12700K at 100C after few minutes of Prime95

[Rob6502TPU]

I'm not surprised by the 100C. I have to downclock to 4.2GHz for the Prime95 VRM tests just to keep it below 290 watts. Solution? Don't run Prime95 because it will pull 330+ if allowed.

If you don't thermal throttle with CineBench R20 after looping it for a bit, you are golden.

Come on this is a nonsense, Intel & Mobo vendors have together conspired to overclock by default for benchmark scores.
These CPUs are NOT throttling to TDP due to rotten tau settings and are NOT restricting to spec PL2.
This "run yer PC" gently is a result of shabby practices. Developers are one slip away from creating runaway power hogs that can burn >300w and the tech press are letting Intel get away with this.
These boards/CPUs are defective not observing design specifications.

 

[AM4isGOD]

yet people say their Intel 12th gen CPU run cool and dont use a lot of power

He was running prime 95, you think they use max power and run hot all the time? mine runs at 80c and uses 200 watts just idle on windows /s. of course it is going to run hot and use a lot of power running prime 95.

 

[Solid State Brain]

Come on this is a nonsense, Intel & Mobo vendors have together conspired to overclock by default for benchmark scores.
These CPUs are NOT throttling to TDP due to rotten tau settings and are NOT restricting to spec PL2.
This "run yer PC" gently is a result of shabby practices. Developers are one slip away from creating runaway power hogs that can burn >300w and the tech press are letting Intel get away with this.
These boards/CPUs are defective not observing design specifications.

In reality there's no real 'spec' PL1/PL2/Tau, only Intel-defined recommendations. You (the system builder/integrator) are supposed to configure them according to your system capabilities and needs; they may be set higher than default/recommended if the system can support them. IccMax however (current limit) is part of the processor's electrical specifications and is not intended to be exceeded.

The problem is largely the defaults settings used by motherboard manufacturers, which assume (or pretend) that end users all have high-end water cooling systems and want performance no matter the cost (power consumption, temperatures, etc).

Images from the pdf version of the datasheet here: https://edc.intel.com/content/www/u...-1-of-2/001/processor-line-thermal-and-power/

 

[ratirt]

He was running prime 95, you think they use max power and run hot all the time? mine runs at 80c and uses 200 watts just idle on windows /s. of course it is going to run hot and use a lot of power running prime 95.

If it was Furmark I would not make a fuss around it but prime95 is ok for testing. If you get 100c on the CPU something is not ok. Not sure what you mean by run hot all the time? if you use it at 100% yeah it will be hot all the time and when you idle it wont. The problem is not to have a temp of a 100c when you have to use it despite time span while it is 100% utilized.

 

[Vario]

Just don't run Prime95. If its <80C for your normal applications (games, software, etc) don't worry about it.

 

[tpu7887]

If it was Furmark I would not make a fuss around it but prime95 is ok for testing. If you get 100c on the CPU something is not ok. Not sure what you mean by run hot all the time? if you use it at 100% yeah it will be hot all the time and when you idle it wont. The problem is not to have a temp of a 100c when you have to use it despite time span while it is 100% utilized.

It seems my posts on prime95 have been missed or overlooked or skimmed by many. The reason people aren't using P95 is because it requires more power than all of their real world loads, and by a large margin. Problem is, none of the programs they're using to check for stability are thoroughly checking for errors. Below is how to overclock with the lowest voltage while ensuring error free operation (which, correct me if I'm wrong here... is what we all want).


Most demanding programs people run (like games) only make CPUs draw 50-70% of the amount of power that prime95 does. A CPU running at 5GHz with 1.3V can run perfectly fine at 80 degrees, but at 95 be throwing errors left and right- 1.34V might be required for proper operation at 95. But then 1.355V is required for temps which are 99 at 1.34, making 1.365V what's actually required. And you're then at 100, thermal throttling.
If none of the real world loads ever demand power levels as high as P95 small FFT on all cores, 1.35V isn't needed for stability, just 1.3V.

