Archived from groups: alt.comp.hardware.pc-homebuilt (
More info?)
On Thu, 23 Jun 2005 16:42:40 -0500, David Maynard <nospam@private.net>
wrote:
>DaveH wrote:
>
>> On Wed, 22 Jun 2005 21:28:32 -0500, David Maynard <nospam@private.net>
>> wrote:
>>
>>
>>>DaveH wrote:
>>>
>>>
>>>>I have a Northwood 2.8 ghz on an Epox 4PDA2+ and recently noticed CPU
>>>>temps bouncing around 65 C and higher. Arctic Silver and Zalman
>>>>7000Cu---don't think that's the problem, temps were stable at ~40 C
>>>>for 2 years.
>>>
>>>Ay idle or under load?
>>
>> *********
>> I think it would climb to 45-48 C under at 100% CPU usage, running
>> Sisoft Sandra and Prime 95 simultaneously, for example. Now it bounces
>> between 65-80 C at idle and bounces around 95-100+ C at 100% cpu
>> usage. I've got to think that if those temps were remotely accurate,
>> my cpu would have given up by now.
>> *********
>
>>
>>>>I suspect the mobo sensors have had it.
>>>
>>>Sensors dying is probably the least likely possibility, unless you've been
>>>in there with a ham handed screw driver and damaged them.
>>
>> The 4PDA2+ is supposedly notorious for producing unstable temp
>> readings. They've bounced all over the place since Day 1. This is why
>> I suspect the sensors or sensing complex has simply competely
>> deteriorated.
>
>Well, this is the first time you've mentioned that they "bounced all over
>the place since Day 1" and if they've never been reliable then nothing
>they've told you is reliable, "stable at ~40 C for 2 years" or now.
>
>I don't know what you mean by 'bounce', though. Heatsinks take time to warm
>up, or cool down, but an internal diode will sense the CPU temp
>immediately, because it's 'there', so if you're monitoring an on-die
>thermal diode it can change by as much as 10C, or so, almost instantly from
>a load change.
>
>
>>>>Machine is perfectly stable
>>>>otherwise.
>>>>
>>>>I could replace the mobo, or CPU or--- mobo and CPU. Logic dictates
>>>>that I replace the mobo and retain the current CPU.
>>>>
>>>>Am I correct in assuming the probability is low that a faulty CPU is
>>>>the problem?
>>>
>>>None of those are likely. Assuming temps really have risen under the same
>>>operating conditions it's more likely either a dirty heatsink, heatsink fan
>>>failing (running slow), thermal compound migration/dry, heatsink
>>>mechanically shifted (either the fan or on the processor), or a case
>>>cooling problem caused by dirty fans, fan failure (or running slow), or
>>>airflow blockage.
>>
>> Everything's clean. I checked that first thing. I've also suspected
>> mechanical shifting or otherwise mechanical compromise of the HSF/cpu
>> arrangement. But this machine is absolutely stationary and screws
>> firmly torques. It's hard to envision the HSF simply drifting out of
>> place.
>
>You're neglecting, even if nothing else, expansion/contraction with
>temperature.
>
>>
>> Is Arctic Silver prone to migration?
>
>Not excessively but with your comment below about the heatsink sliding
>around it sounds like you put too much on.
>
>However, most, if not all, 'fluid' type thermal compounds are potentially
>suspect over a period of years.
>
>
>>>Other causes could be a software/configuration problem preventing the
>>>normal idle power saving or something (e.g. a virus/worm) causing 100% CPU
>>>activity.
>>
>> The problem shows up in BIOS which makes me suspect it's not a Win2000
>> infection problem. Task Manager shows normal 1-2% usage at idle in
>> Windows.
>> BIOS, which I just flashed to latest (no change), could be infected, I
>> assume.
>>
>> I guess I have no choice but to pull the HSF and examine the interface
>> then re-install as per your advice. I've hesitated doing so because
>> the the Zalman 7000Cu was a nightmare to install. It tends to slide
>> around on the compound (used specified amount). I know others have
>> reported ease of installation with that cooler.
>
>You'd have to pull it anyway to replace the motherboard so you might as
>well check it before dumping a wad on money on one.
>
>
>> Yesterday I had a cpu temp alarm at boot even though that function is
>> disabled in BIOS. I re-booted successfully without alarm and USDM
>> showed cpu temps bouncing between 0 and 90 deg C. Such flakiness looks
>> electronic to me in nature, but I'm pulling the HSF to rule that out.
>
>You really have two problems. Or one known problem and one potential
>problem. The known problem is the temperature readout. The other is whether
>it's really hot or not.
By bouncing I mean something like 35- 40- 45 - 41- 36- 39 at roughly 1
second intervals, for an average of ~40 C. That's what I've had since
the beginning, then the recent increase which occurred over an unknown
interval. You're right of course, it's never been reliable, but a
distinct change in trend has occurred.
I've got USDM set to 1 sec re-flash intervals. With Task Manager
indicating 0-2% cpu usage, the temp is "bouncing between ~55 and ~110
C. That cannot reflect actual thermal conditions, so I must conclude
that a temp sensing clearly exists. That being the case, does it
follow that I have an actual high temp condition? If I do then it is
not related and thus coicidental. Insufficient data, but you may see
why I suspect a sensing problem only and not an actual high temp
condition in light of otherwise perfectly stable operation. OTOH I
could be utterly wrong. I assemble our machines but my knowledge is
incomplete.
Thermal contraction and expansion is something that never occurred to
me. The 7000 has always run barely warm or tepid to the touch, which I
always assumed to indicate superior heat transfer capacity; but in
retrospect it could have indicated insufficient heat transfer at the
interface from the beginning, as you suggest. I have to pull the
heatsink, no choice.
I guess it's logical to assume that the CPU is ok.
Thanks for a very thoughtful reply.
Dave