[SOLVED] Are CPU temps acceptable?

BreakfastTea

Distinguished
Feb 22, 2015
7
0
18,510
Just upgraded pretty much everything on my pc. The CPU is a Ryzen 9 3900xt and is running around 48-55 (mainly 55) C at idle and around 85 C during a stress test/ full load. Is this acceptable? I struggled putting the CPU cooler on and may have smushed some of the thermal paste so should I think about reseating it?

Thanks in advance.
 
Solution
I just used the CPU-Z stress test and was using HW to monitor it at the same time. I think I can stress it even more with the Prime95 program but not sure how necessary it is.
Reporting temps on modern high-performance CPU's is becoming a bit of a challenge. They're all tuned to extract as much as the silicon can dish out. They're also temperature seeking; so will continue boosting as long as temperatures are OK.

But Ryzen is especially sensitive because of it's boost algorithm that constantly dithers the boost clocks across multiple cores, boosting one at a time as long as it has temperature margin. It's doing this up to 100 times a second, very fast. It has to work this way because the very small 7nm geometry leaves very...

Karadjgne

Titan
Ambassador
Use HWInfo64 (sensors only) and run Prime95 Small fft (not smallest, AVX technologies disabled) . That's a solid 100% load on all cores , doesn't bounce around. Look at the individual core temps, not cpu as a whole. Temps should be within 10°C of hottest-coldest at worst. Any core more than that and I'd say your paste job is shot.

With a Ryzen, spread method is best, a thin and uniform coat from edge to edge, guarantees coverage and no airpockets.

Ryzen idles run higher than Intel, because the cpu at idle works differently, but software still reports only the hottest core as a single temp reading.

I'd be on the web looking up "how to's" on undervolting your cpu.

Otherwise yes, temps are very normal.
 
  • Like
Reactions: BreakfastTea

BreakfastTea

Distinguished
Feb 22, 2015
7
0
18,510
Depends a bit on what you're using for a 'full load' stress test, but yes it's reasonable for a 3900X.

A tip: get HWInfo64 and monitor the CPU die(average) temperature for the more representative state of CPU thermals.

I just used the CPU-Z stress test and was using HW to monitor it at the same time. I think I can stress it even more with the Prime95 program but not sure how necessary it is.
 

BreakfastTea

Distinguished
Feb 22, 2015
7
0
18,510
Use HWInfo64 (sensors only) and run Prime95 Small fft (not smallest, AVX technologies disabled) . That's a solid 100% load on all cores , doesn't bounce around. Look at the individual core temps, not cpu as a whole. Temps should be within 10°C of hottest-coldest at worst. Any core more than that and I'd say your paste job is shot.

With a Ryzen, spread method is best, a thin and uniform coat from edge to edge, guarantees coverage and no airpockets.

Ryzen idles run higher than Intel, because the cpu at idle works differently, but software still reports only the hottest core as a single temp reading.

I'd be on the web looking up "how to's" on undervolting your cpu.

Otherwise yes, temps are very normal.

Thanks, just did this and got 75 C average and max range of 8C, so looks like everything is OK. Not sure how the temp is lower than before (I did this same test without disabling AVX cores and got up to 95C, but that was before I updated the chipset drivers so maybe that was it). Should be fine as is then looks like. I will look into undervolting too.
 
I just used the CPU-Z stress test and was using HW to monitor it at the same time. I think I can stress it even more with the Prime95 program but not sure how necessary it is.
Reporting temps on modern high-performance CPU's is becoming a bit of a challenge. They're all tuned to extract as much as the silicon can dish out. They're also temperature seeking; so will continue boosting as long as temperatures are OK.

But Ryzen is especially sensitive because of it's boost algorithm that constantly dithers the boost clocks across multiple cores, boosting one at a time as long as it has temperature margin. It's doing this up to 100 times a second, very fast. It has to work this way because the very small 7nm geometry leaves very little surface area to transfer heat across. It also has dozens (hundreds?) of heat sensors and reports the hottest single one at the moment.

That all leads to a lot of temp spiking in one tiny area of the die at a time, each one with relatively little thermal input: like a match in a room. There has been improvements with newer BIOS' but HWInfo's averaging readouts is averaging all the sensors together so you get a better picture of the CPU's true thermal state.

You can run Prime95, expect high temps with small FFT's and if you get them don't fret (too much) because you know it's utterly unreal. Your CPU is safe up to 95C, but naturally you want it to run cooler than that. It will help you determine if you have a weak cooling solution. It's especially valuable for AIO or liquid coolers since they have a long delay before the liquid can thermally saturate. Monitor the average temperature reading across a several hour run and look for a trend line.
 
Last edited:
Solution

mmp09

Commendable
Nov 27, 2021
91
4
1,545
I personally consider anything above 50 C idle CPU temp as not that good when (if) ambient is about 27 - 30C. Is BIOS latest?
Out of curiosity just contact AMD support in your region and ask them if they think 55C idle is normal or not? I was getting 54-55 on my 3500X and when contacted AMD, they were ready to do RMA but later after updating BIOS and using one more fan in cabinet I brought it down to 43 in summer when I built my system (ambient avg 27-28) and later about 6 months down the line to 32 in winter (ambient avg 22-23). Wanted to avoid RMA process and my tryouts worked.