[SOLVED] Ryzen 3900X temperature too high

[hristohh]

My build consists of a Ryzen 3900X, on a Biostar Racing X570GT8 board. The cooler is a Noctua U12S. The case has 2 120mm intake fans, one 120mm exhaust fan and two more intake 120mm ones at the bottom. Under heavy CPU load the package temperature reported in HWMonitor and Ryzen Master quickly jumps to 95C. The cooler fan ramps up to max speed. However the TMPIN0 remains at 75C which is the temp I've set in the BIOS for 100% PWM. I've noticed that the two readings go up or down togehter and the difference between them is always 19-20C. The CPU does not throttle even at 95C. It stays at 3900MHz, occasionally going to 4000MHz. Even at idle the package temp is around 60C. I remounted the cooler applying thermal paste again but that did not change anything. Since the CPU clocks a bit over its nominal frequency even at 95C, I've been wondering if the package temp reading could be off. From what I've read, Noctua U12S should be sufficent to cool the CPU without overclocking. On the other hand if it is really 95C, I'm slowly damaging the CPU.

I'd appreciate your suggestions and thoughts. Do I believe the package temp reading given all the other info?
Here is a screen capture of the HWMonitor and Core Temp readings. https://1drv.ms/u/s!Ao_bSLUvi9Zk-VqPQQqmLBHcXMdm

 

[cin19]

1. For 3900X, the maximum operating temperature is 95° C.
2. The cooler, Noctua U12S, will support up to 165W TDP CPU. https://noctua.at/en/buying-guide-cpu-coolers

From the HWmonitir, the temp looks like real, and fine, because I read other reviews, the temp with a good air coller should be around 70-80° C. You may try to update the AMD chipset driver, then run both software to see what happens, I think the MB may not have the right heat sensor driver so that the temp reading is not right.

Also you had too many intake fans so that the hot air will stay inside the case. Read the linker for more cooling.

View: https://www.youtube.com/watch?v=sh6F2eccMec

 

[hristohh]

1. For 3900X, the maximum operating temperature is 95° C.
2. The cooler, Noctua U12S, will support up to 165W TDP CPU. https://noctua.at/en/buying-guide-cpu-coolers

From the HWmonitir, the temp looks like real, and fine, because I read other reviews, the temp with a good air coller should be around 70-80° C. You may try to update the AMD chipset driver, then run both software to see what happens, I think the MB may not have the right heat sensor driver so that the temp reading is not right.

Also you had too many intake fans so that the hot air will stay inside the case. Read the linker for more cooling.

View: https://www.youtube.com/watch?v=sh6F2eccMec

Thanks I'll check for driver update. But if the difference stays, which reading should I trust - package or TMPIN0? From what I've read, it should be the package. 70-80C would be ok, but 95C is on the verge of throtling, yet it does not happen. And I forgot to mention that HWMonitor gives a 95C package temp reading as well.

 

[cin19]

Chipset driver https://www.amd.com/en/support

 

[hristohh]

The chipset driver update did nothing. The 20C difference is still there.
https://1drv.ms/u/s!Ao_bSLUvi9Zk-VuE4RR0GYDrgvIp

 

[abhishek2696hhh]

Check you voltages using ryzen master. Also watch this video it'll help you out.

View: https://www.youtube.com/watch?v=Ssuqhyqah2k&t=988s

 

[hristohh]

Check you voltages using ryzen master. Also watch this video it'll help you out.

According to Ryzen Master the core voltage hovers around 1.29-35V at full CPU load. My main concern is the huge difference between the reported package temperature and the TMPIN0. The CPU doesn't behave like it's on the verge of throtling. Even at package temp of 95C the CPU is above its nominal frequency by 100 to 200MHz on all cores. I tend to believe that the package temp readings are off by around 20C but on the other hand, if I am wrong I'm damaging the CPU.

 

[drea.drechsler]

.... which reading should I trust - package or TMPIN0?
....

Neither...or maybe either. It's hard to know for sure with HWMonitor so get rid of it. Instead get HWInfo64.

With HWInfo look at "CPU (Tdie/Tctl)" and "CPU Die (average)". They'll not agree unless under load.

Tdie/Tctl is the instantaneous reading of the hottest of any one of several (dozens?) of on-die sensors scattered around the CPU cores (for a 3900X you should also have a Tdie for each of the two CCD's). It will fluctuate a lot at idle as one core gets a bursting processing load and heats up, then quickly cools back down when it's completed.

As the name suggests, the "CPU Die (average)" reading is a moving average of the last several readings and more closely reflects the actual thermal state of the CPU cores. As processing load picks up the cores are given less and less opportunity to cool down and the two readings will come closer to matching. Ryzenmaster also reports a moving average but must use a different algorithm to calculate it since it doesn't match CPU Die (average), at least on my system, but they're close enough to not really matter.

 

[hristohh]

Neither...or maybe either. It's hard to know for sure with HWMonitor so get rid of it. Instead get HWInfo64.

With HWInfo look at "CPU (Tdie/Tctl)" and "CPU Die (average)". They'll not agree unless under load.

Tdie/Tctl is the instantaneous reading of the hottest of any one of several (dozens?) of on-die sensors scattered around the CPU cores (for a 3900X you should also have a Tdie for each of the two CCD's). It will fluctuate a lot at idle as one core gets a bursting processing load and heats up, then quickly cools back down when it's completed.

As the name suggests, the "CPU Die (average)" reading is a moving average of the last several readings and more closely reflects the actual thermal state of the CPU cores. As processing load picks up the cores are given less and less opportunity to cool down and the two readings will come closer to matching. Ryzenmaster also reports a moving average but must use a different algorithm to calculate it since it doesn't match CPU Die (average), at least on my system, but they're close enough to not really matter.

Both readings in HWInfo show 94C steady after 30 min of full load (fold@home). The CPU happily works just as steady at 3,9GHz - all cores.
For a few seconds package temp in HWMonitor goes over 102C without any CPU throtling.

 

[drea.drechsler]

Both readings in HWInfo show 94C steady after 30 min of full load (fold@home). The CPU happily works just as steady at 3,9GHz - all cores.
For a few seconds package temp in HWMonitor goes over 102C without any CPU throtling.

94C is too hot. I'd try to do something to improve cooling if you run heavy loads on the CPU often. And if serious about Folding, I think you will.

Wise Ryzen owners learn to simply remove HWMonitor, it's just too unreliable with this platform to trust.

 

[hristohh]

94C is too hot. I'd try to do something to improve cooling if you run heavy loads on the CPU often. And if serious about Folding, I think you will.

Wise Ryzen owners learn to simply remove HWMonitor, it's just too unreliable to trust on this platform.

I'm wondering could the CPU really be working at 94C at 100% load for hours above its nominal frequency. I use Fractal Design R5 with two stock 120mm intake fans, 120 mm exhaust fan (all working at full speed now), a Noctua U12S with two fans, and two intake corsair ML120 at the bottom which should be tied to the MB temps. They go from 300rpm to 1500 rpm when the CPU reaches 75/95C. The cooler fans blow towards the rear exhaust fan. The intake fans blow right into the cooler. What more could I do? More exhaust fans?

 

[drea.drechsler]

I'm wondering could the CPU really be working at 94C at 100% load for hours above its nominal frequency. I use Fractal Design R5 with two stock 120mm intake fans, 120 mm exhaust fan (all working at full speed now), a Noctua U12S with two fans, and two intake corsair ML120 at the bottom which should be tied to the MB temps. They go from 300rpm to 1500 rpm when the CPU reaches 75/95C. The cooler fans blow towards the rear exhaust fan. The intake fans blow right into the cooler. What more could I do? More exhaust fans?

I'd trust it to be closest to 'right'. You could also get Ryzenmaster...it's temp readout is an average and it's using AMD's methods to read it.

Not sure what to do...have you confirmed voltages are safe? left in AUTO? are you manually overclocking?

