Abnormal CPU Core Temps...

[PowerPie5000]

I recently built a budget gaming rig and have just noticed my CPU temps seem a bit too low! I'm using an 'AMD ATHLON II X4 645' CPU along with an 'ASUS M4A88TD-M EVO/USB3' motherboard. I'm not sure if the temp sensors on the motherboard are wrong or maybe it's a problem with the CPU sensors?


These are the temps i'm getting right now using HWMonitor (idle temps):


CPU Temp - 31C (usually around 55C at load)
Mainboard - 25C (usually around 30C at load)


Core#0 - 16C (all cores are usually around 35C at load... still a bit too low)
Core#1 - 16C
Core#2 - 16C
Core#3 - 16C


I get the same results whatever software i use to check the temps. The motherboard hardware monitoring sensor is being detected as 'ITE IT8721F'. I'm using the stock AMD heatsink and running the CPU at stock speed and voltage. My case (In Win Dragon Slayer) has 2 x intake fans on the front and 2 x exhaust fans (one at the rear and one at the top).


Anyone have any ideas? Cheers


Also... my room temp at the moment is 21C

[peterb]

First thoughts are that it is a software problem. Are temperatures available in the BIOS set up pages, and is there an ASUS utility that allows you to read the temperatures?

[PowerPie5000]

First thoughts are that it is a software problem. Are temperatures available in the BIOS set up pages, and is there an ASUS utility that allows you to read the temperatures?

The CPU temp section in the bios doesn't show indivdual cores, it only shows a single temp reading which is 31C at the moment. I just installed 'Asus Probe II' and it also only shows a single CPU temp reading (currently 31C). I'm guessing the motherboard CPU temp sensor is working correctly (it's probably a diode placed in the CPU socket so it won't be 100% accurate) and the actual CPU's core sensors are way off! It's seems strange how all the cores are the same temp, which makes me think there's only a single sensor reading the temps for all 4 cores.... My old Intel Q6600 had a different temp reading for each core.

[peterb]

U can't remember if AMD have on chip sensors or not. I have a dual core AMD set up (running Linux) and that reports the temperatures of each core.

I would have expected the four core to do the same. However http://www.cpu-world.com/CPUs/K10/AM...FGMBOX%29.html shows that it has an on-chip temperature sensor, but doesn't specify if there is one per core.

However I see your post on Tom's hardware has received a similar comment.

[watercooled]

I wouldn't pay it too much attention, the same thing happens on Thuban CPUs which read about 10-15C too low. So, worst case scenario - add 15 to the core temp. I think I remember reading that the sensors are off, but consistent i.e. if 15C reported = 20C real then 25C reported = 30C real. In that case you could use the change in temperature to get a somewhat useful reading, not forgetting the CPU could be way above ambient when idle depending on the cooler. But don't rely on that either, it could just be an completely false claim I've read and I've not heard anything like that from AMD/Intel.

I would like to keep an eye on CPU temps myself but I have no idea how to do so reliably. Some motherboards have fairly accurate socket probes, but that's going to be a few C off the core temps which matter more for most purposes.

Edit: Oh and Thuban CPUs have a single on-die sensor, but I'm not sure about others.

[PowerPie5000]

Sounds to me as though AMD are trying to make out their CPU's are cooler than they actually are... The temps could be off on purpose! Either that or their quality control has gone down hill (i've never had issues in the past reading temps with AMD CPU's).

Anyway, i'll wait and see what Bulldozer and Ivy Bridge are like before i build another PC

[watercooled]

I very much doubt either - no-one with an ounce of common sense would believe their CPU was sub-ambient. Besides, CPUs don't report absolute temperatures anyway, they report the distance from TjMax, it's up to the software to turn it into a useful display - AMD aren't as open about their TjMax values IIRC, but it wouldn't be hard for the software developers to make them a more reasonable value with a bit of work.

[scaryjim]

My Deneb-based Opteron at work appears to have a single on-die temp sensor.

I'm a little dissappointed to hear that it's probably registering incorrectly though, I was rather chuffed at the fact that my workstation was maintaining a steady 42C at full load - not so chuffed if that's actually a steady 57C