I laid out an easy method to tune to the optimal hypothetical 1.3V using Prime95. Once that's done, the CPU can be power limited so that it never reaches over 80 deg C during operation, and all is good! It will run just as error free in all applications, as if you tested Prime 95 on all cores at once, requiring much less voltage or allowing higher frequencies (take your pick on which you'd like). Caveat: only up to the power levels that normal loads demand of the CPU (which is not a problem because you tuned the power to the real world maximum...)

edit: I put the (very) easy method below for reference. You do with P95 the same as you would if running on all cores. 90-180 minutes coarse adjustments, 30+ hours on the final run

If you don't know already, find out which of your cores are weakest (using prime95 at a lower voltage and clock speed, one that doesn't raise temps above what your real world loads reach). Then, run Prime95 on those cores, adding load to the remaining cores so that your weak cores running Prime95 reach the maximum temperature they do in real-world scenarios. Whatever voltage is required to prevent errors, add an additional 0.02 to 0.03V to ensure stability.

edit: To add heat, google "CPU burn in", in the first result download the 20kB file, each instance is one thread. You can assign them to cores as needed using task manager

edit2: If you're using a high voltage, periodically re-check stability every few months (this goes for any overclock)

 

[Rob6502TPU]

In reality there's no real 'spec' PL1/PL2/Tau, only Intel-defined recommendations. You (the system builder/integrator) are supposed to configure them according to your system capabilities and needs; they may be set higher than default/recommended if the system can support them. IccMax however (current limit) is part of the processor's electrical specifications and is not intended to be exceeded.

The problem is largely the defaults settings used by motherboard manufacturers, which assume (or pretend) that end users all have high-end water cooling systems and want performance no matter the cost (power consumption, temperatures, etc).

Images from the pdf version of the datasheet here: https://edc.intel.com/content/www/u...-1-of-2/001/processor-line-thermal-and-power/

Exactly Intel are allowing configurations which are running extremely hot and just turning a blind eye. That means a race to the max power to "win" benchmarks.
The users should NOT have to worry about what defaults the mobo manufacturs use and let's be clear. INTEL IS AT FAULT HERE.
The processors I see are advertised at far lower TDP than they would actually be run at.

 

[ir_cow]

Come on this is a nonsense, Intel & Mobo vendors have together conspired to overclock by default for benchmark scores.
These CPUs are NOT throttling to TDP due to rotten tau settings and are NOT restricting to spec PL2.
This "run yer PC" gently is a result of shabby practices. Developers are one slip away from creating runaway power hogs that can burn >300w and the tech press are letting Intel get away with this.
These boards/CPUs are defective not observing design specifications.

No they are not defective, Intel is to blame for marketing the CPUs with lower TPD than they really have. If Intel really cared they would have enforced a lower power target. Every Z690 I've come across so far automatically sets P1+P2 = to 4096. Anyway you slice it. If you have a 12900K, it will pull 241 watts in Cinebench, Blender, etc. Motherboards have a option to allow the CPU to boost higher, thus more power. Using P1+P2= 4096 doesn't mean it will draw that much, just means the motherboard is no longer limiting the power load requested by the CPU.

 

[phanbuey]

Exactly Intel are allowing configurations which are running extremely hot and just turning a blind eye. That means a race to the max power to "win" benchmarks.
The users should NOT have to worry about what defaults the mobo manufacturs use and let's be clear. INTEL IS AT FAULT HERE.
The processors I see are advertised at far lower TDP than they would actually be run at.

This is true for the K series as a whole, kind of the whole point of the K series is unlocked everything. If you let your motherboard set everything unlocked, the chip will run hot when you run prime 95 . Not seeing the issue or fault here.

You don't see these threads with non-K CPUs for this reason. Intel locks these parts down and sells them cheaper with non-K branding that run at the at lower TDP within 5-10% of the performance of the K skus.

This has been true since the pentium EE days. Core xtreme or whatever they called it - all the unlocked K chips run hot AF even back down to ivy bridge.

 

[Rob6502TPU]

This is true for the K series as a whole, kind of the whole point of the K series is unlocked everything. If you let your motherboard set everything unlocked, the chip will run hot when you run prime 95 . Not seeing the issue or fault here.

You don't see these threads with non-K CPUs for this reason. Intel locks these parts down and sells them cheaper with non-K branding that run at the at lower TDP within 5-10% of the performance of the K skus.

This has been true since the pentium EE days. Core xtreme or whatever they called it - all the unlocked chips run hot AF even back down to sandy/ivy bridge.

Unlocked is different from arriving with stupid settings. You should be able to increase power, alter tau. If you do that you can be ready for thermals.

 

[phanbuey]

Unlocked is different from arriving with stupid settings. You should be able to increase power, alter tau. If you do that you can be ready for thermals.

I get that -- they definitely could have had a more optimized setting between 288W and 125W. 100C is scary looking but within the safe limits of the chip - but:

I think we can all safely agree they missed the boat here a bit with the 51W for an extra 2.1% performance.

 

[ir_cow]

The 12900KS has a Tj.Max of 115C. Same chip. 100c is looking to be a bit more safe

I agree Intel should have just gone with 190W instead. Enforce it by default and the user has to change it in the BIOS. Oh well , next time I guess.

 

[Solid State Brain]

The 12900KS has a Tj.Max of 115C. Same chip. 100c is looking to be a bit more safe

Unclear where this information came from. On the Intel ark website it is listed at 90 °C and the datasheet mentions that the thermal control circuit (TCC) activation temperature (i.e. thermal throttling) by default is at 90 °C. It can be manually configured up to 115 °C, though.

Product Specifications

quick reference guide including specifications, features, pricing, compatibility, design documentation, ordering codes, spec codes and more.

ark.intel.com

https://cdrdv2.intel.com/v1/dl/getcontent/655258

 

[ir_cow]

@Solid State Brain It comes from the TPU 12900KS review. Every MB I've tested has the default at 100c for the 12900K (none S). Expect one, which was the ASrock ITX. That was 115c and it shouldn't have been.

Intel Core i9-12900KS Review - The Best Just Got Better

The Intel Core i9-12900KS is the company's new flagship Alder Lake processor. After our review, we can confirm that it is the "world's fastest gaming CPU," but that comes at a price not only in terms of dollars, but increased power draw and heat output, too.

www.techpowerup.com

That's why my approach was a bit different from the usual for the best overclock. For the 12900KS, I started with all cores set to the same x49 multiplier for 4.9 GHz all-core and increased the voltage until I got close to the thermal limit of 115°C when running Prime95. Unlike other Alder Lake processors, including the Core i9-12900K, the KS has its out-of-the-box default thermal limit set to 115°C, not 100°C. 115°C is the highest manual setting you can pick in the BIOS for all Alder Lake CPUs. Since this is the default setting, Intel's warranty will cover operating the processor at up to 115°C, which is a good hint for users of other Alder Lake processors based on the same silicon; there's no reason those can't withstand 115°C, too.

Edit: TJunction and T.J Max are not the same thing.

 

[Solid State Brain]

I mean, what is the source for this information from the TPU review (edit: suggesting that 115°C is the default for the 12900KS)? Public data from Intel appears to suggest that the default TjMax for the 12900KS is actually lower than other CPUs in the Alder Lake lineup, which could possibly be beneficial for stability. This could have been a BIOS bug or mishap—again, motherboard defaults shouldn't be trusted to represent what Intel actually recommends.

Edit: TJunction and T.J Max are not the same thing.

The Ark website shows this by clicking on the [?] icon. It's referring to Tjunction as a maximum allowed temperature:

 

[ir_cow]

@Solid State Brain. If the warranty covers 115c, than it isn't a bug.

 

[Solid State Brain]

On my MSI Z690 motherboard an ominous message appears if I set the temperature limit above 105 °C. It doesn't inspire much confidence on 115 °C being a safe continuous limit. It is also counterintuitive that the more power-hungry i9-12900KS based on the same silicon would be fine at higher temperatures:

 

[ratirt]

On my MSI Z690 motherboard an ominous message appears if I set the temperature limit above 105 °C. It doesn't inspire much confidence on 115 °C being a safe continuous limit. It is also counterintuitive that the more power-hungry i9-12900KS based on the same silicon would be fine at higher temperatures:

View attachment 260969

100c is a lot. I would never have gone above that and to be fair I would have done whatever possible to drop the temp at least to 90c.
There must be some misunderstanding. I doubt it can withstand 115c long term without damage.

 

[mclaren85]

Dictating not to run Primetest95 is not a valid suggestion. Primetest is one of the best softwares to emulate render performance. What if the op is a heavy premiere pro or topaz video enhance user?

 

[Solid State Brain]

100c is a lot. I would never have gone above that and to be fair I would have done whatever possible to drop the temp at least to 90c.
There must be some misunderstanding. I doubt it can withstand 115c long term without damage.

The setting has a range of 90–115 °C, but the effective range can be different depending on whether a temperature offset is present, and I believe that on B-series motherboards it cannot be configured. On my system I've set it to 90 °C given that the 12900KS is supposed to have it at the same temperature according to Intel data. Techpowerup appears to be the only source suggesting 115 °C.

A lower temperature limit also means that the CPU will draw less power on sustained heavy loads with high PL2, although Intel suggests that it is not intended to be used as a means to maintain the TDP / PL1.

It's unfortunate that the temperature of the heat spreader (Tcase) cannot be directly monitored, since it is the CPU temperature that actually has a maximum specification in the same datasheets (59–71 °C depending on the model, and guess what: the 12900KS is on the lower end of this). On my motherboard there is "socket temperature" which I think might be a good approximation for that. If that is a good assumption, that temperature has a certain thermal inertia and the default 56s Tau makes a lot of sense with it. With a high ambient temperature (30 °C) at moderate fan speeds it also appears that the limit can be sustained only at a reasonable PL1 (e.g. 125W). In other words, I have doubts that sustained usage at a high PL2, even if core temperature is below TJmax, is actually truly within specifications.

 

[Deleted member 185158]

the temperature of the heat spreader (Tcase) cannot be directly monitored,

That's it right there.

Software is made aware through algorithm, unless a direct reading is used otherwise, CoreTemp generally the most accurate in my opinion.

But nothing beats a probe on an IHS plate.

Tcase max should be awareness to the user that any throttling can happen at or beyond this temp point.

If this where an AM4 AMD chip, the high temp alert is 70c - A default system with a core temp of 70c should command cpu fan 100% duty cycle. This can be changed up to only 75c by the user. (A lot of complaints on stock coolers of fans revving up). Thus the Tcase IS configurable of 5c on AMD systems.

I always viewed the Tcase to be the "suggested" all core sustained load for long periods of time. This is more my opinion because it would never be stated in documentation in this fashion.

Mind you I don't look at TDP as an energy usage, but more so as a thermal dissipation. If you think about 70c

IE: 70c would be converted to roughly 239 BTU/hr.

Maybe more of a challenge of how many thermals can you move in a short period of time rather than seeing if the CPU can hit it's thermtrip (typically around 110-115c) and be the golden winner of a burnt cpu.

 

[Solid State Brain]

Here is a graphical representation of socket temperature on my motherboard during a Prime95 Load. On my CPU, PL1=125W, PL2=190W, Tau=56s , which is according to Intel recommendation. However, I set TJmax to 90 °C as I mentioned earlier.

With this setup and 29 °C ambient temperature, socket temperature initially ramps up relatively quickly (but not instantly) to a certain level while core temperature is pegged at 89–90°C. After the time at PL2 is exhausted (with a 56s Tau), socket temperature slightly backs down, then starts increasing slowly again. In other tests I observed that on the long term this levels off at about 61 °C.