If you leave BOTH VCore and Multiplier settings in AUTO in BIOS then the boosting algorithm will lower both clocks and voltage to keep the processor operating parameters in a safe range. 95C is Tjmax so maybe it's doing what it has to to keep it below that. When at 94C like that, are clocks still at or above the rated base clock for a 3900X (3.8Ghz I think)?

Even a vaunted Noctua can't cool when being fed hot air off of a GPU that's Folding a Cuda workload (if an NVidia). Try opening up the case and put a fan...a room fan...blowing cool air right at the intake to the CPU cooler. If that helps a lot, then more/better case cooling would be in order.

 

[hristohh]

I'd trust it to be closest to 'right'. You could also get Ryzenmaster...it's temp readout is an average and it's using AMD's methods to read it.

Not sure what to do...have you confirmed voltages are safe? left in AUTO? are you manually overclocking?

If you leave BOTH VCore and Multiplier settings in AUTO in BIOS then the boosting algorithm will lower both clocks and voltage to keep the processor operating parameters in a safe range. 95C is Tjmax so maybe it's doing what it has to to keep it below that. When at 94C like that, are clocks still at or above the rated base clock for a 3900X (3.8Ghz I think)?

All settings are at auto. I've just set one of the XMP memory profiles. At 94C the clocks of all cores are between 3875 and 3900 MHz steady, except for some blips of 4000MHz on some of the cores. I've been staring at the screen for 10 minutes now . Ryzenmaster reports 95C, 79% PPT of 142W, 95% of 140A EDC, voltage is at 1.25V, CPU @3900MHz. This is way I tend to think the 95C reading is off. I don't know if it could be because of the MB. BTW it works with AGESA 1.0.0.3ABBA.

 

[drea.drechsler]

All settings are at auto. I've just set one of the XMP memory profiles. at 94C the clocks of all cores are between 3875 and 3900 MHz steady, except for some blips of 4000MHz on some of the cores. I've been stearing at the screen for 10 minutes now . Ryzenmaster reports 95C, 79% PPT of 142W, 95% of 140A EDC, voltage is at 1.25V, CPU @3900MHz. This is way I tend to think the 95C reading is off. I don't know if it could be because of the MB. BTW it works with AGESA 1.0.0.3ABBA.

Good..but still bad cooling. Like I added above, It could be due to the GPU feeding the Noctua hot air when they're both Folding. Open the case up and put a room fan blowing cool air directly at the input to the CPU cooler. If temps drop, then better case ventilation is in order.

Another choice would be to go to a 240mm AIO (at least), front mounted fans as intakes. That's very forgiving of poor airflow inside a case.

Voltage is nice and low (1.25V) but 94C is still way too hot for long-term operation like Folding@Home. It will degrade your CPU and shorten it's life, almost certainly. Got to improve cooling.

 

[hristohh]

Good..but still bad cooling. Like I added above, It could be due to the GPU feeding the Noctua hot air when they're both Folding. Open the case up and put a room fan blowing cool air directly at the input to the CPU cooler. If temps drop, then better case ventilation is in order.

Another choice would be to go to a 240mm AIO (at least), front mounted fans as intakes. That's very forgiving of poor airflow inside a case.

Voltage is nice and low (1.25V) but 94C is still way too hot for long-term operation like Folding@Home. It will degrade your CPU and shorten it's life, almost certainly. Got to improve cooling.

Thanks for the suggestions. I'll make some experiments. But this happen even when only the CPU is loaded and the front fans blow cool air right into the Noctua. Even the idle temps are 40-50C with the CPU doing just the background Windows stuff. Do you think a beefer air cooler will help? I don't like AIO.

 

[drea.drechsler]

Thanks for the suggestions. I'll make some experiments. But this happen even when only the CPU is loaded and the front fans blow cool air right into the Noctua. Even the idle temps are 40-50C with the CPU doing just the background Windows stuff. Do you think a beefer air cooler will help? I don't like AIO.

