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

[Grog6]

You've got me wondering if using a slab of water as TIM on a LN2 rig would be more effective...

Hard to test, and messy if it fails, but Hmmmmm.

 

[Deleted member 185158]

You've got me wondering if using a slab of water as TIM on a LN2 rig would be more effective...

Hard to test, and messy if it fails, but Hmmmmm.

I've run LN2 without paste. A single drop of water. The cpu pot freezes to the cpu instantly. Best effect is with lapped surfaces.

Like EarthDog mentioned, with LN2 you want big copper mass and lots of holes (I guess) for surface area.

The movement of heat is the opposite direction, your heating towards the motherboard vs away from it.

 

[R-T-B]

Do the equation of thermal transfer when you think about using Liquid metal; there are always disadvantages; a thick layer of nickel does not help with overclocking.
It's just keeping your thermal compound from eating your heatsink.

Is it really that much of a factor? I ask not because I use liquid metal (other than under the CPU heatspreader once long ago, where Nickel is present no matter), but because my present Noctua heatsink is Nickel coated. Pretty thin layer, but present all the same.

Keep in mind, I have a little thermal understanding, but you guys are way over my head here.

 

[Grog6]

You know, I'm actually wrong about pure electroplated nickel; stainless is much much worse than Nickel.

Nickel is about 100W/mK, and stainless is about 20.

The Gallium/Indium liquid metals are in the 20-40 range, so the Nickel layer isn't that bad, other than the roughness, which would have been easier for me to lap smooth than remove completely.

Another discovery today is that the solder can be better than the Liquid metal, depending on lead and indium content.

 

[Zach_01]

You know, I'm actually wrong about pure electroplated nickel; stainless is much much worse than Nickel.

Nickel is about 100W/mK, and stainless is about 20.

The Gallium/Indium liquid metals are in the 20-40 range, so the Nickel layer isn't that bad, other than the roughness, which would have been easier for me to lap smooth than remove completely.

Another discovery today is that the solder can be better than the Liquid metal, depending on lead and indium content.

Conductonaut is stated as 73W/mk. Is that a lie or do they mean something else than you?

 

[Grog6]

73 would mean there's a lot of indium in it; gallium is 20, so that's a lot of indium.

Indium is about 140+, so it's probably not as liquid as others.

 

[Zach_01]

73 would mean there's a lot of indium in it; gallium is 20, so that's a lot of indium.

Indium is about 140+, so it's probably not as liquid as others.

Indeed, it is stated on package... "Increased Indium content"
It's great stuff IMO, even in un-delidded applications. I'm using it for about 25days now with H110i 280mm AIO.
Needs some caution tho on the apply process, but its not rocket science...

As for liquidity look the video on 1:24...

 

[Wickedt]

The cpu die area is split into pieces on Ryzen 3000 chips. Zen+ and previous only had a single die.
Which die area is Noctua referencing? the I/O chip?

I liked the point you brought up earlier about Tim 10 Wm2K vs LM 100 Wm2K thermal transmittance.

You'd think the solder I removed and replaced with TIM would had made a HUGE difference. Nah, not like you think.

I'm running a De-lidded 2700x, solder replaced TIM with the stock cooler and it changed the effects of the processor none.

It's not the IHS plate, or the TIM / LM..... It's the density of the transistor count on small die. This is where the "temp spikes" come from.

It's the density of the transistor count on small die. <--This

As the die sizes get progressively smaller, with a higher concentration of transistors, your getting much less cooling surface. I think this is one of the reasons my nzxt x62 is doing so well, its the placement of there water channels at the exact best spot for maximum cooling of the cpu dies.

 

[Deleted member 185158]

It's the density of the transistor count on small die. <--This

As the die sizes get progressively smaller, with a higher concentration of transistors, your getting much less cooling surface. I think this is one of the reasons my nzxt x62 is doing so well, its the placement of there water channels at the exact best spot for maximum cooling of the cpu dies.

Right. Makes it damn near impossible to run a high temp gradient. 1.45v and idles 37c. lol.

