Crazy CPU sensors? Wondering if I need to RMA anything?

ron2d287

Honorable
Apr 11, 2013
20
0
10,510


Should I just run a blend test?
 

zyky

Distinguished
Sep 12, 2006
174
0
18,710
With SpeedFan, you're supposed to define what's considered "hot" for each temperature - and the default is a pretty low 40c desired / 50c warning.

HW Monitor was showing you having a sensor labeled CPUTIN at 89c, peak of 92c. Most people would consider that hotter than they'ld like for any component, if that is a valid sensor.

You can also enable/disable particular sensors from showing/counting in SpeedFan's configuration if it's not a sensor that's actually connected to something.
 

CompuTronix

Intel Master
Moderator
In Hardware Monitor, "CPUTIN" is indeed a valid sensor, which in SpeesdFan, is typically "Temperature 2 offset". Keep in mind that there are 5 thermal sensors in a 4 Core processor; one Analog Thermal Diode (CPU temperature), and four Digital Thermal Sensors (Core temperatures).

The sensor in question is the single Analog Thermal Diode. It's located in the center of the lower layers of the processor package substrate material. The sensor's DC voltage level is converted from Analog to Digital (A to D) by the motherboard's Super I/O (Input / Output) chip, then calibrated by look-up table in BIOS. Unfortunately, BIOS is all too often mis-coded, and sometimes extremely so.

Since there are a wide variety of 2nd and 3rd generation Core i processors which can drop into a wide variety of socket 1155 motherboards, BIOS vendors frequently get the offset values wrong. However, BIOS updates sometimes include fixes for inaccurate CPU temperature.

The Core temperature sensors, which are factory calibrated, are located directly at the heat source on the top of the Cores, but the CPU temperature sensor is not, so there's a 5C gradient between them. So in a properly calibrated system, CPU temperature should normally run about 5C below average Core temperature.

Download Real Temp 3.7 to monitor your Core temperatures.
 

zyky

Distinguished
Sep 12, 2006
174
0
18,710


Premade sensor io chips generally have multiple inputs with predefined labels such as "cpu" "aux" etc, but it doesn't necessarily mean the motherboard manufacture hooked up these sensors in this way. Generic sensor reading programs such as SpeedFan, AIDA, HWMonitor, HWInfo, Sandra, etc, can report the IC's name and the value it's reading off that sensor, but it doesn't mean that the sensor labeled CPU temp in is actually a thermal diode underneath the CPU socket.

These analog sensors are not located on the CPU and BIOS does not need to be calibrated to each cpu. Perhaps you should read the documentation that lead up to the release of the RealTemp program you recommended. It's the other on-cpu thermal sensors that are not accurate at reporting temperatures and were only designed to catch when the CPU got to the point where it would have to throttle itself, they were not designed to report temperatures.

Meanwhile, these generic ICs with temperature input probes that motherboard manufactures use, aren't guaranteed to use every temperature, voltage, fan speed sensor actually connected to something. Therefore, some readings can therefore be invalid.

With a reading that high compared to the others, I'd wager that temperature is for the VRM on the motherboard.

if the original poster has a aftermarket motherboard meant for end-user consumers, they likely have utilities for monitoring these values that will properly name each sensor which is actually used on their motherboard - which he/she can then correlate to the values listed in other programs to know which sensor is what.
 

CompuTronix

Intel Master
Moderator
ZYKY ... I beg to differ. I am the author of the 16 page Core i and Core 2 Temperature Guide, which was the most popular Sticky here at Tom's for over 3 consecutive years. You can Google it.

I have consulted with the author of Real Temp, Kevin Glynn, on many occasions, as well as the authors of CoreTemp and SpeedFan. I've invested many hundreds of hours of research on this topic, and I've read hundreds of pages of Intel's White Papers and various documents ... so I know what I'm talking about.

Perhaps you should read what I wrote more carefully:

"Keep in mind that there are 5 thermal sensors in a 4 Core processor; one Analog Thermal Diode (CPU temperature), and four Digital Thermal Sensors (Core temperatures).

... the single Analog Thermal Diode. It's located in the center of the lower layers of the processor package substrate material. The sensor's DC voltage level is converted from Analog to Digital (A to D) by the motherboard's Super I/O (Input / Output) chip, then calibrated by look-up table in BIOS."

 

cmi86

Distinguished


That was a good read and really helped me understand my Q6600 back in the day. ZYKY you should really listen to computronix, he definitely knows whats up on this matter.
 

zyky

Distinguished
Sep 12, 2006
174
0
18,710


SpeedFan, AIDA, et al, do not ask BIOS/ACPI for the temperature. They read it directly, usually via i2c/smbus, from the super IO IC, if any post calibration needs to be done, it's up to those programs - and there is no guarantee which input was used for what on the Super IO IC - meaning CPUTIN may not be the CPU temperature.
 

