the top (for non benchmarking individuals) deliding. Unless you go subambient, not much will make a difference because of the inherent constraints of the silicon. Even under subambient, these chips can get hot! Especially the higher core count CPUs.

Intel chips have solder again... delidding and reapplying there yields a couple of C.

 

[JackCarver]

Think about it.

Because the Intel TIM is rubbish, at least till the first coffee lake generation, the new ones are soldered. And I achieved more than I expected. So everyone can think different here, but with Intel CPUs, delidding was the place to go to get stable overclocks.

 

[EarthDog]

Because the Intel TIM is rubbish, at least till the first coffee lake generation, the new ones are soldered. And I achieved more than I expected. So everyone can think different here, but with Intel CPUs, delidding was the place to go to get stable overclocks.

Keyword... WAS. Any of these that are soldered barely show an improvement. Things can help, but there is a point of diminishing returns is my point... and this discussion is at that point. Not to mention one of every 1000 pc owners would even consider any kind of deliding the first place. Even at a so called 'enthusiast' site like tpu, few do it.

Intel also stated this was an intentional design consideration...so there is that. The K series that overclock use it. Not sure if the locked ones do... but they are locked...so that isn't relevant.

Edit: to be clear, when I said it isn't the tim, nor the ihs or cooler on top, this is assuming a cpu that overclocks with solder tim. You can strap on a 480mm cooler versus a 240... the deltas are closer than one may expect compared to yesteryear with larger dies, etc. Certainly, some improvements can be made, but, to what end, Jack? A few C isn't 100 Mhz on an overclock...

 

[JackCarver]

As far as I know, the die of the new coffee lakes is larger and thicker than that of the old ones. The thickness may here also a factor why the new coffee lakes run that hot, der8auer made a video here, where he smoothened the die with sand paper to get better temps here after delidding.

So result? Whatever cooler, heatsink, TIM/ water/air cooling you are using is pointless as the problem is the silicon? Won’t believe that...

 

[infrared]

Respectfully, I still don't grasp the concept of this thread..

If you aren't overclocking almost any cooler will do (meaning any deep thinking on the subject is a waste of time/energy).. Or if you are overclocking cooler is always better with almost no exceptions.. I'm not going to gain anything by allowing my processor to run near it's thermal limit, why tf would you want to do that??

 

[EarthDog]

As far as I know, the die of the new coffee lakes is larger and thicker than that of the old ones. The thickness may here also a factor why the new coffee lakes run that hot, der8auer made a video here, where he smoothened the die with sand paper to get better temps here after delidding.

So result? Whatever cooler, heatsink, TIM/ water/air cooling you are using is pointless as the problem is the silicon? Won’t believe that...

Be careful, you're putting words in my mouth...or maybe I wasnt clear. My context is with Intel CPUs that can overclock/unlocked. Those that have the sTIM. Because, as was mentioned just above, the locked chips will be just fine and run as they were designed to. Same with the K series. The OPs original point (I'm not entirely sure what it is either, honestly) is to run the cpu hotter...is just jenky snake oil to me. The results gained from such seem, to me, not worth the effort. I wouldnt call it pointless... just a large time sink for what amounts to little gains.

The concept of killing the temp spikes as discussed earlier seems limited by the silicon... not the cooler. Its just too intense too quick. And no amount of a few C difference isnt going to prevent the spike from happening.

 

[Zach_01]

Be careful, you're putting words in my mouth...or maybe I wasnt clear. My context is with Intel CPUs that can overclock/unlocked. Those that have the sTIM. Because, as was mentioned just above, the locked chips will be just fine and run as they were designed to. Same with the K series. The OPs original point (I'm not entirely sure what it is either, honestly) is to run the cpu hotter...is just jenky snake oil to me. The results gained from such seem, to me, not worth the effort. I wouldnt call it pointless... just a large time sink for what amounts to little gains.

The "idea" of hotter CPU is that its increasing the temp gradient/delta between silicon and all parts after towards the cooler, whatever that may be (air, water, chiller, LN2) Because of higher delta the heat transfer increases.
But for me (sorry @mtcn77) its completely dumb, because it contradicts the idea of silicon cooling and preserve its integrity along with highest clocks and voltage possible. If you want high temp delta, find a way to further cool the IHS or cooler's cold plate, and not keep CPU temp high...

Conventional cooling hits the ambient temp.
Or use a TEC with extra W consumption.
Or use a chiller which will hit the water icing temp.
Or a special chiller with sub-zero (°C) liquid temps.
Or LN2

There is no other way around it. And Intel does not "run" its CPU's at 95°C on purpose because of high delta. Its because users choose to do so, by wanting more clock, and those CPUs can usually operate safely at these temp levels. 7nm will change this mindset.

2 videos should all watch closely...

 

[mtcn77]

Respectfully, I still don't grasp the concept of this thread..

I understand that. That is what is striking about it. Heat transfer coefficient is limiting a 250w cpu cooler, such as BeQuiet! Dark Pro 4, from operating up to its thermal specification.
We have 2 alternatives to make up for lost heat transfer coefficient;

1. at thermal transmittance level, higher thermal gradient makes up for lower thermal resistance,
2. at heat transfer coefficient level, higher thermal conductivity of better than copper transfer media make up for lower thermal resistance.

You have missed a sizable portion of the OP, but GN walks you through the explanation perfectly - AMD's TDP calculation is analogous to thermal series resistance formula. You get thermal resistance lower and get instantaneously better performance. This is how this overclocking algorithm operates.

Spoiler: AMD Community post, verbatim:

After applying metal liquid between the cooler and IHS things improved by almost 20°C in the range of 65-80°, that was when the CPU draws from 80 to 120watts, (80° became 60 at 120watts!) but improved only by 3°C when the wattage jumps over 150watts, (90° instead of 93-94°), also frequencies improved accordingly.

 

[infrared]

I understand that. That is what is striking about it. Heat transfer coefficient is limiting a 250w cpu cooler, such as BeQuiet! Dark Pro 4, from operating up to its thermal specification.
We have 2 alternatives to make up for lost heat transfer coefficient..

In a hypothetical universe where the best coolers we had were barely enough, then yes, in that scenario running the silicone as hot as it could endure would allow the highest amount of heat to be transferred. However back in the real world there's plenty of extremely good coolers so we have the luxury of being able to cool to lower temperatures which results in higher clock speeds, slighly less power consumed and longer life span.

 

[John Naylor]

You said copper practically plays no part, I say the copper mass plays a part. Most tower coolers lack it.

Thermal mass plays a part but not with regard to overall heat removal rates. PC components deal with varying loads. With small mass, the cooling system will immediately respond to increases in thermal load often leading to the fans 'chasing their tail .... load rises.... heat rises ... fans spin up ... component cools ... fans sow down. With larger coolanyt capacities, or larger tower mass, this effect is mitgated against. Typically, most cooling systems are limited by the surface area of the block. Heat conductivity is the major factor here and that's why copper excels in this application with almost twice the thermal conductivity coefficient. However .... in the tower, the process of heat transfer is called convection and to a lesser extent, radaiation

From engineering toolbox

"Conduction as heat transfer takes place if there is a temperature gradient in a solid or stationary fluid medium. "
"Heat energy transferred between a surface and a moving fluid with different temperatures - is known as convection. "
"Heat transfer through radiation takes place in form of electromagnetic waves mainly in the infrared region. " ... emissivity coefficient

When transferring heat to a fluid (liquid or gaseous) .... the thermal conductivity of the material is not in play here, but the transfer of the surface to the fluid is what determines heat transfer. With the same fluid, the material properties are not even in the equation