NHD14 maybe? Dark Rock Pro? maybe. don't really know for sure. an awful lot depends on whether you're moving air through the case well enough since any air cooler only works as well as the ventilation of the case. That's why I like liquid coolers, they truly are forgiving of bad airflow.

 

[hristohh]

NHD14 maybe? Dark Rock Pro? maybe. don't really know for sure. an awful lot depends on whether you're moving air through the case well enough since any air cooler only works as well as the ventilation of the case. That's why I like liquid coolers, they truly are forgiving of bad airflow in the case.

With all the quality fans for intake, I thought the airflow was enough. I'll try the open case and blower thing and I'll report the results.

 

[drea.drechsler]

With all the quality fans for intake, I thought the airflow was enough. I'll try the open case and blower thing and I'll report the results.

I think it's how 'dead air' spaces get created inside a closed space. It gets pretty complicated, engineers spend lifetimes modeling fluid dynamics, but even in the presence of a lot of airflow vortexes and eddy flows can be created that result in hot air just circulating around instead of flowing through.

I've never had luck with cases and arranging fans in all sorts of strange ways and various speeds up to 'tornado'. That is, not until I put a cheap coolermaster 240mm AIO in front of a cheap coolermaster mATX case. This thing folds quietly, both CPU and GPU, with my 3700X keeping to 70C.

 

[hristohh]

I think it's how 'dead air' spaces get created inside a closed space. It gets pretty complicated, engineers spend lifetimes modeling fluid dynamics, but even in the presence of a lot of airflow vortexes and eddy flows can be created that result in hot air just circulating around instead of flowing through.

I've never had luck with cases and arranging fans in all sorts of strange ways and various speeds up to 'tornado'. That is, not until I put a cheap coolermaster 240mm AIO in front of a cheap coolermaster mATX case. This thing folds quietly, both CPU and GPU, with my 3700X keeping to 70C.

Let me report the results of the little experiment I did. Opening the case resulted in the CPU getting closer to 4000MHz instead of 3900MHz. The temp did not change - 95C. The I stuck a handheld fan blowing at full speed at the cooler's intake fan. The air at the handheld fan's intake seemed cool to my hand . I kept the fan blowing right into the Noctua for 5 minutes. 0 degrees change. Still at 95C and the CPU at 4000 MHz.
All of this just makes me think something is wrong with the reading, maybe because of the MB. It's just weird to me!

 

[drea.drechsler]

Let me report the results of the little experiment I did. Opening the case resulted in the CPU getting closer to 4000MHz instead of 3900MHz. The temp did not change - 95C. The I stuck a handheld fan blowing at full speed at the cooler's intake fan. The air at the handheld fan's intake seemed cool to my hand . I kept the fan blowing right into the Noctua for 5 minutes. 0 degrees change. Still at 95C and the CPU at 4000 MHz.
All of this just makes me think something is wrong with the reading, maybe because of the MB. It's just weird to me!

The clock rising up is good! that's the algorithm pushing it harder with cooler air for the Noctua to work with. But pushing it harder means temp's go right back up to the 94/95C TJmax limit where the algorithm has to stop again.

To my mind, that suggests cooling it better will only help!

 

[hristohh]

The clock rising up is good! that's the algorithm pushing it harder with cooler air for the Noctua to work with. But pushing it harder means temp's go right back up to the 94/95C TJmax limit where the algorithm has to stop again.

To my mind, that suggests cooling it better will only help!

Indeed! The question is what better cooling when even a hairdryer style fan stuck into the fins of the cooler can't push the temp down from 95C. If it is really 95C. Opening the case I've let the hot air go out. I've provided copious amounts of cool air to the cooler. It is still 95C. BTW I looked at the fan curves in the BIOS. They react to the TMPIN0, not the package. What's even funnier, I stopped folding, the package temp went straight down to 35C. The TMPIN0 did not move from 73C and the cooler blows full speed. That has never happened before. The case is still open by the way. Restarting the PC did not help.

 

[drea.drechsler]

... BTW I looked at the fan curves in the BIOS. They react to the TMPIN0...

