[SOLVED] 3900X heating up to 95 C in stress test,normal?

[frozensun]

Hey friends.
Well actually I thought it's normal but googling about it saw it isn't.
I ran Aida64 stability test yesterday with HWInfo and I remember that T die was reaching 96 C which I suppose is way too hot.
I'm using H100i Platinum which should be more then enough for this CPU.
Cooler is mounted okay I suppose,screws don't loose and all was installed brand new back in 7/2020 (my work).
Since I dismantled the case I will put it back tomorrow and try to test in prime95 small FFT and report back or you want me to test in some other benchmark.
I will write you the voltage of the CPU,readings from BIOS I suppose are most accurate.
I didn't OC the CPU,nor did I flash with latest BIOS (never did).
I mean I don't use PC in synthetics tests but I just want it to work normal.
Could be the issue that I set up fans of an AIO to balanced?
What should I do if this is heating issue?
p.s. I remember jay2cents doing OC with 3900x and Crosshair Hero VIII and telling that the boards sets by default too high voltage to the CPU (was it 1.4 V I really can't remember),I can try to find that vid.
I'm on Crosshair Hero VIII.

 

[drea.drechsler]

h100i should be enough cooler, although 95C seems a bit hot even for Aida64 stress testing a 12 core CPU. But do keep in mind that stress tests like that are highly unrealistic. The idea running them is to simply survive it without crashing, not necessarily maintain desireable 24/7 temperatures.

Try something a bit more realistic to know if your cooling is adequate: I'd run Cinebench23 in the 'stress test' mode that runs about 30 minutes. It's a real-world rendering application that makes extensive use of AVX instructions.

If you've left the processor on AUTO clocks and voltage then it will protect itself when it gets really hot. When temps are up in the 90's in a stress test like Aida, for instance, the voltage should be down around 1.25V or maybe even less. Look at the (SVI2 TFN) core voltage as that's the more important to watch. It will only go up to 1.4V (and higher) when one core boosts in light threaded workloads.

Some motherboards and CPU's can benefit with slightly lowered voltage though. If you want to lower voltage do it with offsets only and a little bit at a time. You should test single thread performance though as you can easily kill it with too low of a voltage. Single thread is what games need most so it's important.

In BIOS there should be a setting for platform thermal limit. Find it and set it to 85 or 90C and temp won't go above it even running Aida.

 

[frozensun]

h100i should be enough cooler, although 95C seems a bit hot even for Aida64 stress testing a 12 core CPU. But do keep in mind that stress tests like that are highly unrealistic. The idea running them is to simply survive it without crashing, not necessarily maintain desireable 24/7 temperatures.

Try something a bit more realistic to know if your cooling is adequate: I'd run Cinebench23 in the 'stress test' mode that runs about 30 minutes. It's a real-world rendering application that makes extensive use of AVX instructions.

If you've left the processor on AUTO clocks and voltage then it will protect itself when it gets really hot. When temps are up in the 90's in a stress test like Aida, for instance, the voltage should be down around 1.25V or maybe even less. Look at the (SVI2 TFN) core voltage as that's the more important to watch. It will only go up to 1.4V (and higher) when one core boosts in light threaded workloads.

Some motherboards and CPU's can benefit with slightly lowered voltage though. If you want to lower voltage do it with offsets only and a little bit at a time. You should test single thread performance though as you can easily kill it with too low of a voltage. Single thread is what games need most so it's important.

In BIOS there should be a setting for platform thermal limit. Find it and set it to 85 or 90C and temp won't go above it even running Aida.

I dropped CPU VCore to 1.3 V and exit from BIOS.
Then I ran Prime95 small/smallest FFT and instantly in a minute temperature reaches 92 C and increases.
This for sure can't be good,but what is happening,I lowered the voltage.

 

[drea.drechsler]

I dropped CPU VCore to 1.3 V and exit from BIOS.
Then I ran Prime95 small/smallest FFT and instantly in a minute temperature reaches 92 C and increases.
This for sure can't be good,but what is happening,I lowered the voltage.

Under something like Prime95 voltage might go as low as 1.2V if not fixed at 1.3V like you've done. And expect very high temps with Prime95 smallFFT.

 

[frozensun]

Under something like Prime95 voltage might go as low as 1.2V if not fixed at 1.3V like you've done. And expect very high temps with Prime95 smallFFT.

Nope,only me having this issue.Other guys get max temp at 80-82 C.
The voltage is constant.

 

[drea.drechsler]

...
The voltage is constant.

That's not helping it. Put it back to auto and see where the voltage goes to in the exact same Prime95 FFT.

