18 Feb 2012 yeupou   » (Master)

Getting accurate temperature reading for the CPU

On my main workstation, lm-sensors provides apparently contradictory temperature reading for the CPU, depending on the sensor:

radeon-pci-0200
Adapter: PCI adapter
GPU Temperature:  +62.0°C  

k10temp-pci-00c3
Adapter: PCI adapter
CPU Temperature:  +17.0°C  (high = +70.0°C)
                           (crit = +70.0°C, hyst = +68.0°C)

atk0110-acpi-0
Adapter: ACPI interface
[...]
CPU FAN Speed:          1890 RPM  (min =    0 RPM)
[...]
CPU Temperature:         +32.0°C  (high = +90.0°C, crit = +125.0°C)
MB Temperature:          +42.0°C  (high = +45.0°C, crit = +90.0°C)

17°C, as reported by the CPU sensor, seems very low especially as the temperature of the room the computer is running inside is at least 17°C already. Clearly, the Motherboard sensor (atk0110 / IT8716F chip) readings, same as what the BIOS reports, are more sensible.

There’s actually lot of misinformation on the web. For instance, CoreTemp author, a proprietary software for MS Windows to provide CPU temperature readings, states on his front page that “all major processor manufacturers have implemented a DTS (Digital Thermal Sensor) in their products. The DTS provides more accurate and higher resolution temperature readings than conventional onboard thermal sensors”. Possibly, probably, right: k10temp may be more accurate than atk0110. However, when the same author reply, on his forum, to a user asking about inconsistencies in CPU temperature readings, clearly interested in real and not relative temperature (he wrote :”I’m running water cooling and the temps aren’t high during load but just wondering about the accuracy”, high temperature is meaningless on an undefined relative scale), that “I’d say that Core Temp is more accurate, especially at higher temperatures. The ASUS programs sensors are based on the motherboard and depend on an external chip. The sensors Core Temp reads are located in the CPU itself and the values are read directly from the CPU registers.”, he clearly shows misunderstanding of what superior accuracy CPU sensors really mean.

As documented by AMD mentioned in the k10temp linux module doc, “[k10temp] is the processor temperature control value, used by the platform to  control cooling systems, [...] is a non-physical temperature on an  arbitrary scale measured in degrees, [...] does not represent an actual  physical temperature like die or case temperature. Instead, it specifies  the processor temperature relative to the point at which the system must  supply the maximum cooling for the processor’s specified maximum case  temperature and maximum thermal power dissipation”.

I was about to publish this article without paying attention to Intel sensors, but a quick search lead to me even worse: a comment about Core Temp in a doc titled CPU Monitoring with DTS/PECI stating : “These tools provide a convenient way to see the temperature variation reported by the sensor [...] There are several issues with these tools. First the assumed value for Tj may not be correct and thus impact the accuracy of actual temperature reporting. Secondly the DTS is only accurate when in the adjacency of Tj. Not knowing the intention and effective range of DTS, the tools try to compensate with the inaccuracy of low temperature reading, which may not be a correct interpretation.”

However accurate they may be, relative readings provided by CoreTemp for AMD K10 are almost meaningless to an end user (while great for the system for fancontrol and such), likely expecting to be able to compare them to other (motherboard|hard disk|etc) readings. In my case, surely k10temp mean something (17°C is low) but it makes no sense to compare it to the room (20°C), GPU (62°C), PATA Hard Disk (39°C), Motherboard (42°) or any other else temperature. In short, except if you know exactly what you’re doing, use Motherboard sensors and if you’re looking for an alternative to CoreTemp, try Open Hardware Monitor.


Syndicated 2012-02-18 15:28:31 from # cd /scratch

Latest blog entries     Older blog entries

New Advogato Features

New HTML Parser: The long-awaited libxml2 based HTML parser code is live. It needs further work but already handles most markup better than the original parser.

Keep up with the latest Advogato features by reading the Advogato status blog.

If you're a C programmer with some spare time, take a look at the mod_virgule project page and help us with one of the tasks on the ToDo list!