PowerMac G5 quad-core, CPU A1 DIODE TEMP: 90.8 C (for example): How to handle? [Mac OS X behavior]

[Mark Millard]

In addition to the below notes that I sent to Justin and Nathan I installed the Processors system preference pane (from CHUD 4.6.2) and did an experiment with turning off and on "Allow Nap" under Mac OS X 10.5.8 on the problematical G5 quad-core. I'll note that the whole time Mac OS X was keeping the RPM's near maximums so those are not much of a parameter set for this note.

With Allow Nap disabled the result for idle-conditions (well, Temperature Monitor running and the configuration's normal background activity going) was: the temperature gradually increased overall over 2hr+ and then started the temperature oscillation pattern once it reached something around around 88C (for the same core as before). No benchmarks or other loads added.

Reenabling Allow Nap lead to the temperature dropping and staying down (compared to reaching 88C). (This matches earlier results were Allow Nap was implicitly enabled and so idle-conditions did not get to such high temperatures.)

Earlier activity reports establish that Allow Nap is not sufficient to keep the processors/cores cool when they are under load (with the high RPM's always involved).

===
Mark Millard
markmi at dsl-only.net

On 2015-Jan-17, at 10:56 PM, Mark Millard <markmi at dsl-only.net> wrote:

Hi. [The mail list will not get the pictures Justin and Nathan get with this.]

Mac OS X 10.5.8 has the pump and fans going nearly full tilt at idle once the cores have been warmed up for the problematical G5 quad core:

  CPU A Core 1 56°C
  CPU A Core 2 65°C
  CPU B Core 1 44°C
  CPU B Core 2 44°C
  CPU A Intake 3104rpm
  CPU B Intake 3104rpm
 CPU A Exhaust 3200rpm
 CPU B Exhaust 3200rpm
    CPU A Pump 3600rpm

And Mac OS X 10.5.8 does not prevent the temperature rise. Mac OS X just does more to cool things down temporarily once critical temperatures happen, trading performance silently by having some idle time. Having FreeBSD be more like the above for the lower temperatures would still overheat before "make -j 8 buildworld buildkernel" would finish. (I think Justin's various RPM changes are great improvements to PowerPC FreeBSD but the problematical G5 quad core is just too defective to reasonably work around: I doubt that FreeBSD would want the silent performance tradeoff the later plots for Mac OS X indicate.)

By contrast the good G5 quad core is more like what one would expect when booted from the exact same Firewire 800 Mac OS X 10.5.8 disk (and warmed up):

  CPU A Core 1 42°C
  CPU A Core 2 43°C
  CPU B Core 1 46°C
  CPU B Core 2 48°C
  CPU A Intake  970rpm
  CPU B Intake  970rpm
 CPU A Exhaust 1000rpm
 CPU B Exhaust 1000rpm
    CPU A Pump 1250rpm

[Past this point in these notes is likely of no interest from the mail list but may be for Justin and possibly even for Nathan.]

The later picture for (A) below shows the high temperatures reached for a good G5 quad core context.

I've collected Temperature Monitor data from Mac OS X for both a good G5 quad core and for the problematical one (for running 6 variants of the HINT benchmark at once). The information is in the form of screen shots of its temperature graphs. I've included:

A) GoodG5QuadCoreTemps6HINTs.jpg to show the stages (0) warmup, (1) 6 variants of HINT going at once, (2) cool down after stopping those 6 variants for the good G5 quad core. (Each HINT has some input that I type so very beginning of the start up is not smooth: idle for a bit after being busy for a bit.)

B) BadG5QuadCoreTemps6HINTs.jpg for the same warmup to cool down coverage but for the bad G5 quad core. But the high-temperature time frame is not sampled fast enough to show its structure correctly. There are lots of other differences compared to (A).

For the bad G5 quad core I've also included some short-term graphs that sampled fast enough to give some structure for the high temperature time frame...

C) BadG5QuadCoreTemps6HINTsStart.jpg showing the starting of the 6 HINTs and just after.

D) BadG5QuadCoreTemps6HINTsBeforeStop.jpg showing how the frequency and structure of the oscillations had changed by the time I was going to stop the 6 HINTs. This shows that some of the oscillation structure that I reported earlier eventually goes away and that the oscillation gets faster than what it starts out as --something I'd not reported before.

E) BadG5QuadCoreTemps6HINTsCooldown.jpg showing the cool down detail after I stopped the 6 HINTs. (Compare the beginning of this cooling with the oscillation cooling?)

From all this I get that the problematical G5 quad core is definitely defective but Mac OS X 10.5.8 goes out of its way to keep operating anyway, trading off other things to make that happen. Other contexts might not want those same tradeoffs.

===
Mark Millard
markmi at dsl-only.net

On 2015-Jan-17, at 09:36 PM, Justin Hibbits <chmeeedalf at gmail.com> wrote:

I noticed on my quad running OS X spins up the fans all the way down at 55C. Perhaps you can try that, set the initial multiplier to 300 or more.  I can add a sysctl to change the multiplier, if that would help people.

-Justin

<GoodG5QuadCoreTemps6HINTs.jpeg><BadG5QuadCoreTemps6HINTs.jpeg><BadG5QuadCoreTemps6HINTsStart.jpeg><BadG5QuadCoreTemps6HINTsBeforeStop.jpeg><BadG5QuadCoreTemps6HINTsCooldown.jpeg>