Sensors detective

I have an Asus M5A97 PRO motherboard, which includes the common set of sensors. The lm-sensors free software correctly detects two chips, k10temp-pci-00c3 and it8721-isa-0290, corresponding to my Phenom II and my motherboard, respectively. However, by default the output of sensors looks like:

Well, not exactly useful. But it turns out that we can do better than that with a little work. The lm-sensors wiki page about voltage labels and scaling describes a simple method by which one may make sense of their motherboard’s sensors.

My motherboard’s sensors chip model is IT8721F, which measures raw values ranging from 0 to 3.06V in steps of 12 mV (8 bits of precision). In order to fit its range, voltage rails such as +5V get a scaling factor. Our job is to figure out which is the scaling factor and which input of the sensor is measuring it.

We start by collecting samples of the correct data. I didn’t bother resetting my computer to look at my own BIOS, and instead just searched the web for screenshots of M5A97′s BIOS screen. In the end I had collected the following samples:

The +3.3V sensor had already been correctly named by lm-sensors (in3). The Vcore sensor (CPU) is quite easy since it’s not scaled (in2). +5V and +12V are the tricky ones.

As you can see on the snippet above, I’ve sorted the samples I’ve collected and calculated the differences in mV between them. Looking at the +12V samples first, we see a pattern of alternately adding 51 or 52 mV, thus we deduce that the measurements have steps of 51.5 mV between them. This works out to a scaling factor of 51.5/12. The +5V is analogous and we calculate a scaling factor of 21.5/12 for it.

(Up to this point, something has been bothering me. If we take a look at sample 12.200 V, for example, we’d expect it to be a multiple of 51.5 mV. However, the closest multiple is 237 * 51.5 mV = 12.2055 V. No matter how you round that number, you’ll never get 12.200 V. I wonder if I calculated something wrong.)

There remains the problem of correctly labeling the inputs. Labelling the fan and temperature sensors is quite straightforward. On the other hand, labelling +5V and +12V correctly is quite tricky, since both in0 and in1 have raw values that look okay for both +5V and +12V. At this point I’ve rebooted my machine and took note of my own data:

Then I’ve booted into my Arch Linux and took the raw data from lm-sensors (pasted at the end of this post). Based on it, I guesstimated that in0 corresponds to +12V while in1 corresponds to +5V. This leaves us with in4 to in6, which I chose to just ignore.

And, finally, it seems that the temperature reported by the k10temp isn’t real temperature, instead it’s a relative temperature. It’s mostly meaningless since the motherboard correctly informs the CPU temperature.

The end result is:

Much better! If you have the same motherboard as mine and want to have this beauty, just drop this snippet in your /etc/sensors.conf:

I’ve posted the configuration snippet above to the lm-sensors mailing list as well, so hopefully it will make it to the lm-sensors configurations wiki page :).

The raw data after the BIOS screenshot: