Originally Posted by Timur Born
Thanks for the answer, but this does not even remotely fit the readings in HWinfo. That is unless the "CPU socket thermistor" is different to the "CPU (Socket)" reading in HWinfo, then we would need a proper readout of that.
If "CPU (Socket)" is indeed the "CPU socket thermistor" then it seems that Tctl+5 is used up to about 60C and then some averaging algorithm is used, unfortunately with a not so clear weighting (but mostly weighted for Tctl).
Sorry, but I am not convinced. I repeatedly reproduced that "Sense Skew" is the deciding factor between the CPU safety measures happening or not happening.Sense Skew disabled
: Soft throttling (down to about x30) happens at Tctl + Offset = 95C. Hard throttling (x0.5!!!) happens at Tctl = 95C. Thermal shutdown happens at about Tctl = 115C.Sense Skew enabled
: Soft throttling happens with certain loads regardless of Tctl or at very low Tctl (if CPU at stock clock). Tctl + Offset = 95C or Tctl = 95C cannot be reached at stock clocks/voltages, as a consequence no throttling happens. Tctl = 115C cannot be reached at stock clocks/voltages, as a consequence no thermal shutdown happens!
Instead the CPU crashes to Code 8 when these high temperatures happen.
In the following screenshot Tctl is reported as 66.3C, but the real CPU temperature is likely higher than 110C!
Shortly after taking this screenshot I got a Code 8.
And here is Tctl at over 110C for the same load, same (disabled) cooling and especially same "CPU (Socket)" temperature without Sense Skew. Shortly after taking this screenshot I got a thermal shutdown (last reading was Tctl = 113C, SIO = 109C, couldn't save last screenshot before shutdown).
More important: Disabling Sense Skew leads to proper throttling and thermal shutdowns without producing a Code 8
. Again, I reproduced this repeatedly!
This is reassuring to know. Nevertheless I would prefer if CPU safety measures would work, especially proper thermal throttling.
For the time being I disable Sense Skew and hope that Asus will change fan control to allow more than 75C and HWinfo (Martin) can come up with some logic that makes "Tdie" follow the dynamically changing Tctl offset (+0/+10/+20) and thus better compensate for it in its readings. The whole offset thing is a rather inelegant way of handling certain CPU instructions' load.
Anyone using Sense Skew with the 1800X should watch their "CPU (Socket)" temps, if they climb over 60-70C then your CPU likely is burning hot towards 100C already, regardless of what Tctl and SIO CPU claim.