• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Can anyone explain me how Ryzen CPU temps work?

ZenTimings is just for memory subsystem observation. Sorry I wasn’t clear about that.
SoC voltage can be modified in BIOS\UEFI of course.
I should lower SoC voltage from 1.23 to 1.2? Thanks.

@Zach_01 I set the SoC voltage to 1.2v. Now, it's showing 1.185v in HWiNFO. How to check for instability? Thanks

.
ZenTimings_Screenshot.png
 
Last edited:
I should lower SoC voltage from 1.23 to 1.2? Thanks.

@Zach_01 I set the SoC voltage to 1.2v. Now, it's showing 1.185v in HWiNFO. How to check for instability? Thanks
Same as usual... but dont over run heavy loads too much. Sometimes everyday usage, gaming... can reveal weaknesses too.
Take you time and work your way down to the lowest SoC voltage. 1.185V is pretty good too. But undervolt it to the limit can only hurt stability at some point, and always it's the best for thermals, power consumption and longevity.

So what is the "CPU VDDCR_VDD Voltage (SVI3 TFN)" under load?
For the Ryzen system all you need is HWiNFO64, ZenTimings. Forget everything else.

Untitled_82.png

I mentioned this below earlier and its a must too...

1708413462057.png

If you have any instability, app/game crash or even OS crash/restart, you can check System Event viewer for what kind of error it was (for example WHEA, and what kind of it)
So you can somewhat pin point the culprit.
Also, HWiNFO is displaying if any WHEA errors occured when its running in background

1708414439218.png
 
I have the air conditioner at 20°C. I think 15°C could be accurate. I hate (here it's) summer and it's very humid.
20°C (68°F) is nice and cool for ambient (room) temperatures. No problem there. And air conditioning works, in effect, by "conditioning" the air by removing the humidity as part of the cooling process. That said, unless moisture is dripping into your electronics causing a potential "short" problem, humidity really has no effect on your electronics.

It may make you feel miserable, but your computer does not care.

If anything, there is a significant advantage to high-humidity environments. Static build-up and thus ESD (electro-static discharge) is essentially non-existent. :)
 
Same as usual... but dont over run heavy loads too much. Sometimes everyday usage, gaming... can reveal weaknesses too.
Take you time and work your way down to the lowest SoC voltage. 1.185V is pretty good too. But undervolt it to the limit can only hurt stability at some point, and always it's the best for thermals, power consumption and longevity.

So what is the "CPU VDDCR_VDD Voltage (SVI3 TFN)" under load?
For the Ryzen system all you need is HWiNFO64, ZenTimings. Forget everything else.

View attachment 335485

I mentioned this below earlier and its a must too...

View attachment 335486

If you have any instability, app/game crash or even OS crash/restart, you can check System Event viewer for what kind of error it was (for example WHEA, and what kind of it)
So you can somewhat pin point the culprit.
Also, HWiNFO is displaying if any WHEA errors occured when its running in background

View attachment 335490
Hi, i can't find "Snapshot CPU Polling". In fact, my HWiNFO looks a little bit different. What can i do? Thanks.
 
Snapshot polling hits the cpu harder.. I never use it myself.
 
Snapshot polling hits the cpu harder.. I never use it myself.
Anyway, i can't find that option in HWiNFO.
From what've you read; do you think i'm good to go? Thanks!
 
I ran a 30 minutes stability test of CB R23 with a -45 CO (no crashes).
Just out of curiosity, did you run CB R23 multi-threaded or single-threaded? From my experience, 1T will easily crash on CO settings that are stable running nT.
-45 CO? I thought the max was -30? not sure what's going on here....
AMD allowed the use of CO up to -60 a while ago. IIRC, it requires a fairly current AGESA.
I should lower SoC voltage from 1.23 to 1.2? Thanks.
Lower VSOC allows for better infinity fabric overclocks, but somewhat limits overclocks of the memory clock. At the moment 1.2V SoC seems to be the current default for most AM5 motherboards since it should give enough headroom to avoid exploding CPUs.

