Measuring CPU temps with AMD’s internal diode. — Joe
SUMMARY: Reading on-die diodes for both Intel and AMD CPUs for enhanced temperature measurements.
With AMD XPs and Palominos now incorporating on-die diodes similar to Intel’s PIII and PIVs, thermal measurements and heatsink testing is moving into a new arena. Derek over at Void Your Warranty, for example, is pursuing direct diode measurements as well. While motherboards are coming that will support direct die readings from XPs and Palominos, reading the die directly gives heatsink testers more accuracy than relying on measurements filtered through a motherboard.
One way to do this is with Analog Devices Evaluation Board ADM1023. This board is mentioned in both AMD and Intel documents as suitable for direct die readings. I decided to purchase the ADM kit, as it comes with a nice software package and is very simple to use – it plugs into a parallel port.
It does require that you tap into the on-die diodes directly by soldering connections on the back of the motherboard. You should also use shielded twisted pair wire to avoid any EMI interference with the readings. Also note that if the motherboard taps into these already, you have to cut the traces. I am using an Iwill XP333 rev 2.0 board for AMD direct die readings – this board does not tap into the diode pins.
Both AMD and Intel have documentation to locate the appropriate pins (AMD pub # 24228, Nov 2001, and Intel PIV Datasheet, pub # 249887, Aug 2001). Be very careful in locating them – particularly note that “I” and “O” are NOT used in the alpha pin location schemes – that one fact drove me nuts at first! Shown below is what I did to the XP333:
This close shot shows gives some idea of the tolerances – the top is the shield braid which is the ground, and the other two attach to the diodes + and – pins. The are then attached to the Evaluation Board for direct die readings.
Note that you need two PCs for this – one to run the board and software and one to serve as the test vehicle.
The software that comes with the board is quite nice – it allows a lot of flexibility on graphing, data capture (all data can be output as an Excel file) and accessing the test chip’s registers; this is the data capture part:
This is a view of Prime 95 running on an Iwill XP333, XP1700+, cooled by a Glaciator 2, temperature samples taken every 1/8 of a second – it can be set from 1/8 to a reading every 16 seconds.
The software allows users to vary the register contents of the ADM1023 chip directly:
I’ll be using this package to give readers more detailed information in addition to our other testing methods. Also, this is another test vehicle which gives a level of discrimination to more accurately test temperature related issues.
Why some continue to insist that the in-socket diode is an acceptable test instrument is beyond me.