 

[getochkn]

Ignore prime. Your cpu will never be under that stress and prime stresses it even more. If you're obsessed with synthetics, which you say you aren't, then st try cinebench. Fyi, I run a 3900x with stock cooler, auto oc turned off, 1.25v, 4.2ghz set clock speed and with 24 threads rendering in an actual app and not synthetics, I still only hit 70c or so. I'm sure prime would hit me to 100c with my stock cooler, but who cares. Unless the sole purpose of this pc is to run prime all day, its pointless. Does it run, is it stable gaming, yes, move on.

 

[frozensun]

getochkn yea totally right...got my self obsessed with Prime95 which is syntethic benchmark and CPU will probably never push in any app in that way.
So I switched the Voltage to Manual and put 1.3000 V.
Should I revert to AUTO because board raises pretty high auto voltage to even 1.47 V.

 

[drea.drechsler]

...
Should I revert to AUTO because board raises pretty high auto voltage to even 1.47 V.

Absolutely yes, put it to auto. AMD has said it's perfectly normal to hit voltage as high as 1.5V when it boosts in light, bursty workloads. It's designed to do that so let it work as designed.

Let it work as designed and it performs best. Locking voltage prevents it from working right and even hurts performance in some cases. When processing heavy workloads it lowers voltage even lower than 1.3V and not letting it do that is much worse for it anyway.

I get 4.425ghz while gaming and 4.2-4.25Ghz while rendering (handbrake and Cinebench) on my 3700X with a very slight negative offset (-0.0125V) but in AUTO and PBO on. By many accounts a 3900X should be able to giet 4.5-4.6Ghz in gaming and 4.2Ghz in AUTO and PBO when cooled with a 240mm AIO. If you want to fiddle with all-core overclocking go ahead, but start at 1.2V unless you're willing to risk your CPU.

 

[frozensun]

Absolutely yes, put it to auto. AMD has said it's perfectly normal to hit voltage as high as 1.5V when it boosts in light, bursty workloads. It's designed to do that so let it work as designed.

Let it work as designed and it performs best. Locking voltage prevents it from working right and even hurts performance in some cases. When processing heavy workloads it lowers voltage even lower than 1.3V and not letting it do that is much worse for it anyway.

I get 4.425ghz while gaming and 4.2-4.25Ghz while rendering (handbrake and Cinebench) on my 3700X with a very slight negative offset (-0.0125V) but in AUTO and PBO on. By many accounts a 3900X should be able to giet 4.5-4.6Ghz in gaming and 4.2Ghz in AUTO and PBO when cooled with a 240mm AIO. If you want to fiddle with all-core overclocking go ahead, but start at 1.2V unless you're willing to risk your CPU.

Yeaah I agree.
Bit if I limit voltage to 1.3000 manually why again does it boost the voltage to 1.47xx ?
Is it because PBO enabled?
So this voltage is completely not harmful to 3900X?

 

[drea.drechsler]

Yeaah I agree.
Bit if I limit voltage to 1.3000 manually why again does it boost the voltage to 1.47xx ?
Is it because PBO enabled?
So this voltage is completely not harmful to 3900X?

If you've set a manual fixed voltage at 1.3 it shouldn't do that. What BIOS setting are you using to do it with?

1.47V is unharmful ONLY if the CPU's boost algorithm chooses it. The algorithm is monitoring hundreds of sensors and knows when it's safe to raise voltage and boost clocks. IF you ever set that voltage manually (to keep the processor stable at the same clocks the CPU boosts to) then you're definitely doing the CPU harm as the CPU can then never lower voltage to safe levels (1.20-1.25V) when it gets really hot.

 

[frozensun]

I put it to render round 1h 4K video to mp4 H.264 and it did with max temp of 75C.
I suppose that is quite normal.
But why it won't turbo boost to 4.2 GHz on all cores,only 3.8?

 

[drea.drechsler]

I put it to render round 1h 4K video to mp4 H.264 and it did with max temp of 75C.
I suppose that is quite normal.
But why it won't turbo boost to 4.2 GHz on all cores,only 3.8?

75C is a perfectly good temp for rendering on a 3900X.

If it's not boosting it must be because of BIOS setup. So first thing to do would be a CMOS reset to clean up BIOS. Also be sure to install latest chipset drivers and get them only from the AMD web support web site.

Once CMOS is reset go into BIOS and enable CPPC and CPPC preferred cores, and enable CoolnQuiet and Global C States. Leave core frequency and voltage on AUTO.

Reboot and then go back into BIOS and set XMP so your memory works right. Sometimes you have to manually set the right DRAM voltage that's listed with your memory.

That should help the CPU boost eagerly. I'd uninstall RyzenMaster as it's not really all that helpful if not enthusiast level overclocking, if just using it for a monitoring utility it can hurt performance quite a bit. Instead get HWInfo64.