12700K at 100C after few minutes of Prime95

[Outback Bronze]

Just curious for others like myself despite fixing limits that still hit 100c during prime95, what program did you end up using to ultimately determine if the cpu is stable in the long run as a result?

I use Asus Realbench. Haven't used Prime for years now.

 

[freeagent]

I use Linpack Xtreme, I haven’t used prime in eons..

 

[Deleted member 185158]

I use Linpack Xtreme, I haven’t used prime in eons..

Same. I like Intel Burn Test though. That before prime95.

None of this simulates general to heavy usage though. I mean we can stress a cpu all day, doesn't mean it'll game 4 hours straight (not that I ever do that, but some people do).

 

[Solid State Brain]

I'm probably too late, but I have a similar configuration with the same motherboard and CPU. The default load voltage on this motherboard is very high and easily causes thermal throttling and high power consumption in stress tests like Prime95. This is caused by a high "AC Loadline" which can be found under "CPU Lite Load" in Advanced CPU settings. Decreasing AC Loadline decreases the load voltage, while keeping voltage at low loads or idle mostly unaffected.

 

[ThrashZone]

I use Linpack Xtreme, I haven’t used prime in eons..

Hi,
Yeah not sure why people recommend using p95 it's just a power virus.

 

[freeagent]

Hi,
Yeah not sure why people recommend using p95 it's just a power virus.

In a way, Linpack is too. In the sense that there is not much, if any other load out there like it, at least that I have come across. But you get something out of it, that's why I use it.

 

[Solid State Brain]

It's probably not too good for thermal stressing since there are no other sustained loads like it, but Prime95 can be useful for making sure that CPU is truly stable during intense transient loads. Think your CPU is stable? Fire up Prime95 for a few seconds, and chances are that the PC will freeze if load voltages are insufficient.

If it does, you could then either increase load voltages or decrease the CPU current limit so that the CPU will downclock slightly during such intense loads and avoid locking up.

 

[ir_cow]

Hi,
Yeah not sure why people recommend using p95 it's just a power virus.

Prime95 is still the "ultimate" stability test. The issue with these newest Intel CPUs is that many AIOs can't handle it. Secondly, many waterblocks can't handle it lol. Basically unless you enforce PL1+PL2 = 250, it will easily hit 100c.

I'm in the mid 90s with a stock 12900K (1.15v) with a custom waterloop and the only way I can run prime95 so far with a overclock is to delid the CPU and use liquid metal. Thats exactly what I did with a 12700K.

So in short, dont run prime95 with the Intel 11th / 12th gen cpu.

 

[ThrashZone]

Hi,
I just use blender opendata and call it a done deal

 

[ir_cow]

Hi,
I just use blender opendata and call it a done deal

Pretty much if you can loop Cinebench or Blender your are good for the CPU. Use Y-Cruncher for a memory+CPU combo and that's about all you need to do. Is it guaranteed a good overclock? No, but its pretty good.

 

[tpu7887]

The following is the best advice by far. That is, if a stable system with the lowest possible voltage is what you're after...

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 method you choose to use)

 

[ThrashZone]

Pretty much if you can loop Cinebench or Blender your are good for the CPU. Use Y-Cruncher for a memory+CPU combo and that's about all you need to do. Is it guaranteed a good overclock? No, but its pretty good.

Hi,
For my use case it's fine

 

[tpu7887]

Hi,
For my use case it's fine

Did you miss my post? Not interested in guaranteed stability?

 

[freeagent]

No, but its pretty good

Or close enough

Passes everything except one thing? I just pretend the program doesn't exist, unless a bout of OCD strikes.

 

[ThrashZone]

Hi,
Not crunching
Not folding anything besides laundry
Not mining
My benchmarks are just fine I run plenty and never had any stability issues stop me from running a benchmark
So if you think p95 is some sort of score board benchmark we will agree to disagree there because it's not so no point in using it.
But please feel free to use p95 I'll use what I want to.

By the way W1zard uses blender to
Why because it's a realistic workload.

 

[Mussels]

This sadly is what i've been trying to explain to people about how the current intel systems work

They flip flop between PL1 and PL2, and have issues sustaining multi threaded performance unless you have a lot of cooling - you get the performance, but unless you meet certain criteria (unlocked power settings and very high end cooling) you cant sustain it.


There are definitely workarounds, your main choices would be lower PL1 and PL2 limits so the cooler has a chance to dissipate heat before it saturates completely, or an all core overclock at more efficient settings

(Ex: my 5800x runs at 4.4 (all core) - 5.05 (1-2T) at 75C, but a static 4.6GHz runs at 50C)


Also since intel platforms often have AVX offsets, make sure you're testing with AVX and non AVX programs and both sets of clocks are stable.
There's no use only testing a system for non AVX workloads, since games are using them too - you're either going to get lower clocks and performance than you expect, or instability by ignoring it.