Not sure if you are comfortable with manually setting memory timings, but looking at your ZenTimings, you should be able to increase tREFI to something way higher to reduce the time the memory spends on refreshing. Especially since your ambient temps are overall fairly low.
 
Just out of curiosity, did you run CB R23 multi-threaded or single-threaded? From my experience, 1T will easily crash on CO settings that are stable running nT.

AMD allowed the use of CO up to -60 a while ago. IIRC, it requires a fairly current AGESA.

Lower VSOC allows for better infinity fabric overclocks, but somewhat limits overclocks of the memory clock. At the moment 1.2V SoC seems to be the current default for most AM5 motherboards since it should give enough headroom to avoid exploding CPUs.

Not sure if you are comfortable with manually setting memory timings, but looking at your ZenTimings, you should be able to increase tREFI to something way higher to reduce the time the memory spends on refreshing. Especially since your ambient temps are overall fairly low.
Hi. I'm going to run a single-threaded CB R23 (i only run it at multi-threaded). Thanks for the suggestion.
My ASUS board set the VSOC to 1.23v, should i set it at 1.2v? Or is it ok?
And memory wise, i don't wanna mess things up, lol.
 
My ASUS board set the VSOC to 1.23v, should i set it at 1.2v? Or is it ok?
If it's the board's default, it should be fine.
AMD allows up to 1.3V per spec, but especially Asus and Gigabyte tend to be quite careful with that voltage, because both companies had some exploding CPUs on some of their boards that made the rounds on Reddit and YouTube. Also keep in mind that all software voltage readouts tend to be a little bit off (kHz sampling range). Unless you connect an oscilloscope with a high enough resolution (multi MHz sampling range) to your socket, you won't get exact readings.
 
If it's the board's default, it should be fine.
AMD allows up to 1.3V per spec, but especially Asus and Gigabyte tend to be quite careful with that voltage, because both companies had some exploding CPUs on some of their boards that made the rounds on Reddit and YouTube. Also keep in mind that all software voltage readouts tend to be a little bit off (kHz sampling range). Unless you connect an oscilloscope with a high enough resolution (multi MHz sampling range) to your socket, you won't get exact readings.
Oh, ok, thanks. I'm running a 30 minutes single threaded run in CB R23. I'll tell you if any instability occurs.
 
Snapshot polling hits the cpu harder.. I never use it myself.
So you are claiming that it does exactly the opposite of what the author of HWiNFO says it does?

Hi, i can't find "Snapshot CPU Polling". In fact, my HWiNFO looks a little bit different. What can i do? Thanks.
Right click HWiNFO system tray icon >> settings
 
So you are claiming that it does exactly the opposite of what the author of HWiNFO says it does?


Right click HWiNFO system tray icon >> settings
I enabled the "Snapshot CPU Polling". Is it there? HWiNFO said that it could be unavailable.

Untitled.jpg
 
I enabled the "Snapshot CPU Polling". Is it there? HWiNFO said that it could be unavailable.

View attachment 335593
No its says that some per-thread readings are disabled when snapshot is enabled.
Its ok you dont need those per-thread values anyway

Also those blue arrows at the bottom expands the sensors window to more than 1 column.
 
No its says that some per-thread readings are disabled when snapshot is enabled.
Its ok you dont need those per-thread values anyway

Also those blue arrows at the bottom expands the sensors window to more than 1 column.
Yes, i know about the blue arrows. Do you need an expanded view of HWiNFO or is it ok with that?

If it's the board's default, it should be fine.
AMD allows up to 1.3V per spec, but especially Asus and Gigabyte tend to be quite careful with that voltage, because both companies had some exploding CPUs on some of their boards that made the rounds on Reddit and YouTube. Also keep in mind that all software voltage readouts tend to be a little bit off (kHz sampling range). Unless you connect an oscilloscope with a high enough resolution (multi MHz sampling range) to your socket, you won't get exact readings.
No issues with CB R23 single-core run (30 minutes). And as i said before, i did a OCCT (curve optimizer test, long and short), no errors too. I tested AIDA64. I ran a Passmark test, a Geekbench 6 test, i used the PC normally, i let it stay at idle for a period of time, i play some music, some movies, played some games, all ok. I got a maximum temperature of 55°C
 
So you are claiming that it does exactly the opposite of what the author of HWiNFO says it does?
Yup..

Try it out for yourself :cool:
 
Yes, i know about the blue arrows. Do you need an expanded view of HWiNFO or is it ok with that?


No issues with CB R23 single-core run (30 minutes). And as i said before, i did a OCCT (curve optimizer test, long and short), no errors too. I tested AIDA64. I ran a Passmark test, a Geekbench 6 test, i used the PC normally, i let it stay at idle for a period of time, i play some music, some movies, played some games, all ok. I got a maximum temperature of 55°C
No not really, unless you want to ask anything about it. Im happy to explain if I know

Yup..

Try it out for yourself :cool:
I should've ask this earlier...
What do you mean by it hits the CPU harder?
Does it make work higher speed? voltage? usage?
Because I've seen the opposite. It actually makes readings more accurate.
For example previously I had higher effective clocks at idle-middle loads without it, and those clocks contain CPU C-States... so
 
No not really, unless you want to ask anything about it. Im happy to explain if I know


I should've ask this earlier...
What do you mean by it hits the CPU harder?
Does it make work higher speed? voltage? usage?
Because I've seen the opposite. It actually makes readings more accurate.
For example previously I had higher effective clocks at idle-middle loads without it, and those clocks contain CPU C-States... so
Is there a way to make the CPU to work with lower clocks while idling? I find 3400~3600MHz a tad too much. Or maybe that's the way they are supposed to work?
An the last one: what do you think about my CPU temps, voltages, stability (all the tests i did), etc? Thanks very much.
 
No not really, unless you want to ask anything about it. Im happy to explain if I know


I should've ask this earlier...
What do you mean by it hits the CPU harder?
Does it make work higher speed? voltage? usage?
Because I've seen the opposite. It actually makes readings more accurate.
For example previously I had higher effective clocks at idle-middle loads without it, and those clocks contain CPU C-States... so
He means the extra polling rate generally over API uses resources.

So if you benchmark with hwinfo64 running, you would see a drop in scores over an average. If increasing the polling rate reduces scores, then you only use it for base temps and run the benchmark without hwinfo64.

Since hwinfo 64 reads a lot of sensors, it's not always ideal during benchmarks for score accomplishments, maybe use Gpu-Z and read the socket temp only during cpu benchmarks.

[I think this is what Shawn is trying to say]
 
Last edited:
He means the extra polling rate generally over API uses resources.

So if you benchmark with hwinfo64 running, you would see a drop in scores over an average. If increasing the polling rate reduces scores, then you only use it for base temps and run the benchmark without hwinfo64.

Since hwinfo 64 reads a lot of sensors, it's not always ideal during benchmarks for score accomplishments, maybe use Gpu-Z and read the socket temp only during cpu benchmarks.

[I think this is what Shawn is trying to say]
I hear you, but our little debate with @freeagent is not about having HWiNFO on or shut off.
It’s about having “Snapshot CPU polling” on or off when it’s running.
Which the best way IMO for Ryzens is to be enabled.

I find CPU-Z to mess CPU C-States a bit more than this software that monitors so many sensors.

And in my task manager CPU utilization is less than 0.5% when it’s running. Polling interval is 1000ms and Drive polling is set to 5 cycles (SSDs is all I have)
If anyone sets general polling to 2000ms it will be reduced even more.
I’ve tried it with 500ms in the past and it was way too much messing around with C-States.

Also it’s wise to not run any other monitoring software along. Though I do run GPU-Z with it because I want it for my keyboard display.

