CPU tdie vs CCD tdie

crashkaST

New Member
Hi everyone,

Apologies if this has been discussed before, I can't seem to find a comprehensive answer to this.

I am trying to understand how HWiNFO calculates CPU Tdie and CCD Tdie metrics, and why are they behaving differently on my single-CCD CPU (Ryzen 5 3600).

This has all started when I observed the spiking (seesaw) temperature pattern on my CPU. I'm told it's a common thing in Ryzen due to the nature of power management, core control etc. although I'm not 100% sure what the technical reason for the swings is yet - but I don't want to digress.

What puzzles me at the moment is why the CCD Tdie graph seems flat, while both CPU Tdie and motherboard's CPU temperature readings seesaw

cputemp.PNG

At first I attributed the flatness to the short-term rolling averaging of the CCD Tdie metric. That explanation doesn't make sense to me after observing how CPU transitions from full load to no load - see below image. The CCD Tdie seems to essentially drop down to idle-ish temp instantly, while the CPU Tdie has a pronounced 15 second slope. If the CPU Tdie is actual (non-averaged) hottest point temperature of the CPU then I would expect the CCD Tdie to (at least) have a more pronounced slope, especially as the temperature difference is around 35C between full load and idle.

cputemp3.PNG

Also, I do not understand why CPU Tdie and CPU Die (average) have such different slopes. It almost seems as if that CPU Tdie slope isn't representing actual temperatures but it's a slope calculation, and that only the spikes which start the slope are actual temperature readings. This is just me trying to find an answer - pure speculation.

Why are CPU Tdie and CCD Tdie behaving so differently?

How does HWiNFO calculate CCD Tdie, is it an average of all sensors in the CCD or a specific point in the CCD?

Is there another component in the CPU that could be causing temps across the CPU to drop slower than the CCD itself?

Can this be explained by explaining the seesaw pattern of the CPU itself and I'm just missing something?

Cheers!

EDIT: Could IO Die be making the overall CPU temperatures drop slower than the CCD itself? Is the IO Die responsible for the seesawing of temperatures as well?
 
Last edited:
HWiNFO doesn't calculate any of those values, it displays them exactly how the CPU (its SMU firmware) reports them.
So the behavior might also depend on BIOS version (more precisely AGESA which includes SMU firmware update too).
AMD doesn't disclose how exactly it calculates those values (based on dozens of sensors spread across the entire die), but we're pretty sure that "CPU Die (average)" value is a short-term average of the CPU die only (AMD stated in the past that it should cover x86 cores only). Other values most likely include IO die temperatures (and/or cache) as well and Tctl/Tdie is reporting the instant hottest temperature among sensors without averaging.
 
Thanks for that, it makes much more sense now.

CCD temps seem to be nominal, IO die or something else must be causing my CPU temperatures to seesaw, I'll try to dig around.
 
Back
Top