 

[Solid State Brain]

By the way W1zard uses blender to
Why because it's a realistic workload.

The problem is that Blender or Cinebench are not sufficiently intense stability tests. The CPU can be effectively stable with them but then fail within a few seconds (before thermal throttling is even reached) with Prime95 AVX2/SmallFFT, or also specialized MT workloads like Linux kernel compilation.

A CPU that can freeze at any moment depending on the software used cannot be really considered stable in my opinion. A solution to this however is making sure that the CPU cannot reach in the first place such unstable operating points, which could be done not only with increased voltages, but also with current - power - temperature limits. By lowering these limits, lower CPU voltages than normally possible can even be used, since the CPU will be more prone to throttling when pushed to the limit.

As a side note, the opening post was about high temperatures with Prime95 despite using a Noctua NH-D15, and one big reason for that are the high voltages that the MSI Z690-A Pro applies with default settings. Well, at least they will lead to stable CPU operation. That Noctua should be able to handle the thermal load without throttling up to about 200-210W, but it will not be silent.

 

[ratirt]

damn a 100c with a noctua air cooler. Should it even get that high despite what workload it is? 100c is literally throttling and yet people say their Intel 12th gen CPU run cool and dont use a lot of power and yet you have these reports.
Maybe limit the power to 125w for PL1 and PL2 would help since I clearly don't see any other choice. Unless you fiddle with the settings every now and then to make it stay cool enough.

 

[ThrashZone]

The problem is that Blender or Cinebench are not sufficiently intense stability tests. The CPU can be effectively stable with them but then fail within a few seconds (before thermal throttling is even reached) with Prime95 AVX2/SmallFFT, or also specialized MT workloads like Linux kernel compilation.

A CPU that can freeze at any moment depending on the software used cannot be really considered stable in my opinion. A solution to this however is making sure that the CPU cannot reach in the first place such unstable operating points, which could be done not only with increased voltages, but also with current - power - temperature limits. By lowering these limits, lower CPU voltages than normally possible can even be used, since the CPU will be more prone to throttling when pushed to the limit.

As a side note, the opening post was about high temperatures with Prime95 despite using a Noctua NH-D15, and one big reason for that are the high voltages that the MSI Z690-A Pro applies with default settings. Well, at least they will lead to stable CPU operation. That Noctua should be able to handle the thermal load without throttling up to about 200-210W, but it will not be silent.

Hi,
On the same note you can run p95 until hell freezes and bsod on BF5
Again blender uses a realistic workload with avx content so gaming you're better off not worse off using it instead

Besides you can use blender opendata or demo rendering files for cpu and gpu testing so much better.
Blender Open Data — blender.org

Demo Files — blender.org


R15-R20-R23 are indeed silly testing loads, none use much of any avx content to trigger any offsets if used even looped
But one thing these three have in common is they are benchmarks and there are leader boards.

 

[Solid State Brain]

Hi,
On the same note you can run p95 until hell freezes and bsod on BF5

It also depends on how you're running Prime95. For instance, under thermal-throttled conditions you're not maximizing internal peak currents, just thermally stressing the CPU.

R15-R20-R23 are indeed silly testing loads, none use much of any avx content to trigger any offsets if used even looped
But one thing these three have in common is they are benchmarks and there are leader boards.

Cinebench R23 causes about the same power draw as with Blender during rendering (which I use for many purposes, not just benchmarks or stress testing), I think it's about on par as for CPU load.

damn a 100c with a noctua air cooler.

I have about the same configuration as the OP, with same CPU-motherboard and Noctua cooler but one less fan (NH-D15S). At maximum fan speed it is able to dissipate about 200W keeping temperatures slightly below thermal throttling, with a 28C ambient temperature (it will of course work better in the winter). The dual-fan version might be able to fare slightly better. It's a good air cooler but it can't perform miracles.

EDIT: it's not that the cooler gets hot (it is only somewhat warm to the touch under such testing conditions), it has a limit to how quickly it can extract heat from the CPU.

 

[tpu7887]

This sadly is what i've been trying to explain to people about how the current intel systems work

They flip flop between PL1 and PL2, and have issues sustaining multi threaded performance unless you have a lot of cooling - you get the performance, but unless you meet certain criteria (unlocked power settings and very high end cooling) you cant sustain it.