Is there a way to make the CPU to work with lower clocks while idling? I find 3400~3600MHz a tad too much. Or maybe that's the way they are supposed to work?
A the last one: what do you think about my CPU temps, voltages, stability (all the tests i did), etc? Thanks very much.
Set windows power plan to power saving if not already.
I find your system’ status very nice.
Yes Ryzens are not lowering clocks too much. After a point cores are entering halt (C1) or power down (C6) states. That is the reason of effective clocks existence. They account all C-States in.
Also you have the “CPU C-States Residency” readings which show what is the percentage of all core C-State status on the running time/session of HWiNFO sensors window.

C0: Active state
C1: Halt or sleep state
C6: Power down or deep sleep state
 
Last edited:
I hear you, but our little debate with @freeagent is not about having HWiNFO on or shut off.
It’s about having “Snapshot CPU polling” on or off when it’s running.
Which the best way IMO for Ryzens is to be enabled.

I find CPU-Z to mess CPU C-States a bit more than this software that monitors so many sensors.

And in my task manager CPU utilization is less than 0.5% when it’s running. Polling interval is 1000ms and Drive polling is set to 5 cycles (SSDs is all I have)
If anyone sets general polling to 2000ms it will be reduced even more.
I’ve tried it with 500ms in the past and it was way too much messing around with C-States.

Also it’s wise to not run any other monitoring software along. Though I do run GPU-Z with it because I want it for my keyboard display.


Set windows power plan to power saving if not already.
I find your system’ status very nice.
Yes Ryzens are not lowering clocks too much. After a point cores are entering halt (C1) or power down (C6) states. That is the reason of effective clocks existence. They account all C-States in.
Also you have the “CPU C-States Residency” readings which show what is the percentage of all core C-State status on the running time/session of HWiNFO sensors window.

C0: Active state
C1: Halt or sleep state
C6: Power down or deep sleep state
Task manager is garbage for monitoring anything. Why would this be a consideration to follow for cpu usages? And then how can you be sure it's messing with the C-states at all? It'll be evident task manager isn't giving correct readings.

Then you'd back it up by comparing to cpu-z. But using multiple monitoring creates conflicts, as you've stated. And it's true. It's a matter of finding a software that remains accurate and many do not.

Even Hwinfo64 will skew after a given amount of time even if the system is idle for 12 hours.

Not sure about AMD rigs and snap shot polling because I don't have an AMD rig to experience this issue.

Either way, monitoring shouldn't conflict with C-states. Honestly the first I heard of this. The system would need to be at defaults completely I'd think to test this accurately. What's your experience there?
 
I hear you, but our little debate with @freeagent is not about having HWiNFO on or shut off.
It’s about having “Snapshot CPU polling” on or off when it’s running.
Which the best way IMO for Ryzens is to be enabled.

I find CPU-Z to mess CPU C-States a bit more than this software that monitors so many sensors.

And in my task manager CPU utilization is less than 0.5% when it’s running. Polling interval is 1000ms and Drive polling is set to 5 cycles (SSDs is all I have)
If anyone sets general polling to 2000ms it will be reduced even more.
I’ve tried it with 500ms in the past and it was way too much messing around with C-States.

Also it’s wise to not run any other monitoring software along. Though I do run GPU-Z with it because I want it for my keyboard display.


Set windows power plan to power saving if not already.
I find your system’ status very nice.
Yes Ryzens are not lowering clocks too much. After a point cores are entering halt (C1) or power down (C6) states. That is the reason of effective clocks existence. They account all C-States in.
Also you have the “CPU C-States Residency” readings which show what is the percentage of all core C-State status on the running time/session of HWiNFO sensors window.

C0: Active state
C1: Halt or sleep state
C6: Power down or deep sleep state
Power saving plan goes from idling at 3600MHz to 3060MHz. With that, the CPU reduces a little bit of power consumption (like 5~8W) and temps. Running a CB R23 gives the CPU the same score than before. Thanks.

EDIT: Sorry, i've cuoted you wrongly.
EDIT 2: I've noticed that the CPU isn's as fast as it should be in some tasks, so i select Balanced power plan and in AMD Power Slider, then Overlay, i set it to battery saver and now it's idling at 3133MHz.
 
Last edited:
Back
Top