Ryzen 9 3900x Voltage and Temperature

[yentlvdd]

if some of u guys still have these problems, this is what i did :

i use msi x570 gaming plus motherboard and NZXT Kraken X53 aio cooler

i updated the bios to the latest fw, after that i went into dragon center and updated the cpu chipset drivers and... bam solved the problem.

idle temp : 34°c -38°c
idle voltage : 1.104 V

 

[phill]

I've set mine to 1.00vcore and it's stuck at 4.20GHz across all cores.. Being stressed with WCG/Boinc on all 24 threads.. It's a beast of a CPU and I still wish I could have ordered a 3950X instead...

I digress.. I did a load of testing, testing this very thing not too long ago.. Before I updated my bios, I was able to get to 0.900vcore and that didn't used to crash.. Not anything under 1.00vcore the board doesn't seem to like (X570 Crosshair 8 Hero)

From auto vcore (around 1.45v) and that was hitting 90C in around 1 minute of running WCG taking roughly 280w of power for my whole system, to now, taking just 160 to 165w for my whole system under full load, I'm very impressed....

 

[Leonoid007]

Hi, I am a bit confused by all the values posted here. I have these and they are much higher. However I have 4725 Mhz boost on one core and 4.7 on second best (Ryzen 9 3950x). If I lower my CPU voltage system crashes under load or even do not boot at all. MB is ASUS TUF 570+ wi-fi. CPU voltage set Auto - 0.0625 (the best I can do). If I just leave it on auto (no offset, my best core boosts at 4775 Mhz. I decided to let it go at a bit less stress and use offset -0.0625

So, my concerns are: voltages being around 1.44v max (core) , CPU package temps. 50-60 idle in hwm (around 47 in Ryzen Master), 86 cinebench test load.
And individual cores show 1.5v max ???

Do I have a problem? Cooler is Noctua-U12A.

Thanks!

 

[$ReaPeR$]

Hi, I am a bit confused by all the values posted here. I have these and they are much higher. However I have 4725 Mhz boost on one core and 4.7 on second best (Ryzen 9 3950x). If I lower my CPU voltage system crashes under load or even do not boot at all. MB is ASUS TUF 570+ wi-fi. CPU voltage set Auto - 0.0625 (the best I can do). If I just leave it on auto (no offset, my best core boosts at 4775 Mhz. I decided to let it go at a bit less stress and use offset -0.0625

View attachment 160762
View attachment 160763

View attachment 160764


So, my concerns are: voltages being around 1.44v max (core) , CPU package temps. 50-60 idle in hwm (around 47 in Ryzen Master), 86 cinebench test load.
And individual cores show 1.5v max ???

Do I have a problem? Cooler is Noctua-U12A.

Thanks!

You're fine. What are you worried about?

 

[northvisit]

Hi, I am a bit confused by all the values posted here. I have these and they are much higher. However I have 4725 Mhz boost on one core and 4.7 on second best (Ryzen 9 3950x). If I lower my CPU voltage system crashes under load or even do not boot at all. MB is ASUS TUF 570+ wi-fi. CPU voltage set Auto - 0.0625 (the best I can do). If I just leave it on auto (no offset, my best core boosts at 4775 Mhz. I decided to let it go at a bit less stress and use offset -0.0625

View attachment 160762
View attachment 160763

View attachment 160764


So, my concerns are: voltages being around 1.44v max (core) , CPU package temps. 50-60 idle in hwm (around 47 in Ryzen Master), 86 cinebench test load.
And individual cores show 1.5v max ???

Do I have a problem? Cooler is Noctua-U12A.

Thanks!

VCore (SVI2 TFN) voltage is the correct to look when you have set VCore negative offset (0,0625V). VID value is requested voltage and when it is about 1,5V when the real Vcore = 1,5V - offset (0,0625V) (about). Temps are a bit high but maybe just fine for 3950 depending on load and cooling. When using negative offset you have to check also that it does not affect on performance. Run some program like CB20 without any background programs/processes to see if scores stay about the same with and without applied offset voltage. If negative offset is over some limit the CPU could drop real frequency (clock stretching) without showing it in shown clocks in hwinfo. Performance drop would be clearly noticeable. My CPU/MB seem to work up to about -0.100V before perf drops so I use -0,075V offset.

 

[Leonoid007]

You're fine. What are you worried about?

Temps are on high end... Of course prime95 will heat up to 95C without any problems and stay there so I do not bother. But cinebench at 86C is my concern.
For the cooling I use silent profile, so it runs very quite idle and only goes full speed cooling at about 70C.
Thanks, so I guess I am fine. I will lower offset to -0.05v because after that CB20 shows a bit of performance loss at 0.0625v My best CB20 score is 9732 and total cpu package power at about 211WT

Is this OK for high end air cooling? I did not want water for 2 reasons - one more part to fail (pump) and noise at idle. I also use computer for work so when I am typing in code it does not bother me with noise and C++ compilation time is 2.5-3 times faster than my work Intel computer.

