"Core Temp" - monitor temps on K8s and Core/2 CPUs (updated

[Drizzit]

I thought the fan settings in bios used the same readings as the ThermTrip register but guess I was wrong thank you for clearing that up.

[zarathrustra]

This is basically what the difference I'm seeing with temp readings between CoreTemp and ITE SG with my single core, under water, on a DFI LP Expert with no adjustment in the BIOS.

 

The lowest readings are at first boot on a cold morning.

The middle, where the two show a very similar reading, is at normal idle in a average room temp.

The top are at full prime and/or Toast load at the same room temp.

Looks to be a completely different scale:rolleyes:

[red930]

I also found some very interesting information a little deeper in my research.

 

On this link view page 85...

http://www.amd.com/us-en/assets/content_ty..._docs/25759.pdf

 

This means there are a lot of bum processor temp readings out there.

 

Discuss this amongst yourselves.

[Praz]

For affected systems BIOS should add 10C to the signed diode offset value.

In this case the bios would be reporting higher then actual temperatures. That would explain why on rare occasions programs like Core Temp read lower then MBM or the bios.

[red930]

OK

 

Here's the breakdown as best I can figure it out.

 

The motherboard sensor might have a diode located under the processor.

 

The motherboard sensor might have a diode located just outside the processor socket.

 

The motherboard sensor might have a diode under or near the chipset(or Northbridge and Southbridge).

 

The motherboard sensor might have a diode near the PWMIC or MOSFET area.

 

The processor has a diode integrated on the die.

 

Each core has a thermocouple located in the core.

 

No wonder there's such confusion when trying to sort the sensors and I won't get into GPU or SMART sensor reporting.

[Prospector]

Dual Prime SG shows 43-44 CoreTemp shows 53-54?

[Praz]

Is 53-54 each core?

[Guest Kobalt]

154 Incorrect Diode Offset

Description

Some dual-core AMD Opteron™ processors (939-pin package) have an incorrect diode offset

(Thermtrip Status Register [DiodeOffset], Dev 3xE4h[13:8]. The affected units can be identified as

follows:

• CPUID Fn[8000_0001]_EBX(11:0) = 2Ch, and

• Thermtrip Status Register [TCaseMax] (Dev 3xE4h[28:25]) <= 2 (53oC).

Potential Effect on System

Thermal management events that depend on the diode offset field may occur 10 degrees earlier than

expected.

Suggested Workaround

For affected systems BIOS should add 10oC to the signed diode offset value.

 

So then we need a utility that reports the above info, and then we can take the skew of up to 10C into account.

[Praz]

That isn't the only possible error. The accuracy of the thermal diode itself has a possible spread of inaccuracy of ±10C-14C compared to the hottest part of the chip.

[Prospector]

It shows a difference of about 5 degrees from each core. At idle the diff is 1-2 degrees. Also at idle core Temp and SG are virtually the same temps.

[yfital]

i noticed that usually, core temp shows lower temps under idle

and higher temps under load

which makes a lot of sense, cause the mbm's sensor is outside the cpu, and is more "sensitive" to ambient temp + the fact it takes more time for the temp to "crawl" out from the cpu into it...

[Praz]

i noticed that usually, core temp shows lower temps under idle

and higher temps under load

which makes a lot of sense, cause the mbm's sensor is outside the cpu, and is more "sensitive" to ambient temp + the fact it takes more time for the temp to "crawl" out from the cpu into it...

The sensor that MBM reads is located on the motherboard but that isn't where the temperature reading originates. That sensor is an interface between MBM and the thermal diode which is inside the processor. The location of the diode is somewhat of a mystery but from AMD's documentation it appears to be not located at the hottest area of the processor.