If socket temperature can be assumed to be similar to case temperature/Tcase, short periods at TjMax should be fine for long-term reliability.

I honestly believe this is the way Intel intended these processors to be used.

 

[Rob6502TPU]

I get that -- they definitely could have had a more optimized setting between 288W and 125W. 100C is scary looking but within the safe limits of the chip - but:

I think we can all safely agree they missed the boat here a bit with the 51W for an extra 2.1% performance.

Absolutely you get it. Most 12th gen purchasers though see a benchmark "win" and claim it's efficient using less power in normal use. Effectively they ignore that constanty heavily loading these processors with mobo settings as delivered will probably end badly and that more power is used than necessary even when loaded lighter. Chips & Cheese article found the "race to idle" being efficient proved to be total BS

Here is a graphical representation of socket temperature on my motherboard during a Prime95 Load. On my CPU, PL1=125W, PL2=190W, Tau=56s , which is according to Intel recommendation. However, I set TJmax to 90 °C as I mentioned earlier.

..
If socket temperature can be assumed to be similar to case temperature/Tcase, short periods at TjMax should be fine for long-term reliability.

I honestly believe this is the way Intel intended these processors to be used.

A more plausible explanation is that E cores were intended originally to run efficiently for background tasks.
Then the Golden Cove cores bloated in the effort to beat Zen in ST, which meant they could not use more P cores, so E cores were used for the cache light creator type software dominating MT benchmarks used by reviewers. That had the side effect of making AVX512 problematic.
Then what happened is the marketing people whacked up power limits to achieve the imperative benchmarking goal, ignored TDP so the i9 12900k could match Zen3 16c in the benchmarks.
In companies with Zen3 & ADL OEM built PCs used by developers, the ADL ones suffer instability through over heating.

They've pushed the responsibility onto motherboard makers, SIs and OEMs; the 51% power increase for 2.1% performance is NOT something a competent CPU designer would intend.
You're kidding yourself, if you hit a problem all the responsibility is on you for buying an unlocked CPU and not managing thermals better.

 

[cadaveca]

the 51% power increase for 2.1% performance is NOT something a competent CPU designer would intend.

Actually, given how silicon works, it IS something that is planned for. That reading, for example, of 51% for 2.1% performance, is for that chip only. Another chip of the same type and model may only increase power use by 25% under the same workload due to having better silicon quality.

So not directly intend, but planned for, for sure, since this is inherent in how silicon chips work.

 

[Solid State Brain]

A more plausible explanation is that E cores were intended originally to run efficiently for background tasks.
Then the Golden Cove cores bloated in the effort to beat Zen in ST, which meant they could not use more P cores, so E cores were used for the cache light creator type software dominating MT benchmarks used by reviewers. That had the side effect of making AVX512 problematic.
Then what happened is the marketing people whacked up power limits to achieve the imperative benchmarking goal, ignored TDP so the i9 12900k could match Zen3 16c in the benchmarks.
In companies with Zen3 & ADL OEM built PCs used by developers, the ADL ones suffer instability through over heating.

They've pushed the responsibility onto motherboard makers, SIs and OEMs; the 51% power increase for 2.1% performance is NOT something a competent CPU designer would intend.
You're kidding yourself, if you hit a problem all the responsibility is on you for buying an unlocked CPU and not managing thermals better.

I only meant that running the CPU a high PL2 for periods in the order of a minute or so and then keeping it at a relatively low PL1 for the rest of the computation time makes sense if the intended aim is taking advantage of the "available thermal capacitance". From the datasheet linked earlier:

Clearly, they're not referring about core temperature, since it has almost no thermal inertia and reacts immediately to load changes. On the other hand, externally-measured temperatures very close to the CPU itself (e.g. socket temperature as I showed in the image posted the other day) react with speed consistent with the Intel-recommended Tau time.

Of course one might argue that the 240+ watts is an excessively high power draw, even if for short periods. That is a related but different subject.