Spurious thermal shutdowns on Dell Studio 1557
smithi at nimnet.asn.au
Sun Apr 4 13:07:53 UTC 2010
On Sat, 3 Apr 2010, Bartosz Fabianowski wrote:
> Alexandre wrote:
> > I would not override _CRT
> Intel seem not to have documented the maximal operating temperature for
> this CPU (Core i7 Q720M). But overclocking enthusiast forums mention
> automatic throttling from 100°C onward. So while I cannot be sure, I
> think the CPU should survive just fine with a _CRT of more than 85°C.
I think so too, but suspect those overclockers would be talking about
desktop boxes, not laptops? I've seen few laptop _CRT levels at 100C.
Personally I suspect they've just got the _PSV and _CRT backwards (95
vs 85C) and that swapping those should give safe passive and critical
values .. it could never make sense to have _PSV hotter than _CRT; if
they'd intended making _PSV unavailable it should be set -1 like _HOT.
Some things to try anyway, along with Alexandre's good clues re thermal
polling interval etc - and maybe the cheap cooling pad to boot, or even
a more heavy duty one; you likely don't run buildworlds on your lap? :)
> > Something to the tune of:
> > hw.acpi.thermal.tz0.passive_cooling=1 hw.acpi.thermal.user_override=1
> > hw.acpi.thermal.tz0._PSV=75C
> I can do that. But since the CPU is running at ~60°C when completely
Sure, but there still should be a discernable fan speedup around 71C; if
not then your active cooling may not be working right either. A length
of plastic tubing makes a useful 'stethoscope' for listening to fans.
> idle, this will essentially force throttling whenever I do anything that
> demands a bit of CPU. I bought a quad-core i7 specifically because I want CPU
> power. Forcing the CPU to throttle all the time would remove any advantage of
> having bought such an expensive CPU.
Indeed. You might ask on freebsd-mobile@ is anyone else has one of
these beasties, to see whether yours is behaving true to type?
As others have mentioned recently as well as Alexandre since, CPU duty
cycle throttling seems mostly ineffective on modern multi-cores, and is
likely best disabled, leaving stepping between fewer real frequencies to
make positive differences in power consumption, thus heat generation.
If you do reduce the number of frequencies as mentioned, you may want
to reconsider powerd's default values .. only playing around will get
you the best optimum between responsiveness, performance and cooling.
> > * Is tz0 the only thermal zone you have on this machine?
> All of this is TZ1. TZ0 does exist as well but reports a constant 26.8°C
> with a _CRT of 127.0°C - not very interesting.
Definitely no active or passive cooling parameters at all for tz0?
> I am very reluctant to do this. The laptop is barely three months old. As
> long as it is still under warranty, I really do not want to take it apart.
> Also, shaving off 3-4°C will not cut it anyway. To prevent the spurious
> shutdowns without overriding _CRT, I would need the CPU to run at least
> 10-15°C cooler.
I guess google would reveal any overheating problems people with this
model may be having under Windows, or not? You may just have a faulty
unit, and if so installing W on it may be needed for a warranty claim;
sadly, the first thing they might blame problems on is an 'other' OS.
> dev.cpu.0.freq: 1463
> dev.cpu.0.freq_levels: 1597/35000 1463/31000 1330/27000 1197/23000 1064/19000
> 931/15000 814/13125 698/11250 581/9375 465/7500 349/5625 232/3750 116/1875
I'd be interested to see exactly which of those frequencies remain after
you disable p4tcc or acpi_throttle, but maybe you could post dmesg.boot
to see which cpufreq drivers are now being used, to save some guesswork?
Also, is that 35 Watts at full speed for the whole package, all 4 cores?
More information about the freebsd-acpi