Real Temp - New temp program for Intel Core processors

[unclewebb]

JohnZS: The most Extreme thing about your Quad seems to be the sensors. My Q6600 isn't great but the Distance to TjMax which represents the raw data coming from the sensors only varies by about 4 or 5 degrees at idle from core to core. Your center two cores show a difference of 14!

All I can say is good luck to Intel to try and come up with a simple, user friendly formula to cover the wide variety of sensors that I've seen. Trying to cover a difference of 14 is stretching RealTemp's capabilities for covering up this mess.

If you're bored John could you try running a test with the calibration factors you're using? Run at whatever your normal MHz / core voltage is. Turn on the RT logging feature and set it to 1 second. Run 1 minute idle, followed by about 2 minutes of Prime95 small FFTs on all cores and then back to idle for a minute. E-mail or PM me the log file when you're done.

It is normal during this test for core0 / core1 to closely track each other and then for core2 / core3 to closely track each other. Core2/core3 being about 5C less while running Prime compared to core0/core1 is also fairly normal. I'm just curious to see how all these cores track each other during transition from idle to full load and back again.

Prime95 v25.6 32 bit
Prime95 v25.6 64 bit

One theory I have is that when a processor has badly mismatched sensors, Windows XP or maybe the processor itself will schedule a bigger work load on the core that reports the lowest temperature. When uncorrected, core2 reports that it is running so cool that it will get to do some extra work. To test for this run Prime95 and for Number of torture test threads to run enter 2. This will run 2 instances of Prime but they will be constantly spread to your 4 cores so no one core gets too hot. Do a CTRL+ALT+DELETE and bring up the Task Manager and you should get to see what core is getting the biggest chunk of work. I'm not sure if that's important but I found it kind of interesting.



My core2 reads highest Distance to TjMax so it gets to work the hardest. The load seems to be balanced for both sets of cores. Core0/core1 share half the load and core2/core3 share the other half.