There are definitely workarounds, your main choices would be lower PL1 and PL2 limits so the cooler has a chance to dissipate heat before it saturates completely, or an all core overclock at more efficient settings

(Ex: my 5800x runs at 4.4 (all core) - 5.05 (1-2T) at 75C, but a static 4.6GHz runs at 50C)


Also since intel platforms often have AVX offsets, make sure you're testing with AVX and non AVX programs and both sets of clocks are stable.
There's no use only testing a system for non AVX workloads, since games are using them too - you're either going to get lower clocks and performance than you expect, or instability by ignoring it.

I like your method mixed with mine.


"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"



Then, set a CPU power limit in the BIOS at 3-5 watts over the amount reached during the above stability test (the power dissipated in the (usually) most demanding program). Then, if something out of the ordinary happens, your system doesn't hang! (the extra 0.02-0.03V easily allows for the extra 3-5 watts dissipation)


edit: like a lot of people I do think running P95 on all cores is a stupid way to ensure stability for systems that don't run loads anywhere near as demanding, but P95 applied as above should be the standard method

 

[ratirt]

It also depends on how you're running Prime95. For instance, under thermal-throttled conditions you're not maximizing internal peak currents, just thermally stressing the CPU.




Cinebench R23 causes about the same power draw as with Blender during rendering (which I use for many purposes, not just benchmarks or stress testing), I think it's about on par as for CPU load.




I have about the same configuration as the OP, with same CPU-motherboard and Noctua cooler but one less fan (NH-D15S). At maximum fan speed it is able to dissipate about 200W keeping temperatures slightly below thermal throttling, with a 28C ambient temperature (it will of course work better in the winter). The dual-fan version might be able to fare slightly better. It's a good air cooler but it can't perform miracles.

EDIT: it's not that the cooler gets hot (it is only somewhat warm to the touch under such testing conditions), it has a limit to how quickly it can extract heat from the CPU.

That is why it is weird. I got two TR 3970x's with the Noctual U15 (if I remember correctly) and I can run them at full load no problem and the temp never gets to a 100 and still it is a 32c CPU.
OP got something wrongly mounted or thermal paste problem. Maybe air circulation in the chassis sucks so bad.

 

[pavle]

Not sure how CPU-demanding God of War is, but that sounds pretty good?

IIRC all-out AVX2 loads like default Prime95 small FFT will just run like an inferno on 11th and 12th gen. Aside from a small AVX offset (do they even do that anymore) I don't think Intel significantly throttles power usage during P95. The only reason Ryzen doesn't do the same is because AMD's boost algorithm is hardcoded to throttle insanely hard upon detecting small FFT AVX. AT tested the 12900K when it came out and it was also running balls-to-the-wall at 272W in their own AVX2 test, which I don't expect to be more demanding than P95 Small.

12700K is getting toward the upper end of what an air cooler on Alder Lake can realistically handle anyways.

The socket bending issue shouldn't a problem for a brand-new system that was just assembled 2 days ago. The warping happens over time, and I'd expect your temperatures to be noticeably bad in other workloads as well such as games.

As the quoted poster said and I add: you're running a basically at least 200W CPU (at 100%) cooled by an air cooler. Of course it is going to get hot.

 

[Solid State Brain]

That is why it is weird. I got two TR 3970x's with the Noctual U15 (if I remember correctly) and I can run them at full load no problem and the temp never gets to a 100 and still it is a 32c CPU.
OP got something wrongly mounted or thermal paste problem. Maybe air circulation in the chassis sucks so bad.

The AMD Threadripper 3970X should have a significantly lower heat density and will be easier to cool for the same power than CPUs with smaller dies.

Furthermore, as I mentioned earlier, with default settings the same motherboard as the original poster overvolts the CPU and uses very high limits (288W / 512A after choosing "Tower Air Cooler" on the initial setup), and under such conditions with Prime95 AVX2 / Small FFT power draw can easily go above 250W with the 12700K, at least while it is below the thermal throttling point.

 

[ratirt]

The AMD Threadripper 3970X should have a significantly lower heat density and will be easier to cool for the same power than CPUs with smaller dies.

Furthermore, as I mentioned earlier, with default settings the same motherboard as the original poster overvolts the CPU and uses very high limits (288W / 512A after choosing "Tower Air Cooler" on the initial setup), and under such conditions with Prime95 AVX2 / Small FFT power draw can easily go above 250W with the 12700K, at least while it is below the thermal throttling point.

With the die size sure. Smaller is harder to cool but 12700k is not a small one. 215mm2 is quire big in comparison to a 5800x which is 81mm2 and you can still cool it with any noctua cooler no problem. Even though there were some complaints about the 5800x being hot as hell which I could not see. Hotter than a 5900x for instance but overheating and hot as hell it was not.

Obviously, if you cooler can't manage, limit the processor till you feel comfortable using it.