I have no idea what TMPIN0 or Package is but one of them could be the TDie from the CPU core as reported by HWInfo. You could 'calibrate' against HWInfo to find what it's precisely following.

Package could be a temp sensor located under the CPU if your mobo has one. I'd look for that in a section of HWInfo for your Super IO chip which has more temp readouts. Many of thos could be non-functional and reading randomly.

Most (all?) BIOS's use Tdie for fan control which is very troublesome since it pulses so much at idle. You can't use a fan profile that follows it closely at temps below 65C or so as the fans will zoom and then drop back down all the time.

And as you push more cool air through the cooler it's obviously giving the algorithm room to push the CPU harder and harder but still bumping off TJmax. You probably have to go to the big guns for this thing. But if you CAN get the temp to somewhere in the 80's you should also be rewarded with better performance at high processing levels.

But what should you have expected? It's 12 core/24 threads running AVX -rich routines when Folding. It's going to push some ergs, that's just the nature of it.

 

[hristohh]

I have no idea what TMPIN0 or Package is but one of them could be the TDie from the CPU core as reported by HWInfo. You could 'calibrate' against HWInfo to find what it's precisely following.

Package could be a temp sensor located under the CPU if your mobo has one. I'd look for that in a section of HWInfo for your Super IO chip which has more temp readouts. Many of thos could be non-functional and reading randomly.

Most (all?) BIOS's use Tdie for fan control which is very troublesome since it pulses so much at idle. You can't use a fan profile that follows it closely at temps below 65C or so as the fans will zoom and then drop back down all the time.

And as you push more cool air through the cooler it's obviously giving the algorithm room to push the CPU harder and harder but still bumping off TJmax. You probably have to go to the big guns for this thing. But if you CAN get the temp to somewhere in the 80's you should also be rewarded with better performance at high processing levels.

But what should you have expected? It's 12 core/24 threads running AVX -rich routines when Folding. It's going to push some ergs, that's just the nature of it.

From what I observe, the fan control follows tdie from hwinfo with an offset of about 20c. I have set the curve to 100% at 75c. The fan goes to 100% at tdie of 95c.

 

[drea.drechsler]

From what I observe, the fan control follows tdie from hwinfo with an offset of about 20c. I have set the curve to 100% at 75c. The fan goes to 100% at tdie of 95c.

Noctua does have nice quiet fans so 100% @ 75c probably isn't as scary as it sounds to me! I've never been certain if it's been removed or not but Ryzen used to have a 10 or 20 degree offset from Tdie, to be reported as Tctl temperature. They'd do that for high-powered CPU's, like the 1800X, to assist in cooling somehow. I thought that had been stopped with later BIOS' though. It's not that way with my 3700X.

But one thing has occurred to me. In BIOS in the area for making PBO settings (it's called Advance CPU Settings in mine) you can set a "Max allowed CPU temperature". Normally the max allowed temp is Tjmax, 95c, which is where it's stopping at right now for you. You could set that to 90c, for instance, and it will not let temperature get above 90c. It does it, of course, by limiting clock speed and voltage. I did this out of curiosity once and I found it did work as advertised on my motherboard.

If you set that to 90c when Tdie or Taverage holds at or about 90c when running that same heavy load then you'll have a good idea that it's reporting the 'true' temperature.

 

[hristohh]

Noctua does have nice quiet fans so 100% @ 75c probably isn't as scary as it sounds to me!

But one thing occurred to me. In BIOS in the area for affecting PBO you can set a "max allowed CPU temperature". Normally the max allowed temp is Tjmax, 95c, which is where it's stopping at.

You could set that to 90c, for instance, and it will not let temperature get above 90c instead. It does it, of course, by limiting clock speed and voltage. I did this out of curiosity once and I found it did work as advertised on my motherboard.

If you do this, when Tdie holds at or about 90c when running that same heavy load then you'll have a good idea that it's the 'true' temperature.

Well, that is a great idea. I'll dig in the biostar bios to see if there is an equivalent setting. Thanks for spending time to help me. BTW the BIOS HW Monitor reports the lower of the two readings.