 

[Wickedt]

Id love to see a user replaceable cover for the Ryzen's, pure copper, that envelops all the dies, including the sides. We stopped using pennies in Canada, i know there's tons of them out there waiting for this. Penny for your thoughts? lol

 

[mtcn77]

Right. Makes it damn near impossible to run a high temp gradient. 1.45v and idles 37c. lol.

The cpu heatsink material having an aluminum base with nickel coating on top could play a part. We can do this, although I have to decipher the 3d model conductance formula. At first, I was trying to tie things with thermal conductance, instead it looks more feasible with thermal transmittance. I still don't get what the denominator means in the alternative thermal conductance formula: (cal/sec)(cm2C/cm)
Copper looks suspiciously like its unit measure with 0.99.

 

[Deleted member 185158]

The cpu heatsink material having an aluminum base with nickel coating on top could play a part. We can do this, although I have to decipher the 3d model conductance formula. At first, I was trying to tie things with thermal conductance, instead it looks more feasible with thermal transmittance. I still don't get what the denominator means in the alternative thermal conductance formula: (cal/sec)(cm2C/cm)
Copper looks suspiciously like its unit measure with 0.99.

Well I'm trying a few side experiments. Using the stock cooler. However the fan is upgraded. I set manually 2700x to 3.5ghz and running 1.155v with a 20% fan curve to 100% at 75c. The bios will not allow me to change the max RPM any lower than 100% past 75c.

This effort is going to be difficult to control with high clock speeds and voltage. I'm already at 72c a few minutes in OCCT Linpack AVX2 16 threads. The fan is about to be loud. lol.

 

[mtcn77]

The bios will not allow me to change the max RPM any lower than 100% past 75c.

This is easier than thought. In the bios there is a smart q-fan option. Once you limit it to 70/60°C & 60/30% you have just reestablished a new 100% threshold in windows.

I don the opinion that somehow lining a good conductor next to a bad one from the start should increase the temperature and serve as a one way valve to increase the heat flow in a specific direction, if it wasn't copper all throughout on the motherboard side. If it makes a temperature gradient - let's face it aluminum will always run hotter to meet the same transmittance as copper - does it keep the motherboard cool, as it does because higher conductance mean copper will return to gradient faster, or spread farther than aluminum? I'm shot.

 

[Deleted member 185158]

This is easier than thought. In the bios there is a smart q-fan option. Once you limit it to 70/60°C & 60/30% you have just reestablished a new 100% threshold in windows.

I don't use software tuning.
One app for monitoring, OCCT does the stress and the monitoring. Less conflicts accurate readings.

Prefer to follow the guidlines according to what I read in the white papers. I aim for cold temps. 75c is hot enough, I alarm at 80c. Anytime I see these high numbers is generally only benchmarking and stress testing.

Not sure how you are going to accomplish the high idle temperature to achieve high temp gradient thresholds.

So a thermal paste that is less conductive when at a cooler temperature and then becomes more conductive at a higher temperature would be one approach. Water is an example of this. The thermal conductivity gets better as it warms up. (This is an example, perhaps a paste designed around these types of properties?)

Water - Thermal Conductivity vs. Temperature

Figures and tables showing thermal conductivity of water (liquid and gas phase) with varying temperature and pressure, SI and Imperial units.

www.engineeringtoolbox.com

 

[mtcn77]

Not sure how you are going to accomplish the high idle temperature to achieve high temp gradient thresholds.

The software won't allow for it but it is simply by zerofan setting taking off only when the heatsink thermal load is max and the cpu is about to thermal throttle if not.

Water is an example of this. The thermal conductivity gets better as it warms up. (This is an example, perhaps a paste designed around these types of properties?)

Water - Thermal Conductivity vs. Temperature

Figures and tables showing thermal conductivity of water (liquid and gas phase) with varying temperature and pressure, SI and Imperial units.

www.engineeringtoolbox.com

Water is a good example: high specific heat, low conductance. If we line it next to aluminum, aluminum will always work at a deficit to water.

 

[Deleted member 185158]

The software won't allow for it but it is simply by zerofan setting taking off only when the heatsink thermal load is max and the cpu is about to thermal throttle if not.

Well, I was able to passively cool, but that's around sub 3ghz (at least with my Zen+) about 1 volt. barely maintain under 90c. adjusted VRM fan for assistance.

I ran an AM2 Sempron (one core) once completely heatsinkless (it did have the IHS plate however) and a small 80mm fan just a few inches away. 1.2ghz 0.75v Sub 50c temp loaded.

Water is a good example: high specific heat, low conductance. If we line it next to aluminum, aluminum will always work at a deficit to water.

Most passive heatsinks are aluminum and very large.

Please note this is not a high wattage passive heatsink. Only handles 47w. Pretty much the max output anyone could passively cool. This is the figure I aim for using HWinfo64.

Once I get closer to 60w usage, gotta have a fan.

ARCTIC COOLING ACALP00022A Alpine AM4 Passive, Silent AMD CPU Cooler - Newegg.com

Buy ARCTIC COOLING ACALP00022A Alpine AM4 Passive, Silent AMD CPU Cooler with fast shipping and top-rated customer service. Once you know, you Newegg!

www.newegg.com

 

[mtcn77]

We could just replace the aluminum ihs with a vapour chamber and call it a day.

In a coarse calculation or CFD simulation, it is not uncommon that a uniform and isotropic thermal conductivity, say 10000 W/m-K which is 25 times the thermal conductivity of copper, is assigned to the entire volume of the vapor chamber.

We need some sort of directionality to force heat into a one way shunt. Placing it one with a vapour chamber would bring both the temperature gradient and necessary thermal transmittance crossection in short order. If it expands 25 times, it is like making a 7nm die's features 5 times wider, like 22nm.
Basically place an aluminum heatsink and it won't perform a lot different because apart from density, the chips don't actually use a lot of power.

74mm² consuming 120w with liquid metal applied - that is 160w/cm². Remember, 300w is the vapor chamber. That is 25x better than copper.
Now, in the chart at 95°C. That is 170w/0.74cm²= 230w/cm²! I think we are getting somewhere.

Design Guidelines for a Vapor Chamber Heatsink
The following table shows the suggested operation conditions for typical applications. They are not necessarily the maximum capabilities of vapor chambers.

Vapor Chamber Ambient temperature	0 - 85 ºC
Power	20 - 300 W
Heat Flux	Up to 300 W/cm2

Since the cpu is an active heat source the same gradient trick doesn't work here and we are necessitated to work from heat flux. Otherwise it would incur a nonstop temperature gradient between the cpu and the ihs unless it throttles.

Anybody with experience in vapour chamber cpu heatsinks at the cold plate? ID-Cooling involves us with their 2015 creation, however it is 130w tdp specified. It is neat, though. I like neat.

ID-Cooling Releases Vapor Chamber CPU Cooler for Mini-ITX Systems

ID-COOLING, a cooling solution provider focusing on thermal dissipation and fan technology research and production for over 10 years, releases a newly developed Vapor Chamber CPU Cooler for Mini-ITX system: IS-VC45. When small PC systems get more and more power hungry and house much more...

www.techpowerup.com

This article describes the temperature gradient as "conduction loss".

Vapor Chamber Heat Sink Design Guidelines - More Advanced

This article is intended as a follow-on to a couple of posts we made in the past about using vapor chambers rather than heat pipes. If you haven’t read those and are not familiar with designing with vapor chambers, we suggest you reference this link before continuing.

www.linkedin.com

I'm interested in using a vapour chamber for ihs, or dynatron r25 for that matter. Apparently, the vapour chamber bulges outward, if specified heat output is surpassed, so they structurally strengthen it using an outer exoskeleton in the form of skived fin layer of solid copper.

 

[Deleted member 185158]

Well the vapor chamber isnt a bad idea. 130w should get you to throttle at max load depending on what you use to build heat. Obviously WPrime 1024m wont build as much heat as Prime95 or OCCT.

Not a bad idea.

 

[mtcn77]

I had forgotten about amd's own stock cooler, the wraith 'spire' having a copper vapor chamber base. It would be a simple understudy to test with some liquid metal paste.

 

[EarthDog]

It really doesn't matter what goes on top of these CPUs.... the problem is getting the heat out of the dense die and small space. It can only put out so much heat no matter what is on top.

 

[mtcn77]

Forcing the issue is probably not a good idea. I'm less than thrilled about heat reflected onto the motherboard vrms. It's not nice when underreported temps kill a device suddenly. Why let it happen, given a choice.
I concede the notion on previously stated motherboard vrm safety. Still, if zerofan will enable temps to go high, vapor chambers have instructions manuals that go as far as 85°C.
I did bring down the shutters on one so infamous cnps9500 when I tried to protect it from rust by *oven-dessication*. Let's just leave a word out to the wise - don't try to go overboard with boiling temp.

 

[Deleted member 185158]

Let's just leave a word out to the wise - don't try to go overboard with boiling temp.

And why I was mentioning to run passive, get down about 50-60w and work your way up from a safe "er" place.

Who said you had to push air onto the heat sink? Flip the fan over homie.

 

[mtcn77]

Who said you had to push air onto the heat sink? Flip the fan over homie.

If you unplug the fan, I think the motherboard gives you the graceful exit.

This is a stupid idea. We are trying to push 250w just to justify aftermarket coolers. I am still not totally satisfied by the overall absence of vapor chambers in cpu coolers. Just attach to lower tier models for further reduction in thermal resistance opening up a brand new mid tier.

 

[Grog6]

Doing the thermal calculation is hard; but thermodynamics is hard.

If you want to be totally accurate with your calculation, you have to instrument each layer of material in your stack.

So: CPU die, liquid metal, copper shim, liquid metal, layer of nickel on HS, HS thermal spreader, fins or water (or LN2) temperature, ambient air.

Every layer comes into the equation.

If you do w/m/ degree K on each material, you can make it all come out reasonably easy.
Thickness in meters is easy, degrees k is easy; Watts is not because of the varying area the thermal flux is passing thru is not constant, and it's not evenly spread across the entire area of a layer either.

Measuring it all at various loads with thermocouples and a NI card, and dumping it into Excel is pretty easy.

If you treat the stackup of thermal resistances as a series electrical circuit with the thermal difference of the CPU die to ambient as the overall Thermal driver, you can see where in the stack the highest thermal gradient is, and that's your limiting material.

Get rid of it and try again, until you can't get rid of anything else.

If the biggest gradient is the cooling fin or water to ambient, you need more fin, more radiator, or more cowbell.

I have to say; the only real reason we do this whole execrcise is the more cowbell thing; I consider myself ripping off intel or AMD when I manage a real OC over the "advertised" clocks.

Fuck those assholes, that want to charge us Serious precious dollars for meager incremental gains.

I'll wait a few years, and rock my x58 system with a previously $3k processor, you bastards.

Now I'm rockng my x97 system the same way.

 

[mtcn77]

If you do w/m/ degree K on each material, you can make it all come out reasonably easy.

You know what came out when I did it? It turns out you are super cool.

If you treat the stackup of thermal resistances as a series electrical circuit with the thermal difference of the CPU die to ambient as the overall Thermal driver, you can see where in the stack the highest thermal gradient is, and that's your limiting material.

You are a good aid and an inspiration. Now, I got to think what it makes sense for this formula;

What is Thermal Insulation - Thermal Insulator - Definition

Thermal insulation is the process of reduction of heat transfer between objects in thermal contact or in range of radiative influence. Thermal Insulation

www.thermal-engineering.org

The blank in the statement is how much heat flux expands at each in the series. I need to calculate heat flux density from transmittance, get a gradient toward the next in the chain and reach a total outlier between that and the core temperature.
I had the simple notion of conductance*gradient gives the magnitude of heat flux, but there is also the area in the proposition.