Multiplier drops under sustained load

[gregorius]

I have a Zalman AIO water-cooled 4770K on an ASUS Z87I-Pro, overclocked to 4.2GHz (from stock 3.5) It’s been stable at that overclock for several years, but recently it’s started to lose the overclock (in the sense that the multiplier drops from 42 to 35) after a few minutes at full load.

I’ve been testing it with a prime95 load to see how this rig behaves in different circumstances, and things appear to be pointing towards a thermal problem. If I up the voltage, the multiplier drops to stock more quickly; if I down the overclock to 4GHz, it loses the multiplier more slowly. The multiplier returns to 42 after a few minutes rest (computer left on, but no applied load).

HWInfo is reporting that the CPU core temperatures reach 60C (which shouldn’t be a problem for Haswell, surely). The reported motherboard temperatures appear to be nonsense (which I understand to be a problem with this generation of boards), going down as time-under-load progresses! However, the multiplier drop appears to happen at a consistent (nonsense) motherboard temperature; also, the multiplier comes back up to 42 at a consistent temperature reading.

So, my questions are:

(a) What order of steps should I go through to try to cure this problem? I’ve tried a few obvious things like cleaning out the case and re-seating leads.

(b) Are there any obvious candidates for what would be overheating to cause these symptoms?

(c) Or do I have to admit that overclocked systems do eventually die of over-exertion?

 

[gregorius]

It sounds like you're saying a specific system temp reading is affecting the clock speed.

I suspect that based on the readings, but also based on the way the time it takes for the multiplier to drop when I change the voltage. More voltage, quicker multiplier drop.

So, regardless of having the wrong value I would figure out what that reading comes from, and point a fan at it. If it's VRMs overheating, and the BIOS reads fine but Windows reports it wrong, that's a start. Thermal pads could have dried out for instance.

The readings are labelled "temp3", "temp4" and "temp5" in HWInfo, which doesn't give me much of a clue!

Are they nonsense (irrelevant/random readings) or are they good readings with a dodgy offset on them?

That's quite possible, as they do behave consistently over time under load - they start very high (>100c) and drop down under load (to around 80C), so it's quite possible they're just being mis-decoded. This appears to have been a common problem (google "speccy motherboard temperature", for example).

What are you using to read temperatures? Some software can have bugged numbers on some boards. You may be able to find out if they really are going over a certain point then limiting clocks.

I'm using Speccy and HWInfo.

Edit: what are the readings in the BIOS menu?

Bios only appears to report CPU core temps, which are normal. I'll follow up on that idea of dried out thermal pads; time to take it to bits! Thanks for the help!