Real Temp - New temp program for Intel Core processors

[unclewebb]

axis: Just to put things into English. What my low MHz / low voltage test is designed to do is to lower your core temperatures as low as they can possibly go. There are 101 variables that will effect your load temperatures but at Idle, if you can control these two variables, you should find that your reported Idle temperatures are a little above your room temperature. The digital thermal sensors are calibrated by Intel to trigger thermal throttling at TjMax and generally speaking they are very accurate but as you move farther and farther away from TjMax, the sensors no longer move at the same rate as changes in your core temperature. Read the explanation in post#1 and let me know anything that doesn't make sense.

ChrisZ: Which brings me to your data ChrisZ. Your sensors are doing a great job of proving the point that they shouldn't be fully trusted at idle.

Looking through your data the first thing I notice is that core1 is stuck at 22C and can't go lower. The next thing I look for is two cores that are running mates. By that I mean that when you apply a Prime load and then go back to idle the two cores will track each other very closely. This happens with core0 and core1 once the temperature goes above 22C and core1 becomes unstuck.

I think with a +2 Idle correction for both of these cores that you should get some reasonably believable and accurate data. Core0 will be very accurate from idle to TjMax and core1 will be just as accurate as soon as it gets above the sticking point. It is presently sticking at DTS=73 so whenever the DTS goes lower than this number on core1, it should be fine.

Core2 and core 3 are a bigger mess. Here's some data from the log you sent:

11:27:52 25 25 13 17
11:27:53 29 29 18 22
11:27:54 30 30 19 23
11:27:55 31 31 20 23

I have some faith in core0. In the above data, core2 is reporting 11C to 12C lower than that.

11:27:24 16 22 10 10
11:27:25 15 22 10 10
11:27:26 15 22 10 10
11:27:27 15 22 10 10

In this data core0 is still moving freely but I think your other 3 sensors are stuck. If core2 wasn't stuck at 10 I think it would be showing readings about 11C less than core0 which is about 4C. I've seen an E8400 with similar issues where one core wasn't just too low at idle but way too low and was even reporting negative numbers though the user was only air cooled.

I usually don't have 100% faith in these sensors until core temps are over 60C and because you are water cooled you are a long way from that. Below this point, some sensors can be slightly manipulated to produce some reasonably accurate temperature data but the sensors on core2 and core3 of your Quad don't fit that category.

For core2 and core3 I would bump TjMax to help make the numbers a little more accurate even if they're not ever going to be 100% accurate.

TjMax0=0
TjMax1=0
TjMax2=1
TjMax3=1
Idle0=2
Idle1=2
Idle2=2
Idle3=2

If you were air cooled and had some data up near 60C then there might be a better set of correction factors for your cpu. For your situation, I think the above factors is your best compromise for accuracy in the range you operate.

Core3 will track core0 and core1 much closer and core2 won't be too far behind.

fng77: Congratulations on your super E2160! The problem with my original benchmark is the timer I used is totally proportional to boot MHz so if a user used SetFSB after booting up, it wouldn't properly show any improvements in the benchmark score. My replacement timer may have been good in theory but obviously I need to find something better. Thanks for the heads up. Does it always report that score or is it normal most of the time?

Here's three rounds of XS Bench run simultaneously on the Quad. It works most the time.