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

[Zach_01]

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.

I think that 74mm² is the 8core chiplet.
I can agree for the density of those small dies, but having high thermal resistance (TIM over LM) would only make things worst.
Not having any difference from delidded 2700X I think its because of soldered IHS on the die, over other CPUs that only having "regular" TIM under the hood by stock (and not soldered).

Not all the cpu heat is dissipated to the heat sink. A great deal of it yes, but some is dissipated through the board as well.

Yes yes... Thats one reason (among others) there is a difference between TDP and max power draw.

 

[Deleted member 185158]

I think that 74mm² is the 8core chiplet.
I can agree for the density of those small dies, but having high thermal resistance (TIM over LM) would only make things worst.
Not having any difference from delidded 2700X I think its because of soldered IHS on the die, over other CPUs that only having "regular" TIM under the hood by stock (and not soldered).


Yes yes... Thats one reason (among others) there is a difference between TDP and max power draw.

response to in bold.....
I'm running LIDLESS PGA 2700x - means no solder. I physically removed it, replaced it with TIM.

Thermal Design point is for your max P-state.
Do note a cpu-z txt pull will give you absolutely NO boosting specs at all. (This differs from past chips such as the FX line, the Boost states are listed as P-states)

You can convert electrical wattage to BTU.
The reason they use say 105w TDP, is because this number is much lower than the BTU per hour you need to remove, 358 BTU/hr. (stock max p-state)

P(BTU/hr) = 3.412141633 × P(W) // 1W = 3.412141633 BTU/hr (the P = power)

 

[mtcn77]

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?

Their reference is to the zen 2 dies, 7nm, not the matisse 14nm IF port.

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.

That note was really illuminating, me included. Copper practically plays no part, since it far outpaced the rest and is playing the least significant part in the thermal resistance series calculation, therefore, ironically.

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.

I'm not 100% clear if it will resolve temperature spikes, however if a graphene led meets steady state at 2500°C, you get to wonder if thermal transmittance is making a fool out of 250w tower coolers for never meeting their crossection up to a temperature higher than pb2 ever permits. Now that would be an industry insider joke.

 

[Deleted member 185158]

Their reference is to the zen 2 dies, 7nm, not the matisse 14nm IF port.
That note was really illuminating, me included. Copper practically plays no part, since it far outpaced the rest and is playing the least significant part in the thermal resistance series calculation, therefore, ironically.


I'm not 100% clear if it will resolve temperature spikes, however if a graphene led meets steady state at 2500°C, you get to wonder if thermal transmittance is making a fool out of 250w tower coolers for never meeting their crossection up to a temperature higher than pb2 ever permits. Now that would be an industry insider joke.

It likely is an industry joke. I cannot recommend a better cooler over the stock because the Cpu is already maxed boosted to it's stability point under throttle 95c. The rest just doesn't matter while sensemi is a temp driven algorithm, while user slight changes only give slight effects.

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

I've thermally tested down to -30. SenseMi boosting did not change without user input.

 

[mtcn77]

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

You are right, I was just looking at its inverse corollary, the thermal resistance difference of tim from liquid metal to fit the equation.
I try to tie things with the OP. In this instance, the original study is where it is mentioned - 4th in the list of things.
15°C delta is what it translates to and not because it pulls any less heat, in fact is using 50% more. Sometimes I wonder how people do these things better than its vendor.

How difficult could it be to get your cpu to steadystate at 90°C and see if it passes a test there at that voltage&MHz... pttl permitting, I think we can benchmark this manually.

Undervolting study@90°C, no less.

 

[Grog6]

This is a ridiculous take on cooling, IMHO.

Yes, Math and thermodynamics state that maximum cooling and heat transfer takes place at the highest temperature gradient, the one at issue is to ambient air, so as high as is theoretically possible is best.

This is not a case for longevity or stability, however.

As low as possible while delivering maximum performance is what drives LN2, chillers, and other extreme cooling solutions; I would postulate that their success obviates the paper at issue in every way.

Every CPU degradation mechanism increases with increasing temperature; at 125C, the diffusion process actually reverses, changing the actual die properties.

Leakage was mentioned; this is the primary driver, in conjunction with capacitance, with heat production in modern dies.

