Quote Originally Posted by unclewebb View Post
I'll have to check up on this. I thought it was reading a sensor that was located in the CPU socket but don't quote me on that.

Looks like the sensors on core0 and core1 are both working fine. My best guess is that your actual core temperature is between 28C and 30C during your test. If you want to keep things simple you could assume 28C so you wouldn't need to do any calibration on core1 and then you could use a -1.0 calibration on core0 which should bring it down to 28C also. If you choose to go up a degree or two, that's fine. Just use calibration factors so core0 and core1 are balanced during this test. I like the simple approach where you only have to use one factor. When you return to your normal MHz, you might need to do a slight adjustment. Core0 and core1 should be pretty much equal at idle even when overclocked and overvolted.

I like using Prime95 v25.6 running small FFTs. Are you using a multi-core version of Prime or running 4 individual instances of Prime? I'm not sure if this would make any difference to your results. At full load I generally find that core0 and core1 are equal while core2 and core3 are equal but the two sets of cores might be off by a few degrees. One user I helped had load temps similar to yours where the two outside cores ran hotter than the two inside cores and I think when he re-did his thermal paste this changed. Anything like installation procedure or an IHS / heatsink that is not perfectly flat can cause slight differences at full load as well as less than perfect sensors. Probably not worth worrying too much about though.

The other test I use is to use Prime95 and to go into the task manager and set the affinity to limit it to run on 1, 2 or 3 of the 4 cores at a time and to switch the load around while Prime is running. If you get some interesting numbers then post them.

I don't believe the Xeon sensors are any different. The 45nm sensors just aren't always capable of giving us the information that we'd like to be able to get out of them. It's sort of random luck and I think you're doing better than most users have. At least you took the time and did your homework so you know exactly what your sensors are capable of and the temperature range where they give you accurate data.
Thanks for the info! I ended up having to use a -1.8 calibration on Core0 to get it roughly matched up with Core1.

To answer your question, I generally run 4 instances of Prime95 and use the Affinity (-A) option to tie each instance to its own core. Is there a multi-core version of Prime95 out there? That would make it much easier...

I think I'm going to take your advice and not return the CPU...too much hassle especially if I'm likely to get another one like this. I'll be building a few more systems with almost identical specs so it will be interesting to see how the next ones match up with this one.