Large gap between Tcase (cpu temp) and Tjunction (core temp)

peacemaker-

Distinguished
Jun 10, 2009
30
0
18,540
I am reading the temperature guide by Computronix because I have

Tjunction Load = Tcase Load + 13c (80c = 67c + 13c)
Tjunction Idle = Tcase Idle + 10c (37c = 27c+ 10c)

My ambient is currently 19c but the idle delta is true at 22c as well.

Based on the guide my take is that either Tjunction or Tcase is inaccurate because the difference should only be 5c. Do I understand correctly?

I have a p6t deluxe v2 and am currently overclocked to 4.0GHz using 21*191 at vcore = 1.25v and vtt = 1.25v. I am hitting 80c using Prime95 small FFT and this seems high with 1.25v vcore. I am trying to figure out if the problem is:
1) case (4 Noctua 120mm fans) - seems like this should be enough - the mb temp is 35c according to speedfan
2) cpu cooler (Noctua NH-U12P SE2 with dual fans) - need to reseat or fans not enough for the task
3) inaccurate temps
4) luck of the draw
5) Have I missed anything?

I have not yet run the guide's calibration. I want to make sure I fully understand what I am about to undertake, considering it is a fair amount of work. And I want to make sure it helps me solve the problem of 80c load.
 
peacemaker-,

Have you flashed your BIOS? Remember that it affects the accuracy of CPU temperature, but not Core temperature. It's not unusual for the 9xx D0 at 1.25 Vcore to reach Core temperatures of 80c with HT on. Please list the temperatures for all 4 cores in SpeedFan as well as Real Temp at 10 minutes of Prime95 Small FFT's. Ignore the virtual cores 4,5,6 & 7. Make sure you run each utility one at a time, and not simultaneously.
 
I have flashed my bios to the latest (0610). After 10 minutes of Small FFT, Realtemp showed 76-72-74-71. I then closed Realtemp and opened SpeedFan. It showed 76-73-74-71 with CPU at 65c. Ambient is at 67c.

 
BIOS accuracy for CPU temperature is seldom perfect. For a quick calibration in SpeedFan:

Configure > Advanced > Chip > Windbond Wx... > Temperature 2 offset > Set to + 3 > remember it

Configure > Advanced > Chip > INTEL CORE > Temperature 1 offset > Set to -3 > remember it
Configure > Advanced > Chip > INTEL CORE > Temperature 2 offset > Set to +1 > remember it
Configure > Advanced > Chip > INTEL CORE > Temperature 3 offset > Set to 0 > remember it
Configure > Advanced > Chip > INTEL CORE > Temperature 4 offset > Set to +2 > remember it

Since there are always cores which are higher or lower than their neighbors due to slope error, non-linearity and factory calibration deviations, these offsets leave the mean or average Tjunction value of 73.5 unchanged. These offsets are also very close to my own. Your temperatures should line up nicely as follows:

CPU = 68c
Core 0 = 73c
Core 1 = 73 to 74c
Core 2 = 74c
Core 3 = 73c

CPU / Core Gradient = 5c 😀

Run Prime95 Small FFT's and report back with your new calibrated values. We may still need to fine-tune it a bit.
 
While you were replying, I was running the calibration process. I found that:
At Ambient = 20.5c
X=4 (i7 chip)
Y=3 (for Noctua)
(X+Y)/2 = Z
(4+3)/2 = 3.5 + 20.5 = 24 (Tcase Idle)

At all settings in guide and after 10 minutes:
CPU Temp = 28c
Core Temps = 31-31-34-30
I went to Speedfan to configure offset correction for Tcase.

On Advanced tab, when I click on the Chip field, I get choices for:
Intel Core @ $0...
Winbond W83627DH6...
HD0...

When I open up Intel Core, I find that the Temperature 1 through 4 Offsets are set to 15. Is this the default value? Or did I set this value (probably long ago base on some forum poster's advice and have since forgotten it)?

Anyway, these calculations say the CPU Offset should be set to -4. I might as well finish the calibration as per your guide and see what it shows. Either way, it seems that you have helped me discover that my cores aren't as hot as I thought and at the acceptable max - great news! I will report back when my testing is done.

Perhaps you can help (or not) with my last angst...
While doing all my testing to get to 4.0GHz, I accidentally left my DRAM Bus Voltage on Auto. the p6t has the amazingly awful feature to raise it to dangerous levels (I've been told 1.8v or higher) when overclocking on Auto. I ran it this way during OCCT and Prime95 for up to 2 hours. I've read all about the possibility of degradation and damaging the CPU and I know no one can give me real info as to what I potentially did to my chip. Any advice other than, "It's done and there's nothing to do... so forget about it..."

 


I am running version 4.38. I followed the guide. Reviewing it, the only thing I didn't do was open the second door on the far side of the case (on the other side of the motherboard). Should I re-run the tests (bummer).

My test results
cpu: -4
core 0: +11 (it was at +15)
core 1: +13 (it was at +15)
core 2: +11 (it was at +15)
core 3: +15 (it was at +15)

When I ran the load tests at 1.25v and stock voltages I got:
cpu: 55 cores 60-60-60-60

Then I rebooted back into my 4.0GHz settings and ran again load for 10 minutes and I got
cpu: 66
cores: 75-74-74-74

When I went back to idle:
cpu: 28
cores: 35-36-37-37
 
You've found that when Vcore and frequency are changed, that BIOS shifts it's offset, which causes your Gradient to be greater than 5c. It's a common problem with these board. Simply run Prime95 Small FFT's again, and at 10 minutes, recalibrate your Windbond Wx... > Temperature 2 offset for a 5c Gradient. You might end up with a Temperature 2 offset of 0.

Also, version 4.39 will probably correct the +15c offset, so you'll need -4, -2, -4, 0.
 
I re-ran the 1st calibration (all stock settings) with both doors off and it confirmed that I needed a -4 offset for the CPU temp. I am at 22c ambient finally (the fireplace has finally warmed it up).

I then rebooted at 4.0GHz and ran under load as you suggested with the offsets as:
cpu: -4
Temperature 1: 11
Temperature 2: 13
Temperature 3: 11
Temperature 4: 15

Temperature Results:

CPU Temp: 69
Core Temps 75-74-74-74

At Idle at 4.0GHz I get:
CPU Temp: 32
Core Temps 36-37-38-38

Anyway, it seems right on. Do you agree?

Regarding other program core readings (after the offsets applied to Speedfan):
Realtemp now shows core temps about 4c higher than Speedfan
OCCT shows core temps 8c higher than Speedfan

I will now rely on Speedfan for my temps - nice to know they weren't as hot as Realtemp was reporting.

Thanks for your help.