Re: ThreadRipper 1950X context: example kldload amdtemp related temperature reports for idle conditions?

On Oct 22, 2023, at 14:33, Mark Millard <marklmi@yahoo.com> wrote:

> Is anyone familiar with what what they normally see on an idle
> ThreadRipper 1950X via the likes of (presumes kldload amdtemp
> had already been done):
> 
> # sysctl dev.amdtemp | grep "[0-9]\.[0-9]C$"
> 
> (Approximate ambient air temperature for the context would be a
> bonus.)
> 
> I do not have a history of monitoring this, unfortunately.

What I see for idle vs all hardware threads fully busy
(buildworld -j32 's llvm lib* build activity):

fully busy: about 95 degC reported
idle:       low 60's degC reported, say 62 degC:
(varies more than the 95 figure does)

# sysctl dev.amdtemp | grep "[0-9]\.[0-9]C$"
dev.amdtemp.1.core0.sensor0: 62.1C
dev.amdtemp.0.ccd0: 62.5C
dev.amdtemp.0.core0.sensor0: 62.1C

AMD's published 1950X maximum operating Tjmax is 68 degC.
The max reported figure's delta vs. documented maximum:
about 95 degC - 68 degC = 27 degC

Various notes on the web report that in order for the
tCTL used for fan control to have about the same maximum
across various machines, ThreadRipper 1950X's use:

tCTL = operatingTjmax + 27 degC

So, it looks like the dev.amdtemp reported values are
tCTL-like for units, so biased toward fan control use.

Idle shifted similarly down from reported:
62 degC - 27 degC = about 35 degC

Overall: it seems I'm seeing a range for idle through
fully-busy of about (operatingTjmax):

35 degC .. 68 degC

but reported as more like (tCTL):

62 degC .. 95 degC


===
Mark Millard
marklmi at yahoo.com