q9400 bad sensor?

[Oliver_FF]

I've just built my new rig and i'm checking out the temperatures - at full load I seem to have an 8c temperature difference between one of my cores:


Would it be fair for me to assume that the other three are reporting correctly and core0 just reads high?

[unclewebb]

What you are seeing is a difference in TJMax from one core to the next. Intel admits that TJMax is not an exact number but won't tell us how much error there is in this number. My best guess based on observation is a variation of about 10C from core to core on a 45nm Quad. The previous 65nm CPUs seem to have a closer tolerance than that.

Would it be fair for me to assume that the other three are reporting correctly and core0 just reads high?

Unfortunately you can't really assume anything with these sensors. I'm pretty sure that core 1 has a sticking issue at a Distance to TJMax of 71. Core 2 and core 3 match nicely but that doesn't guarantee that they are actually accurate.

For an accurate cool down test you need to use the Small FFT option in Prime 95 and you need to have as little as possible in the background running. You can't be surfing the net or doing anything else while testing and with Vista you need to let it start up and settle down which can take 5 minutes or more on some systems especially after a fresh install.

For your CPU I'd leave core 2 and core 3 as is and I'd use TJMax = 105C for core 0 and then adjust the Idle Calibration factor so core 0 lines up with core 2 and core 3. I think at higher temperatures, core 1 will catch up to core 2 and core 3 so I'd leave TJMax=100C for those 3.

You could use a calibration factor on core 1 to try to balance it better with core 2 and core 3 when the temperature is above 29C (its sticking point) but it's not going to be accurate at idle because it's stuck.

An interesting comparison is from the 34% level to the Idle level. Core 0 moves 6.0 degrees, core 1 only 0.7 degrees, core 2 and core 3 are both 4.4 degrees. Core 0 vs core 2/core 3 shows the difference in slope of the temperature curves Core 1 barely moving helps confirm that it is stuck.