 

[$ReaPeR$]

Temps are on high end... Of course prime95 will heat up to 95C without any problems and stay there so I do not bother. But cinebench at 86C is my concern.
For the cooling I use silent profile, so it runs very quite idle and only goes full speed cooling at about 70C.
Thanks, so I guess I am fine. I will lower offset to -0.05v because after that CB20 shows a bit of performance loss at 0.0625v My best CB20 score is 9732 and total cpu package power at about 211WT

Is this OK for high end air cooling? I did not want water for 2 reasons - one more part to fail (pump) and noise at idle. I also use computer for work so when I am typing in code it does not bother me with noise and C++ compilation time is 2.5-3 times faster than my work Intel computer.

i have a 3900x and my temps are very similar although i use a corsair h115i pro with a room temp of 21C. tbh you made the better choice with the noctua cooler, i never liked water cooling because of the extra things that can go wrong and the fact that unless you go "all in" with custom loop the benefit is negligible. so, i think your temps are ok although if you continue playing with voltages you could lower them even further, it just takes time and a lot of trial and error.

 

[Leonoid007]

i have a 3900x and my temps are very similar although i use a corsair h115i pro with a room temp of 21C. tbh you made the better choice with the noctua cooler, i never liked water cooling because of the extra things that can go wrong and the fact that unless you go "all in" with custom loop the benefit is negligible. so, i think your temps are ok although if you continue playing with voltages you could lower them even further, it just takes time and a lot of trial and error.

System crashes when I do -0.1v and unstable at -0.09ish anything more than 0.05 reduces CB20 score.

BTW my home temperature set to 23.5C not sure about the room with computer, I think it 's a bit cooler but sure above +22C.

 

[thesmokingman]

I undestand the observer effect on TPU article, but why is it the Temp is so high. I mean if it in idle the Temp should be around 35-45 Cel. When it idle it sitting around 45-55 cel.
I am only running one instance monitor tool. Not opening up 2-3 tools to monitor.


Edit:
ALso, when i use or turn on Icue software from Corsair. The voltage will be at contanst 1.35- 1.37V when idol . When i turn it off the idol will drop down to .885v.

View attachment 126707View attachment 126708

That's because it causes observer effects. Icue is hyper aggressive with its polling. Read the article again, it's polling the cpu thus keeping it from idling and allowing cores to sleep.

 

[$ReaPeR$]

System crashes when I do -0.1v and unstable at -0.09ish anything more than 0.05 reduces CB20 score.

BTW my home temperature set to 23.5C not sure about the room with computer, I think it 's a bit cooler but sure above +22C.

cool, if thats the case then i think you are ok with the current settings. out of curiosity, whats your vrm temps?

 

[Leonoid007]

I don't know which sensor is VRM temps, here is the screenshot from hwinfo

cool, if thats the case then i think you are ok with the current settings. out of curiosity, whats your vrm temps?

 

[thesmokingman]

I don't know which sensor is VRM temps, here is the screenshot from hwinfo

View attachment 160889

If you don't see vrm temps in bios that means it doesn't have the sensor. Not a whole lot of boards actually have vrm temp sensors that we can use.

 

[A Computer Guy]

Temps are on high end... Of course prime95 will heat up to 95C without any problems and stay there so I do not bother. But cinebench at 86C is my concern.
For the cooling I use silent profile, so it runs very quite idle and only goes full speed cooling at about 70C.
Thanks, so I guess I am fine. I will lower offset to -0.05v because after that CB20 shows a bit of performance loss at 0.0625v My best CB20 score is 9732 and total cpu package power at about 211WT

Is this OK for high end air cooling? I did not want water for 2 reasons - one more part to fail (pump) and noise at idle. I also use computer for work so when I am typing in code it does not bother me with noise and C++ compilation time is 2.5-3 times faster than my work Intel computer.

I was surprised your CPU package power is so high I thought the default was 142W. Do you have PBO enabled?

I'm running water (including water cooled vrm) but just for comparison...

On water my 3950x will typically bounce between 38c to 48c at idle (vrm about 45c idle) and running CB20 it will bump up to CPU 75c, VRM 72c (80c/80c after multiple runs, pegging PPT, TDC, and EDC to 100% default values) My my case intake temp is 29c. I can peak cold start out CB20 score of about 9400's. Running prime 95 will have 65c to 70c temps with only EDC pegged at 100%.

I had been looking into Power Reporting Deviation after watching GN video (

) and I did some investigation on my motherboard options since I was getting a lower than 95% PRD when running CB20 (I think was around 70%). I don't recall what the option was called off hand but my MB allowed configuration of PPT, TDC, and EDC based on CPU, Motherboard, Custom, Disabled. Also PBO had options to the effect of Disabled, Basic, and Enhanced. I set both to disabled in BIOS and got some increase in PRD up to 85%, better temps and now kept my CPU Package Power (SMU) at 142W default. (Ryzen Master now reports PPT 98% in CB20)

 

[Leonoid007]

Yes I have PBO enabled.

