How do I measure the REAL onDie CPU temperatures (inside core from its inner diode!)


Even if newer HWmon chips (programmable micro-Controllers with analog inputs!) almost all have the input for that, mostly is not or programmed for that diode or even not connected to it and manufacturers still rather use cheap NTC 10k thermistors (therm.probes) mounted somewhere, usually inside Cpu socket (on latest MoBos even outside/close to it), so the temps revealed are not correct ones even if that chip is compensation programmed thru MoBo Bios.

That would be some kind of ok, if the core would be turned inside to socket, just like it was on old Intel Mendocino Celerons; but these days, because of better cooling, the core is flipped on the outer side toward HSF & temp measuring is FALSE mostly! (and if that chip is not programmed for Cpu diode even if has input for it, does not help soldering additional "missing" resistor on the MoBo & connecting non connected connections on it, like I saw somewhere on the Web some guys done it; the chip reveals who knows what then! )

  Since I always wanted to know my CPUs inside temperature & my MoBo was showing temperatures from standard th.probes properly (but not from the Cpu core; my Winbond HWm chip has that ability, but not used Btw!). I checked that time ago with a calibrated external digital instrument, because I wanted to mount a bunch of chosen th.probes in my setup and even one on the balcony, since I wanted to have ability to switch between them & also to monitor outer house temps during summer & winter. Two years ago I started to think about this topic a bit & a guy (a.k.a Morphling1) on some my country Slo-techwebsite forum I also visit, in meantime, made something like this independent solution of the problem with some Maxim micro-Controller chip connected to his MoBo i2c (SM) bus, following the instructions on AMD´s datasheet about this topic (Btw this guy makes excellent waterblocks!), but I did not wanna follow his example of soldering on the MoBo. Since I luckily have my Cpu on the slotket, I wanted to make an analog & my MoBo "dependent" solution!


After some measurings with multi-meter & studying some Winbond & Intel datasheets and so got some knowledge about the problem, I got some ideas! All thermal probes mostly got their bias (working current) from the 3,6V reference source on the MoBo thru a resistor (10k for NTC & 30k for transistor or diode) connected to HWm chip analog input. So for the diode that current is more than 3x smaller & a voltage on it is around 0,6V not changing much with temperature and variations there are very small, +/- few ten mV in 0° - 100°C temp. range. On the contrary, on the standard NTC th.probe it can vary between 0,1 & 3,5V = much more, for example for 25°C there is approx.1,5V on the HWm chip input!

thermistor sensing
. . . These are some snapshots from Winbond datasheet about different type sensoring...
Trans.-Diode sensing
Since diodes & thermistors are made from quite same material & have similar thermal difference voltage curve response from the room temps reference point, according to temps applied to them but at different voltage size & range levels, I got one day a crazy idea on why not to use the HWm chip ordinary th.probe input & that was about inserting in_between Cpu onDie diode & Auxiliary th.probe input on the MoBo /usually used for GPU temp sensing!/ some kind of DC amplifier, mirrorly compensated also for differences between diode & NTC thermal/voltage response curve & gain around 36dB. BINGO! The "Gadget" idea was born!
I started brainstorming and experimenting with different schematics & so I finalized my therm.probe "emulator", piloted by the Cpu diode! Maybe I invented a Hot water, but YES, I actually figured out how to interface a CPU thermal diode to a Winbond chip that only supports thermistors; so I rather call this an innovation & not an invention! The schematics is "idiot proof" simple, but it does its job nice if properly calibrated and connected & if the chain: Aux_in-HWm chip-Bios-OS-MBM program is measuring temps properly for ordinary th.probes (I was sure about that for my MoBo - I heard some Asus boards have problems with that), because as I mentioned before, the Gadget relies on that "chain" properly showing temps! (it is not TL independent stuff like that one with Maxim chip mentioned before!) & but can be done if having some practice with electronics, since it is not so complicated ...

. . . . . . Finally, the SCHEMATICS w/o parts values listed of this mine "wonder_baby" looks like this :


© Zdenko Jerman-Spajky