I've done this to a wide range of mosfet and bipolar devices in my career, none were improved by the treatment. Most died spectacularly.
Silicon Carbide is an outlier; it's going to eventually replace Silicon, IMHO, but it will be a decade or more.

Gallium and Nickel are not a great heatsink combo; nickel is a piss-poor conductor. The gain in thermals you get is offset by the layer of nickel you need to protect your shit.
Check out a bar of stainless steel for comparison, it's mostly nickel.
You can put a torch to one end of a 2 foot bar, and hold the other end for a long time; don't try that with copper or aluminum.

Gallium forms a compound with copper, eats the hell out of aluminum, and is a pretty nasty metal to work with in conjunction with any metal in most cases.

If you want to be bleeding edge, go straight to NaK; it's better, and won't eat the heatsink.

 

[mtcn77]

This is a ridiculous take on cooling, IMHO.

Yes, we are in for a crazy ride however hear me out, please.

Gallium forms a compound with copper, eats the hell out of aluminum, and is a pretty nasty metal to work with in conjunction with any metal in most cases.

Last night, got a good readout on it about its electrochemical battery series once applied to graphite(seriously am I the only one who hoped it will work with graphite heatsink fixation?) and its amalgamation when met with copper. It also creates convective currents as heated by such cpu heat sources. What a weird set of features.

If you want to be bleeding edge, go straight to NaK; it's better, and won't eat the heatsink.

'NaK', good joke. You had me considering there for a moment.

If the cooler slope took heat vs. rpm, instead of °C vs. rpm as its two variables we possibly could do it, although case temperatures will incrementally work against our gradient temperature difference once the fan rpm picks up pace. Though, this too can have its steadystate and unit measure in a series.

 

[Zach_01]

response to in bold.....
I'm running LIDLESS PGA 2700x - means no solder. I physically removed it, replaced it with TIM.

Thermal Design point is for your max P-state.
Do note a cpu-z txt pull will give you absolutely NO boosting specs at all. (This differs from past chips such as the FX line, the Boost states are listed as P-states)

You can convert electrical wattage to BTU.
The reason they use say 105w TDP, is because this number is much lower than the BTU per hour you need to remove, 358 BTU/hr. (stock max p-state)

P(BTU/hr) = 3.412141633 × P(W) // 1W = 3.412141633 BTU/hr (the P = power)

I know what you did with the CPU and that is lidless...
What Im saying is that you did not saw any benefit by going lidless because IHS was solderd and not just TIMed...

I know one can convert watt to BTU and vice-versa because its different names of same thing. Power. My primary education and first job(s) was refrigerant and air condition applications.
TDP, Thermal Design Power is heat translated into Watt and its what a CPU is expected to dissipate towards IHS and cooler. For Intel is at base clock and for AMD its for avg boost clocks (above base clock).
This is from recent GN extensive review and research video.

-------------------------------------
To everyone...

Want to see how R5 3600 behaves from just 2~3C change?

Case 1.
PBO auto (PPT:88W, TDC:60A, EDC:90A, PBOscalar:auto)


-2~3C


Case 2.
PBO manual (PPT:95W, TDC:60A, EDC:63A, PBOscalar:X2)


-2~3C

 

[Deleted member 185158]

But this information is useless as these processors have no change in frequency scale from temps be it high or low.

The transistors are at max operating frequecy from the box.

As stated, you're good to max throttle temp.

Thus my argument, high temp gradient makes no more difference in frequency or very minor variations in overclocks.

We are talking about maintaining a 65c idle temp and a max load temp of 90c.

Why would there be a benefit if the idle temp is nominally higher?

Answer.... Is no benefit.

 

[Bones]

Cpus have two thresholds... where it throttles and where it shuts down. So long as the cpu isnt throttling to protect itself that means intel/amd are ok with those temperatures. If you are throttling and hit thermal shutdown, that is too hot. In our overclocking endeavors over the years, we found occasionally some chips can lose stability a bit before that. This is why we usually say, for Intel chips with 100C throttling point (105C+ shutdown) to keep it under 90c. For headroom under throttling point. I'd gladly pound and have pounded chips in this manner.

While i understand the underlying point of your wire analogy, metal fatigue by mechanical movement and temperature with silicon is quite different.

That's true but what I was getting at is if you push something to it's limits for too long it won't end well.
At least we can agree on that.

 

[cdawall]

Doesn’t matter what TIM you use there are not coolers that can handle the specific heat output of current HEDT. The smaller and smaller dies are causing larger issues since the surface area to dump the heat off keeps shrinking. While not a major issue with these little baby 200w chips those of us living with chips pulling 800w are seeing major problems.

 

[Zach_01]

But this information is useless as these processors have no change in frequency scale from temps be it high or low.

The transistors are at max operating frequecy from the box.

As stated, you're good to max throttle temp.

Thus my argument, high temp gradient makes no more difference in frequency or very minor variations in overclocks.

We are talking about maintaining a 65c idle temp and a max load temp of 90c.

Why would there be a benefit if the idle temp is nominally higher?

Answer.... Is no benefit.

I'm not trying to back-up the title of this thead by any means. I'm against running CPU hotter for high temp gradient. This indeed has no real benefit (other than some heat transfer numbers), CPU is running hotter uneccessary, and in ZEN2's case it would hurt performance by a respected amount.
What I demonstrate above with those screenshots is real and mesurable. That was only 2~3C reduction by cooling increase. Imagine going from 90+C down to 60~65C or lower if possible. Depending the 3000 SKU it would boost +150~250MHz, if not more, from that kind of temp reduction.

 

[Deleted member 185158]

I'm not trying to back-up the title of this thead by any means. I'm against running CPU hotter for high temp gradient. This ideed has no real benefit (other than some heat transfer numbers), CPU is running hotter uneccessary, and in ZEN2's case it would hurt performance by a respected amount.
What I demonstrate above with those screenshots is real and mesurable. That was only 2~3C reduction by cooling increase. Imagine going from 90+C down to 60~65C or lower if possible. Depending the 3000 SKU it would boost +150~250MHz, if not more, from that kind of temp reduction.

Do that all the time. Running max pstate 3.7ghz 1.2v. 60c load (ish - depending on ambient) fans are always nice and quiet, very efficient processing figures.

Max clocks 1.250v is what Id seek with Zen 2. The boosting is nice, but not really necessary.

Yes I agree, its about heat transfer performance than it is about processor performance, but we still have to measure that as well. I get it.

Just seems overly thought into.....

If we wanted to transfer heat more quickly we'd use silver. If slower Aluminum.

I believe good testing would come from passively cooled heat sinks, not tower coolers, but who am I to judge. I can confirm these chips can be run with direct TEC cooling under a water block. Just no one has really tried it yet with a proper configuration. (Example ideas)

 

[Zach_01]

I can confirm these chips can be run with direct TEC cooling under a water block. Just no one has really tried it yet with a proper configuration. (Example ideas)

How thick that TEC plate would be? Any directions?

 

[Deleted member 185158]

Doesn’t matter what TIM you use there are not coolers that can handle the specific heat output of current HEDT. The smaller and smaller dies are causing larger issues since the surface area to dump the heat off keeps shrinking. While not a major issue with these little baby 200w chips those of us living with chips pulling 800w are seeing major problems.

And see this is where the IHS plate plays a huge role. Its not thick enough. A percentage of BTU is stored.
The reason is because we only apply cooling to a single side of a 3 dimensional surface.

 

[Grog6]

A CPU, or any semiconductor, (or anything else) really has a finite lifetime.

An equation of state, based on Arrhenius' equation, defines that.

Arrhenius equation (for reliability) | JEDEC

www.jedec.org

These guys write the standards, so they are the Law, in my book, for design.

I break these laws for personal use, but my employers would have serious problems if I incorporated my overclocked CPU into a design for a customer, lol.

Basically shit dies faster if it's hotter. I can't say it easier than that.

Diffusion, which all IC's are based on, is reversible.

The closer you run to ~125C for Silicon, the faster the stuff that makes it N and P type silicon leaves the building; when Elvis has left the building, it's no longer a semiconductor, but a Silicon Resistor.

Silicon resistors have no switching characteristics, so the first transistor in a critical path that loses Elvis, bricks your processor.
These are usually in clock circuits, that may run up to 8 or more times faster than most of the silicon.

In short, love Elvis, keep him cool, and life will be wonderful.

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.