Prime95 has many options. 2 first options will heat cpu the most and in my opinion it's too much stress on CPU which will not happen in real usage.
Also my system uses EDC bug to disable throttling on that parameter which gives a bit more performance but also drives temps and watts very high.

CPU going at 211WT uses twice more power than TDP ... Ouch...

I reset everything to default but DOCP profile enabled. And CB20 gives me about 9000. Temps are much better and I hit TDC and EDC limits in CB20 (95A and 140A).
So without PBO I have about 6% less score but much better temps and low noise from computer.

I was surprised your CPU package power is so high I thought the default was 142W. Do you have PBO enabled?

I'm running water (including water cooled vrm) but just for comparison...

On water my 3950x will typically bounce between 38c to 48c at idle (vrm about 45c idle) and running CB20 it will bump up to CPU 75c, VRM 72c (80c/80c after multiple runs, pegging PPT, TDC, and EDC to 100% default values) My my case intake temp is 29c. I can peak cold start out CB20 score of about 9400's. Running prime 95 will have 65c to 70c temps with only EDC pegged at 100%.

I had been looking into Power Reporting Deviation after watching GN video (

) and I did some investigation on my motherboard options since I was getting a lower than 95% PRD when running CB20 (I think was around 70%). I don't recall what the option was called off hand but my MB allowed configuration of PPT, TDC, and EDC based on CPU, Motherboard, Custom, Disabled. Also PBO had options to the effect of Disabled, Basic, and Enhanced. I set both to disabled in BIOS and got some increase in PRD up to 85%, better temps and now kept my CPU Package Power (SMU) at 142W default. (Ryzen Master now reports PPT 98% in CB20)

 

[A Computer Guy]

Prime95 has many options. 2 first options will heat cpu the most and in my opinion it's too much stress on CPU which will not happen in real usage.
Also my system uses EDC bug to disable throttling on that parameter which gives a bit more performance but also drives temps and watts very high.

CPU going at 211WT uses twice more power than TDP ... Ouch...

Interestingly I was expecting Prime95 to give me higher temps. I could have sworn I got much higher temps when testing weeks ago but I didn't get anywhere near that today for some reason.

 

[$ReaPeR$]

its similar to this.

I don't know which sensor is VRM temps, here is the screenshot from hwinfo

View attachment 160889

its similar to this.

 

[Leonoid007]

Looks like my MB does not report VRM temps... It has 12+2 Dr.Mos.. I guess this is quite good for power needed for 3950x and should not be a problem.

its similar to this.


View attachment 160916
its similar to this.

I think temp9 is the closest to VRM temps with

37 30 39 38

Does it make sense?

its similar to this.


View attachment 160916
its similar to this.

 

[thesmokingman]

Looks like my MB does not report VRM temps... It has 12+2 Dr.Mos.. I guess this is quite good for power needed for 3950x and should not be a problem.



I think temp9 is the closest to VRM temps with

37 30 39 38

Does it make sense?

It doesn't have vrm temp sensors at all.

 

[$ReaPeR$]

Looks like my MB does not report VRM temps... It has 12+2 Dr.Mos.. I guess this is quite good for power needed for 3950x and should not be a problem.



I think temp9 is the closest to VRM temps with

37 30 39 38

Does it make sense?

what @thesmokingman said.

 

[Leonoid007]

Is there any concern with vrm temps?
12+2 Dr.Mos drivers will give at least 350Amps - enough to fry CPU

what @thesmokingman said.

 

[$ReaPeR$]

Is there any concern with vrm temps?
12+2 Dr.Mos drivers will give at least 350Amps - enough to fry CPU

not really in your case, its just nice to have that info.

 

[Max(IT)]

That voltage is WAY too high! For first gen Ryzen 1.35 to 1.4 was common, but these new gen Ryzen CPU's should only be at 1.2 or 1.25.

See this and start at 3:32;

Jay will walk you through what to set and why.

Great great video !
I just assembled my Ryzen 3900X with an Asus ROG STrix B550-F and I was wondering why idle temperatures were so high (about 50°).
I noticed CPU Core Voltage was set at 1.440 V by default in the BIOS

Since I’m not an expert of AMD CPUs (my last was an Athlon 64) I wasn’t sure about the value.
After seeing the video I set an offset of -0.125 V and idle temperature is 41° !!!
Much better.
During Cinebench R20 previously I was reaching 1.48 V and 85°, with a good cooler like a Dark Rock 4, now it tops at 1.37 V and 76°.
Maybe I could try even a bigger offset...

 

[lexluthermiester]

After seeing the video I set an offset of -0.125 V

Maybe I could try even a bigger offset...

I'm betting you can get away with -0.25 or even -0.35 and still stay stable.

 

[Max(IT)]

I'm betting you can get away with -0.25 or even -0.35 and still stay stable.

Most probably but what about Turbo speed ?
dont you think it could be reduced if I lower the voltage ?

 

[lexluthermiester]

Most probably but what about Turbo speed ?
dont you think it could be reduced if I lower the voltage ?

The stock turbo speeds should not be affected.