If anyone's interested in testing their SocketA motherboard to determine whether the CPU temperature readings are internal or not, here's how I did my test on my A7V333:
Stage 1: Monitor the rate/amount of change in reported temperature as the CPU load (and therefore the actual core temperature) goes up and down
- start a monitoring utility like MBM5 and set the updates to something like 1 or 2 seconds
- let the temperature of the system stabilize at idle and note the temperature
- apply a sudden sustained load to the CPU, such as SETI@Home, Prime95 or whatever
- take notes on how fast the temperature climbs and how long it takes to stabilize
- remove load, take notes on how fast the temperature drops and how long it takes to stabilize
In the case of my A7V333, the Stage 1 test showed the temperature taking two minutes to make its full swing from low to high, and back to low again. Furthermore, the full swing was only 3C from idle to full load. The slow response time and small temperature swing suggested that the CPU temperature was being indirectly measured. On to Stage 2...
Stage 2: Monitor the rate of change of the reported temperature as the temperature of the motherboard PCB itself goes up and down
- get your motherboard out of your case (if your ATX cable is long enough you can do this while your drives stay in place)
- again, start your monitoring utility and set it to update every 1 or 2 seconds
- let the temperature stabilize
- get something cold, like a medical cold pack, wrap it in a washcloth or something, and hold it against the underside of the motherboard's SocketA area, while watching your temperature utility. edit: I don't recommend doing this for any longer than you have to, ten seconds at most. Don't want to crack any solder connections due to the thermally-induced contraction of the PCB...
In Stage 2, the A7V333's response time was practially instantaneous and the temperature drop was radical. This shows that the thermistor is much more sensitive to the temperature of the motherboard than to the temperature of the CPU, indicating that the temperature reading is indeed coming from a surface-mounted thermistor.
It is also possible that a board could have a socket thermistor on a thin, flexible stalk which presses against the underside of the CPU and would be far enough from the motherboard surface that this test might not give conclusive results. A good example would be the Gigabyte 7VRX/7VRXP. However, Gigabyte doesn't claim direct diode readings in the first place.