I wasn't joking; NaK isn't worse than Gallium, IMHO. It just burns if you mess up, as opposed to eating your heatsink.
NaK is a viable alternative; you just have to put a bead of silicone caulk around the edge of the CPU, and clean up the excess after you torque it down.
Don't miss any, lol.
As long as it never sees air or water while it's hot, it works great.
Nuclear Reactors have been cooled with it for a half century.

I have cooled Semiconductors with NaK; it's not easy to work with, but it doesn't eat most metals.


EDIT: these are Peltier coolers that can cool around 200W; take a look.
Realize, the peltier devise is ~50% efficient; If you're cooling 200W, it eats over 400W.

comapact thermoelectric air conditioners with up to 250 watts capacity

thermoelectric panel coolers protect electronics, cool and heat enclosures in variety of industries like communications and transit boxes, outdoor control panels, outdoor kiosks, security alarm housing, incubators, gloveboxes

www.thermoelectric.com

I made a LN2 replacement with Peltiers, for a camera; to cool a 10 gram device using 10 W, it took me over 1200W to do it.

 

[Deleted member 185158]

How thick that TEC plate would be? Any directions?

All I can say is the IHS plate replaced with a cold plate twice the size and mass gave amazing results.

I can take a picture of a couple cold plates I have used and rough numbers.

How thick that TEC plate would be? Any directions?

OK Zach, here's an example picture of a couple of cold plates I use.

The big one (top), It fits most motherboards I've come across. I have one that is larger and has only fit a couple boards throughout the last decade+.

The one on the right (bottom) is similar to performance of the IHS plate and is very poor for holding any kind of heat.

I am not interested in setting this cooling up for testing higher vs lower temp gradients again for die hard figures, you'll have to take my word on it.

That IHS plate came from a soldered Ryzen 3 1200.

EDIT: The picture on the right is a lapped Morgan silver dollar.

 

[Grog6]

When you say "Cold Plate" do you mean copper shim, or a plate with coolant running thru it?

That copper plate would really help cooling, if that's all it is.

Take a water block, fill with NaK, put that on top, and cool conventionally, and it would be awesome with no water.
The fine scarf cut fingers would do awesome heat transfer.


Can you tell I have a fixation on potentially explosive cooling materials?

EDIT: What did you attach the Morgan Silver to the HS with?

There's always this:

Aquacomputer cuplex kryos NEXT with VISION AM4, nickel/.925 silver

The cuplex kryos NEXT marks a new milestone in CPU water block development. Every detail has been analyzed, optimized and tested to achieve perfection in cooling performance, installation procedure and product features. The result is not...

www.aquatuning.us

 

[Deleted member 185158]

When you say "Cold Plate" do you mean copper shim, or a plate with coolant running thru it?

EDIT: What did you attach the Morgan Silver to the HS with?

I mean a plate of copper. What's cooling that plate is a matter of choice, I mainly use something that large to transfer heat, but not above 0 heat, more so the aim is sub zero heat, but has the same effect when running warm, this plate would run a higher temperature gradient than the IHS plate because it stored more BTU which then needs less time for transfer or dissipation. You can take more time to transfer heat or dissipate. (I'm not finding the proper wording for the transaction. It's late o'clock)

When the waterblock and Silver plate where lapped to 3000 grit, I could pick the coin off the table without anything except the direct contact. In the picture, thermal paste.

EDIT: these are Peltier coolers that can cool around 200W; take a look.
Realize, the peltier devise is ~50% efficient; If you're cooling 200W, it eats over 400W.

Oh for sure Peltiers are horrible for efficiency. I only play with them as a hoby and far from being an electric engineer of any kind lol.
However, I know enough to get me by. A very good understanding of metals coming from the Iron workers industry, I've smelted a few things.

So the neat thing about running a Peltier is it's actually (often times) more difficult to cool it than the device it's trying to cool.
Then you have a target temp you're after. So with your case, lacking specifics, 1200w to cool 10w makes absolutely no sense to most people, While I feel you where aiming for some pretty low temps.

Low deltas. The key to any cooling device.

I've seen tower coolers outside them Canadian windows. -10c tower cooler with a TEC may yield -70c on the cold plate and -50c at the cpu. (just random numbers)

NaK, melting temperature of -12.6c but only 22.4 W/mK -
Water at 90c is a whopping 672.88 W/mk -

Why don't we just use water for TIM? lol.

 

[EarthDog]

I say the copper mass plays a part. Most tower coolers lack it.

it does play a role, but again, it simply delays the same end result (saying just increasing copper mass of a block). Think about it in an ln2 capacity. Remember when the higher core chips came out and suddenly the lighter pots couldn't hold temperatures as steady? The answer was to be a great pour bitch (lol) or move to a heavier pot. What this yielded was temps being more stable during runs. However, if you let it run temps will end up right back in the same place. Do your math... in this case it just takes a longer time to saturate a larger item.

So yes, mass matters, but only manages to delay the same end result in the case of thicker water blocks or base plates.

You dont want a thick/thicker ihs.. otherwise it would compound the problem we currently have (getting the heat out of the small die)... that is until it is heat saturated. Part of lapping the die has to do with getting the coating off AND thinning out the surface for better transfer. It takes less energy to saturate a 1.5" x 1.5x" x 0.05 IHS than a 0.25" thick plate. Once the plate is saturated then it's simply the metal properties in play. I'd bet good money says if you tested with a thick plate and thinner plate end temps would be the same after the system saturates (and to be clear, we're talking ambient cooled aio/air, water).

Edit: now this is within reason, obviously. If you add 250ml of water to a loop 1L loop, temps in the end wont change (just take longer to saturate). But if your resivior turns into a pool, clearly the meager loads wont heat that up to where it can 1L. Or throw a piece of copper the size of a cinder block, that may change. But the differences we are talking in blocks is largely irrelevant for normal cooling methods.

Edit2: Also something to consider... the grit. The finer the grit the more mirror like finish and the more smooth surface. However, surface area plays a role too. There is a point of diminishing returns though. The more mirror finish it is, the less surface area it actually has (less/finer ridges). So I wonder if it is really the finish or if additional sanding yields an even thinner ihs/base which improves results.....

 

[TheoneandonlyMrK]

The chips are not built using the same process , the different manufacturers do use similar tools to build their individual IP but the end result is they do some things slightly different.


Temperature reporting being one of these.

Before getting into the weed's on how hot a processor runs to effectively dissipate heat without damaging itself consider what's actually happening more.


These chips are very effectively tested and simulated to wear within a range while used within spec already, the research was done ,this Is how they scientifically resolved the best specs from the silicon.

Also the temperature shown the user means absolutely nothing, the actual Tdie temp is always higher and has an offset applied to it to stop user hysteria , they're already at their peak as many an overclocker can testify and those same types tend to know what degrades a chip.

HEAT and POWER not one in isolation , they're conducive to each other.

 

[freeagent]

I think they are just selling the CPUs pretty much maxed out from the factory, at least from Intel, and probably AMD too. In 2007/08 I was barely able to hit 4900mhz on my E8600 ES (Wolfdale). On my 3770K I can barely hit 4900mhz. On my X5690 I can run 4800, and barely hit 4900. Now you can go buy a CPU from them that will hit 4600-5000MHz out of the box, and maybe get a 200-300mhz overclock out of it, if you can tame the heat. Same goes for AMD, looks like you can get a few hundred MHz out of them, just like the old San Diego and Toledo chips where you got 300-500mhz out of them. So they give you some dividers so at least you can play with your ram to keep yourself entertained.. So maybe its a couple of things holding modern CPUs back. Die size being one, but that in conjunction with a CPU that is already near its maximum potential would be a logical explanation as to why these new CPUs are a bear to tame when it comes to thermals. My CPU is very easy to cool until I get to the end of its capabilities, and when I get to that point is where it sounds like a lot of guys are when they describe stock or mildly overclocked scenarios. In my limited experience CPUs became difficult to cool when the switch from 32nm to 22nm came, at least on the Intel side.. or maybe it was the switch from 22 down.

Don't mind me, I'm just an outsider looking in when it comes to new tech. To combat evil thermals I rely on big heatsinks, big fans, and brute force when needed, because some things don't change when it comes to overclocking.

 

[Deleted member 185158]

it does play a role, but again, it simply delays the same end result (saying just increasing copper mass of a block). Think about it in an ln2 capacity. Remember when the higher core chips came out and suddenly the lighter pots couldn't hold temperatures as steady? The answer was to be a great pour bitch (lol) or move to a heavier pot. What this yielded was temps being more stable during runs. However, if you let it run temps will end up right back in the same place. Do your math... in this case it just takes a longer time to saturate a larger item.

So yes, mass matters, but only manages to delay the same end result in the case of thicker water blocks or base plates.

You dont want a thick/thicker ihs.. otherwise it would compound the problem we currently have (getting the heat out of the small die)... that is until it is heat saturated. Part of lapping the die has to do with getting the coating off AND thinning out the surface for better transfer. It takes less energy to saturate a 1.5" x 1.5x" x 0.05 IHS than a 0.25" thick plate. Once the plate is saturated then it's simply the metal properties in play. I'd bet good money says if you tested with a thick plate and thinner plate end temps would be the same after the system saturates (and to be clear, we're talking ambient cooled aio/air, water).

Edit: now this is within reason, obviously. If you add 250ml of water to a loop 1L loop, temps in the end wont change (just take longer to saturate). But if your resivior turns into a pool, clearly the meager loads wont heat that up to where it can 1L. Or throw a piece of copper the size of a cinder block, that may change. But the differences we are talking in blocks is largely irrelevant for normal cooling methods.

Edit2: Also something to consider... the grit. The finer the grit the more mirror like finish and the more smooth surface. However, surface area plays a role too. There is a point of diminishing returns though. The more mirror finish it is, the less surface area it actually has (less/finer ridges). So I wonder if it is really the finish or if additional sanding yields an even thinner ihs/base which improves results.....

I agree.

And thus why I mention a larger cold plate in place of the IHS plate to be able to maintain a higher temp gradient...... The entire point of the Original post.
The rest, perhaps I was making examples more so than "doing the math" based on experience.

 

[TheoneandonlyMrK]

I agree.

And thus why I mention a larger cold plate in place of the IHS plate to be able to maintain a higher temp gradient...... The entire point of the Original post.
The rest, perhaps I was making examples more so than "doing the math" based on experience.

This is why I like EK full cover block's, I found a large block of copper to be more effective than some other types I tried with a machined thin coldplate in a housing, full mobo blocks exemplify this with liquid metal Tim in that they work well.

Oh and as someone who has run various crunching ,folding and mining machines long term at as high an efficiency as could be conveniently run that tends to equate to between 60-80% of maximum load 24/7 for consumer electronics effectively cooled to run below 80°C(typically 70-80°C in my case at all times including the rig in my specs).
More than this will create issues sporadically and death earlier.

They spec pro and server parts lower apparently for a reason.

 

[Deleted member 185158]

This is why I like EK full cover block's, I found a large block of copper to be more effective than some other types I tried with a machined thin coldplate in a housing, full mobo blocks exemplify this with liquid metal Tim in that they work well.

Oh and as someone who has run various crunching ,folding and mining machines long term at as high an efficiency as could be conveniently run that tends to equate to between 60-80% of maximum load 24/7 for consumer electronics effectively cooled to run below 80°C(typically 70-80°C in my case at all times including the rig in my specs).
More than this will create issues sporadically and death earlier.

They spec pro and server parts lower apparently for a reason.

And this further backs my point about NOT having a high temp gradient, Full cover waterblocks to keep a system running cool vs running the system warm.

Overclocking memory, you actively cool. This is generally a passive design, works well.... Cpus aren't quite there yet.

Larger 15w Ryzen mobo chipsets, the manufacturer puts a fan on the heat sink to keep costs down rather than using a larger heat sink or changing the material type to copper vs Aluminum .

Have done some number crunching(F@H). When I use the Cpu, I pick a lower power state for cooler temps and less worries. It does get the WU done a little slower, but that's cheaper than coming home to burnt hardware that was running a high temperature gradient.

Ryzen chips don't really "need" to run fast and hot. Nor Intel. That's something the consumer market seeks. But this time around, they just handed it to us at purchase. Overclocking, even stated by AMD, is a thing of the past, this is an accurate description. We only "tweak" what AMD has already accomplished with very small variances.

Any ways,
My take, unless I need a hot processor, I just don't run mine that way. I don't need 4.2ghz to browse and light gaming. Heck I can even still play triple A titles right off the max P-state with boost PBO CPB XFR disabled.

---------

Wish I had a 3000 series chip. I'd de-lid that too lol.. would be interesting for me, I've got some ideas for them naked chicks, Chips* (sorry spelling is off today)