In sensors (and respective file in /sys/class/hwmon), I see only the Tctl temperature coming from k10temp driver. I look at the kernel driver code and I am not immediately sure why, since it seems like this model was added a long time ago (kernel 6.0), and this information seems like it is exposed to Windows users. Do others have Tdie/Tccd temperatures? If so, which kernel/distro are you using?
Thanks!
(Edit: Note utilities like KDE System Monitor fall back to Tctl for “CPU core” temperatures when Tccd is missing, so don’t rely on that :-))
Hmm, did this get buried somehow by the forum software? I don’t want to be That Guy who impatiently bumps their own topics, but I didn’t expect it would take days to get one other person to run sensors and say yes/no. Any distribution is fine, my kingdom for a fellow Linux FW16 user’s feedback here!
Great, thank you both for checking! Seems odd since it on the surface it looks like it is supposed to be reading this data. @Mario_Limonciello is it intended that no Tccd is available? I ask specifically only because your name appears as author or reviewer on most of the recent commits. Thanks!
So is this a CPU bug where the valid bit is not being set, and some Windows driver ignores that and will use the value anyway, and it happens to be valid? Otherwise, how are Windows users seeing values? I wish I were in a better position to probe directly by rebuilding driver with some instrumentation to just check myself but I am not right now, for unrelated reasons. Thanks for your quick thoughts!
HWiNFO is not open source so can’t check directly but guess they are doing some thing similar to some other OSS utility called LibreHardwareMonitor that seems to take its own approach of reading the low 12 bits for a few specific models (not 74h), then if it looks like the value is in range they use the value. Gutsy. I guess HWiNFO is doing something similar but since this was mostly a “hey, is this expected?” query, I don’t feel especially motivated to ask them since I think an AMD engineer with direct knowledge would know best. Interesting that those utilities are seeing plausible-looking temperature values anyway. I hope the folks in that other thread are not wasting time chasing phantoms!
I did at least spend a very exciting dinner time looking for a programmer’s guide or open-source register reference for this part of the system and was not successful. I only found OSRR for 17h which only describes base register 0x00059800, which is too bad. (I guess this is to say: I’m currently unsure how external developers would know there is a valid bit since I didn’t succeed at finding the docs about it.)
Thanks for the feedback and the work you do to make everything work so nicely on Linux
I assume you have access to internal datasheet to verify it is the case or not but just wanted to make sure there was not just some assumption based on empirical observation and inference that no longer holds due to changes in undocumented registers, and HWiNFO is actually right since it is returning plausible values.
don’t get too excited with CCD…
It is not per core temp but per “complexe”.
On AMD Ryzen 9 5950X 16-Core Processor wiht 2-CCD / 16 Core (and a IO chip) the report is “correct”
k10temp-pci-00c3
Adapter: PCI adapter
Tctl: +48.1°C
Tccd1: +42.8°C
Tccd2: +29.5°C
On 7N40HS, there is only 1 CCD (with GPU+IO) so it can make sens not have CCD reported.
Per core temps is an other story… can be great to have it be not sure if we can realy.
I remamber (but whit age my memory is bad… so may be wrong) that is have been per core temps in some “old” kernel but it as be removed for unreliability. What I can remamber if it is hardware related (in some/all CPU), or if without clear spec it was to hard to report corect value with correct bias/factor…
Oops! That is certainly my mistake for not understanding well enough what these were measuring (of course, once you point it out, it is obvious ). I guess the question of whether valid bit is still true remains (5700X with one CCX still shows Tccd1), but I guess the real issue is why there is no core temperature implementation. Without public documentation to eventually write it myself I’m not sure what else to do other than politely ask for AMD to put out those docs or send a patch since I don’t have the resources to do more
). I guess the question of whether valid bit is still true remains (5700X with one CCX still shows Tccd1), but I guess the real issue is why there is no core temperature implementation. Without public documentation to eventually write it myself I’m not sure what else to do other than politely ask for AMD to put out those docs or send a patch since I don’t have the resources to do more 