Fluctuating CPU temps / thermal throttling

[shredisn@tdead]

Hi again folks, hoping for some ideas with a problem I'm having with a recent build... this might be a bit long..

Take a look at this:



This is for a 2.8Ghz P4C, running at stock 2.8 no oc no ov, the temps are jumping wildly between 45 & 60 under load (two instances of SETI). The drop off to ~35 is the no-load situation ie both SETIs stopped. I thought perhaps Speedfan was reading the wrong sensors, but this seems to match what I see in the BIOS, with even the no-load temps jumping madly in the BIOS health screens.

Surely I'm not mad to think the temp should be pretty steady, no? It's about 39 (+/-2) under load in another P4C I built a few months back (a 2.6 oc'd to 3.275), and about 28 with no load.

The HSF is a Zalman CNPS7000-Cu, as in my other PC, and I sure didn't expect this kind of weirdness. I initially thought I'd mounted it wrong - eg bad application of Ceramique, getting air trapped under it when trying to screw the thing down etc. However have remounted about 4 times now with the same results. Last night I resorted to fitting the stock Intel HSF (untouched with TIM!) and unsurprisingly the results are worse. So I guess duff HSF is out of the equation now.

The mobo is an EPOX 4PDA2+ Deluxe v2 and I'm now wondering if the fault lies here. My current plan of attack is to switch the CPUs between my ASUS P4P800 and this EPOX and see if the 'good' 2.6 runs funny in the EPOX and hopefully the 2.8 will run stably in the ASUS. If this is what happens then I have a dodgy mobo, otherwise I know the CPU is at fault. Sounds like a sensible plan, no?

My CPU temp expectations aren't mad are they? I've had the system up to 3.5Ghz (250fsb 1:1) but you can imagine what happened with the reported temps! I didn't leave it at 3.5 for long

How are the temps reported to the BIOS? Is there a sensor in the CPU itself? I'd guess so and that this'd be how the thermal throttling works, no? I think I'm seeing this even at stock (SETIQueue reports total time vs CPU time and gives an 'efficiency' which is much lower than for the other PC)

Anyone seen this kind of thing before? Any ideas suggestions welcomed!

[shredisn@tdead]

not even a "yup that graph looks all messed up"?

[Knoxville]

Have you tried setting the fan to a constant speed instead of allowing speedfan to control it, same thing used to happen in my shuttle until i set the fan to a constant speed.

[unrealrocks]

I had the same problem no a A8V Deluxe ... just needed to switch QFan off (same sort of thing)!

[ives]

I know that AMD mobos have a sensor under the cpu ,just touching the underside of the die.
However, I think i read that some pentiums actually have a sensor on the chip itself.

Either way, these are the BIOS temps that you get.

Have you tried other burn-in apps to see whats happening?
Try Prime 95 or Pi to see if the same thing happens.That way you can rule out its not something with SETI.

Other than that ,it could be a fault in the temp probe that reports to the bios.
Try an external temp probe and put it close to the cpu die and see if it really does fluctuate.

[shredisn@tdead]

Thanks folks!

1. Don't think Speedfan is actually controlling my fans. I don't hear any speed changes and the fan speed graphs are pretty much constant

2. Don't have a temperature probe - wouldn't know where to poke it! Under the CPU retention bracket to get under the CPU (if even possible) or do you shove it between CPU & HSF? Might be handy to have one - any recommendations?

Anyways, I swapped the CPUs round in my 2 boxes and the new 2.8 is in the older Asus board and working fine, currently at a cool 33oC under load.

The older 2.6 is now the new Epox board and once again the reported temps are bouncing about, now between 45 and 65.

So, I think this last experiment proves that the mobo is wonky as no matter what CPU I put in it I'm getting funny readings. And both CPUs are cool in the Asus board.

Considering I've had the mobo for under 2 weeks and it's only been in a system for 1wk then I think I'm going to see about RMA-ing it, no?

Also then, anyone know how the thermal throttling works on the P4s? Does the CPU rely on the sensor information from the mobo?