CompuTronix

Intel Master
Moderator
ZYKY,

While I agree with you that the OP's Hardware Monitor "CPUTIN" is a highly suspicious value, it's highly unlikely that it's indicating a VRM, since his rig is at Idle. Nevertheless, just so you understand that I understand, and for the benefit of others who may read this, please allow me elaborate regarding how processor temperatures work:

The signal path for the Analog Thermal Diode originates onboard the processor. Again, this is a thermal diode located in the center of the lower layers of the processor package substrate material, below the cores. It's not part of the motherboard socket assembly, as it was done with a thermocouple in the socket a decade and a half ago.

The signal travels from the processor pads to socket pins, onto traces on the motherboard to the Super I/O chip. The I/O chip then performs the A to D conversion and outputs the digital value onto the ISA Bus. BIOS then reads the CPU temperature data, applies the appropriate offset correction and writes the new data back onto the ISA Bus.

The Analog Thermal Sensor is completely separate form the Digital Thermal Sensors (DTS) located on the top of each Core at the heat source. Each DTS is comprised of an "array" of guess what ??? ... 8 analog thermal diodes which measure the "hot spots" on an individual Core. The A to D conversion is performed within the DTS package, then a comparator outputs the highest measurement from each Core to the ISA Bus.

Although I agree that motherboard manufacturers do make mistakes with a variety of items including basic signal connections, it's the exception rather than the norm. Mistakes with BIOS coding are far more common, especially lookup tables for CPU temperature. Considering the number of motherboard models, socket types and processor variants, it's not hard to understand. This is why BIOS updates sometimes have fixes which may include "CPU temperature corrections".

So if you really want to get technical, there are a total of 33 Analog Thermal Diodes in an Intel Quad Core; 8 per each DTS, or 32, plus 1 in the substrate that measures overall CPU temperature, which runs 5C cooler than the Cores due to it's relatively distant location from the heat sources. This CPU sensor is intended to "emulate" the thermocouple that Intel embeds in the IHS for lab testing only, and is where the "Tcase" Thermal Specification originates, which is shown in their Processor Spec Finder - http://ark.intel.com/products/52210/Intel-Core-i5-2500K-Processor-6M-Cache-up-to-3_70-GHz

Keep in mind that Intel conducts it's lab testing under tightly controlled environmental conditions at 22C, which is "Standard Ambient". As I've pointed out so many times in countless threads, "Tcase" is NOT Core temperature ... it's CPU temperature, which in a properly calibrated system, runs 5C cooler than "average" Core temperature at a Steady-State 100% workload. Since there are always thermal inaccuracies among Cores, average Core temperature is most important.

This means that for a given variant, simply add 5C to Tcase (CPU temperature) to get a relative Tjunction (Core temperature). This is the temperature at which Intel certifies a processor variant. Tjuntion Max, or Tj Max is, of course, the Thermal Specification at which a processor variant will shutdown.

For the i5 2500K:

Ambient = 22C.
Tcase (CPU Temperature) = 72C.
Tcase to Tjunction Offset + 5C
Tjunction (Core Temperature) = 77C
Tjunction Max (Shutdown Temperature) = 98C

Excessive heat kills electronics. Personally, I believe that any temperatures higher than about 75'ish are just too hot for extended stability testing, day-to-day operations or long term processor health.
 

CompuTronix

Intel Master
Moderator
ron2d287,

The bottom line is that Idle temperatures are of little concern. Load temperatrures are most important. Also, keep in mind that it is not recommended to run temperature monitoring utilities simultaneously, as they sometimes can interfere and conflict with one another, which can create erroneous values.

You need to run Prime95 Small FFT's to determine how your Core temperatures are responding to a steady-state 100% workload. A 10 minute run is adequate for your processor to reach thermal saturation. It's best to remove the covers and run all fans at 100% RPM. Do not run Prime95 Blend, because it's a cyclic workload which is designed to test memory stability, and does not yield steady-state processor temperatures.

You can run a BIOS update to see if it corrects your high CPU temperature, and you can use SpeedFan to calibrate all 5 processor temperatures. If you can master the learning curve and get SpeedFan set up properly, then it's really sweet. I especially like the "Charts" to visualize temperatures and voltages when thermal testing and benchmarking.

If you find SpeedFan to be a little overwhelming, then as I mentioned in a previous answer, download and run Real Temp 3.7, which I highly recommend as the best Core temperature monitoring utility available. If you are more comfortable with something really basic, then just use CoreTemp 1.0. When you post your results, remember to include Ambient temperature.

Hope this helps,

Comp :sol: