Sign in with
Sign up | Sign in
Your question

Have you calibrated your Core Temps?

Last response: in Overclocking
Share
October 24, 2009 5:02:24 PM

I've calibrated my CPU and core temps using this forums:Sticky: Core i7 and Core 2 Temperature Guide Core i7 and Core 2 Temperature Guide and with the help of CompuTronix (the author). This guide shows how to calibrate CPU and Core within SpeedFan to correct for inaccuracies in the temperatures reported by the sensors.

It would seem to me that since, as he claims "Factory Calibrations from Intel ... are frequently inaccurate," then when we post our temps on the forum, they are inaccurate (if we haven't calibrated). And we may be running cooler or hotter than we think.

Using his guide, my results were that my Core temps (measured by RealTemp) read 4c higher than actual. Instead of Prime95 running cores at 79c (according to RealTemp), I was really running at 75c using this calibration. Clearly I am much happier with 75c. At 79c I was debating whether to reseat my Noctua NH-U12P heatsink or whether I perhaps needed a better cooler. At 75c, I am satisfied with my temps for a 4.0GHz overclock on my i7 920.

I'm curious to know how many of you have used this well written guide to calibrate your temps and what do you think about it? What did you find when you recalculated - were your temps off and how much?

My final adjustments for SpeedFan:
CPU: -4 (under Winbond... Temp 2 offset)
Core 0: 11 (under Intel Core... Temp 1 offset)
Core 1: 13
Core 2: 11
Core 3: 15

Note: The calibration I did in this test is independent of the known issue that SpeedFan is 15c off on its core temp readings at default settings (for me) and as you can see corrects for this.

More about : calibrated core temps

a b K Overclocking
October 24, 2009 5:48:51 PM

What's thy vcore? Imo, 4Ghz i7 is pretty hard on air coolers.
m
0
l
October 24, 2009 6:29:12 PM

vcore = 1.25v
vtt = 1.25
PLL= 1.88
Dram Bus = 1.64
All other voltages = Auto
CPU & PCIE Spread Spectrum = Disabled
SpeedStep = Disabled
C1E = Enabled (I like a little power savings)

m
0
l
Related resources
a b K Overclocking
October 24, 2009 8:48:02 PM

peacemaker- said:
Note: The calibration I did in this test is independent of the known issue that SpeedFan is 15c off on its core temp readings at default settings (for me) and as you can see corrects for this.
Not necessarilly so. Have you upgraded yet to 4.39? When you uninstall 4.38, it will ask you if you wish to retain your settings, so the upgrade to 4.39 is very quick and easy.

As you were running SpeedFan 4.38 when you calibrated, the current version, 4.39, corrects this issue in most instances. Enthusiasts who are running the new i7 8xx or i5 7xx may encounter - 15c core offsets, which should be corrected in 4.40.

Regardless, it's a simple matter to take a moment and compare SpeedFan's core temperatures with Real Temp. Be sure to run each utility independently, and not simultaneously.

Comp :sol: 
m
0
l
October 24, 2009 9:23:29 PM

Hello Comp, thanks for replying.

I didn't uninstall first. I just installed 4.39, and it left all my settings / offsets intact. The temp readings didn't change and so I interpreted this to mean that 4.39 didn't fix the 15c offset issue.

When I tested / calibrated, I ran RealTemp and SpeedFan independently. I run them both together for my normal daily use. SpeedFan only allows one temp in the system tray and so I use it to monitor my MB temp. I use realtemp to put core temps in my system tray (and mentally subtract 4c). What happens if you run both at the same time?
m
0
l
a b K Overclocking
October 24, 2009 9:39:03 PM

On certain platforms they can interfere with one another.

Also, if you read the Real Temp documentation - http://www.techpowerup.com/realtemp/docs.php - and click on Real Temp's "Settings" button, you'll see that Real Temp can calibrate the core temperatures in two ways; at load and at idle. The significance of idle calibration is to correct for slope and linearity problems. It's the only core temperature monitoring utility which provides this level of precise calibrations.

I think it's absolutely the best freeware monitoring utility for core temperatures. I simply prefer SpeedFan because it also monitors CPU temperature, as well as so much more, not to mention the "Charts", which I find so useful for visualizing loads and temperatures, as well as Vcore and Vdroop. So with all these capabilities, SpeedFan provides a "bigger picture". I run SpeedFan configured for Core 0 to "Show in tray", since during single threaded applications, it carries the heaviest loads.
m
0
l
October 24, 2009 10:52:28 PM

Thank you for pointing out that I can adjust RealTemp as well.

I installed SpeedFan 4.40 beta and indeed it solved the 15c offset problem.
m
0
l
a b K Overclocking
October 24, 2009 11:29:29 PM

Excellent! :D  Let me know how it goes. :sol: 
m
0
l
October 25, 2009 5:30:12 AM

Comp,
I decided to recalibrate using your method with SpeedFan 4.0.
Offsets were:
CPU = -2
Core 0 = -1
Core 1 = +1
Core 2 = 0
Core 3 = +3

Back at 4.0 Ghz both idle and load core temps are 10c higher than CPU temp.
CPU / Cores
Idle 27c / 37-37-38-38
Load 65c / 75-74-75-74c

The only maddening thing is that Real Temp doesn't seem to work the same way. It seems that Real Temp uses "Idle Calibration" to set idle offsets and "Set TJ Max" to adjust the load temps. But raising TJ Max for a core also raises that cores Temperature at idle. Since core 0 perfect at idle but it was 3 degrees higher than the other cores at load so TJ Max gets set to 103 -- which raises the reading at idle by 3 degrees. This becomes circular and I don't get it.

I like your method with SpeedFan. I like that temperature variation is reasonably consistent from idle to load - and hope this is the more accurate reading.
m
0
l
a b K Overclocking
October 26, 2009 9:23:03 PM

peacemaker- said:
Comp,
I decided to recalibrate using your method with SpeedFan 4.0.

Back at 4.0 Ghz both idle and load core temps are 10c higher than CPU temp.
CPU / Cores
Idle 27c / 37-37-38-38
Load 65c / 75-74-75-74c
When Vcore and frequency settings are changed in BIOS from auto or stock values to overclocked values, this will cause certain motherboards to change the CPU temperature. This means that under the Test Setup, although you calibrated a 5c gradient between Tcase load and Tjunction load during Prime95 Samll FFT's, when you restored your BIOS to overclock settings, your CPU temperature changed, so that you no longer have a 5c gradient during Prime95 Samll FFT's.

Since the Digital Thermal Sensors (DTS) can't cause the Core temperatures or the 5c gradient to change, the solution is to simply re-calibrate your CPU temperature for a 5c gradient between Tcase load and Tjunction load during Prime95 Samll FFT's, so that your overclocked gradient is again correct. As long as you don't change BIOS settings, your CPU temperature and 5c gradient should remain consistent.

peacemaker- said:
The only maddening thing is that Real Temp doesn't seem to work the same way. It seems that Real Temp uses "Idle Calibration" to set idle offsets and "Set TJ Max" to adjust the load temps. But raising TJ Max for a core also raises that cores Temperature at idle. Since core 0 perfect at idle but it was 3 degrees higher than the other cores at load so TJ Max gets set to 103 -- which raises the reading at idle by 3 degrees. This becomes circular and I don't get it.

I like your method with SpeedFan. I like that temperature variation is reasonably consistent from idle to load - and hope this is the more accurate reading.
The proper method to calibrate Real Temp is to always begin with "Set TJ Max" follow by "Idle Calibration". Since there is some interaction involved, repeat the steps as necessary until you get it dialed in. Cores which are more offset than others will require more steps.

While you're calibrating, here's a valuable guideline to keep in mind:

Of the 27 dual core and quad core 65 nanometer and 45 nanometer Core 2 and Core i7 processor variants I've tested, if Tjunction Max is 100c, then nearly all of the Core 2 variants were actually 97c to 98c "mean" or average Tjunction Max value. However, all the Core i7 variants were very consistenly 100c +/- 0.5c mean Tjunction Max. So when calibrating individual cores for +/- offsets, it's hard to go wrong if your mean Tjunction Max value doesn't stray far from 100c. This is of course based upon 100% workload during Prime95 Small FFT's.

Comp :sol: 
m
0
l
October 29, 2009 2:19:50 AM

Comp,
I really appreciate your advice with RealTemp. Calibrating both programs each using a different method, I'm able to get a better feeling and confidence for what the real temps are. I really like how RealTemp allows for adjusting both idle temps and TJ Max. It seems to adjust the whole curve and be more accurate over the full range from idle to load. Here are my results.


Temperatures are AFTER the offset / adjustment applied
Load Temperatures after 10 minutes Prime95 Small FFTs
18c ambient (64*F)

SpeedFan
Idle
CPU: 29
Core: 32/35/36/35

Load:
CPU: 67
Core: 72/72/71/71

Adjustment
CPU: +4
Core: -2/+1/-1/+2


RealTemp

Idle: 35/35/35/34
Load: 72/72/71/72
Idle Calibration: +1.9/+0.3/-1.2/-1.0
TJ max: 97/101/99/103
m
0
l
a b K Overclocking
October 29, 2009 5:42:26 AM

peacemaker-,

Congratulations! :D  You've just graduated from Comp's Calibration Course! :bounce:  You have now demonstrated a clear understanding of how to calibrate your temperatures! Excellent job! :wahoo:  You've tweaked it right down to a gnat's ass! I couldn't do it any better myself! :whistle:  Your Diploma is in the mail! :sarcastic: 

Please feel free to assist me on the Forums with helping others to understand CPU temperature / Core temperature and calibrations! I know you'll always be a valuable asset here at Tom's!

Once again, well done! ;) 

Comp :sol: 
m
0
l
October 31, 2009 3:51:33 PM

OK so perhaps I get a bit anal about these things. I have a serious drive to understand what is going on and in particular with all the warnings about reducing the life of the cpu.

As it is I unwittingly ran my cpu for a couple of hours with vdimm Voltage on auto (probably at 1.9v) - although from what I have read this is no big deal. The warnings for above 1.65 vdimm however get my attention.

And I am on my 2nd Asus board with the first one having crapped out after 6 months. I'd like the cpu to stick around for a few years.
m
0
l
a b K Overclocking
November 1, 2009 1:51:14 AM

Asus P6T BIOS does not allow auto Vdimm to exceed Intel's 1.65 volt specification, regardless of whatever memory is plugged into the slots, so rest easy ... your rig is OK.
m
0
l
November 1, 2009 6:37:23 AM

I calibrated mine using the core 2 temp guide sticky.

When I calibrated the Tcase, I made sure:

-The Vcore was 1.1000v (actually 1.0754 in CPU-Z)
-The CPU Speed was 1.6GHz (267FSB x6 = 1602MHz)
-The system was idle for 10 minutes (actually ~15 min.)

The ambient temperature 6 inches from the front intake case fan was 22c
My cooler is an Arctic cooling freezer 7 pro and the CPU is a C2D E6750 (65nm/8w Idle) Using the formula Z+Ambient=Tcase - where "Z" is (cooler scale + CPU idle Power)/2:

(4+2)/2=3+22=25 (Tcase, right?)

With the minimum tcase calibrated to 25, I set everything back to stock which gave me:

-1.3375 Vcore (actually 1.296 in CPU-Z)
-2.66GHz CPU Speed

I ran Prime95 Small FFTs (with round-off checking) for 10 minutes+ and I calibrated my Core (Tjunction) temps to Tcase+5 (Tcase load was 37c so Tjunction was offset -6 in speedfan to give me 42c) It seems my Tjmax is 94c. Speedfan defaults to 100c as well as realtemp.

When I finished, I reapplied my previous stable OC (3.2GHz 400FSB x8; I can go higher but I see no need)

Did I do everything correctly? Because right now my Tcase is reading 23c. The ambient temp is ~20c. The Tjunction idles at 27c right now. Load temps with ambient 20c (Prime95 Small FFTs: Tcase: 37c/Tjunction: 42c. I used the Intel Burn Test and I got Tcase: 45c/Tjunction: 50c. I'm happy, but these seem a tad to good to be true. Any advice?

m
0
l
a b K Overclocking
November 1, 2009 8:04:11 AM

broketechjunkie,

Very astute observation. You apparently followed the Test Setup, yet it seems your calibrations are - 2c.

As I mentioned above, when Vcore and frequency settings are changed in BIOS, this can cause certain motherboard and super I/O chip combinations to skew the CPU temperature up to several degrees +/- in BIOS. This means that under the Test Setup, although you calibrated a 5c gradient between Tcase load and Tjunction load during Prime95 Samll FFT's, when you restored your BIOS settings, your CPU temperature shifted, so that you no longer have a 5c gradient during Prime95 Samll FFT's.

Since the Digital Thermal Sensors (DTS) typically can't cause the Core temperatures or the 5c gradient to change, the solution is to simply re-calibrate your CPU temperature for a 5c gradient between Tcase load and Tjunction load during Prime95 Samll FFT's, so that your 5c gradient is again correct. As long as you don't change BIOS settings, your CPU temperature and 5c gradient should remain consistent.

My own CPU temperature requires a -7c offset.

Comp :sol: 
m
0
l
November 1, 2009 1:34:03 PM

CompuTronix said:
broketechjunkie,

Very astute observation. You apparently followed the Test Setup, yet it seems your calibrations are - 2c.

As I mentioned above, when Vcore and frequency settings are changed in BIOS, this can cause certain motherboard and super I/O chip combinations to skew the CPU temperature up to several degrees +/- in BIOS. This means that under the Test Setup, although you calibrated a 5c gradient between Tcase load and Tjunction load during Prime95 Samll FFT's, when you restored your BIOS settings, your CPU temperature shifted, so that you no longer have a 5c gradient during Prime95 Samll FFT's.

Since the Digital Thermal Sensors (DTS) typically can't cause the Core temperatures or the 5c gradient to change, the solution is to simply re-calibrate your CPU temperature for a 5c gradient between Tcase load and Tjunction load during Prime95 Samll FFT's, so that your 5c gradient is again correct. As long as you don't change BIOS settings, your CPU temperature and 5c gradient should remain consistent.

My own CPU temperature requires a -7c offset.

Comp :sol: 

My load tcase to tjunction gradient is exactly 5c with this overclock (42(Tjunction load) - 37(Tcase load) = 5 (Gradient) Even the difference between Tcase and Tjunction at Idle is correct according to the guide. (4c)

Unless I'm misinterpreting your post, you're saying that the gradient between tcase and tjunction at load is ALWAYS 5c. That's exactly what I'm experiencing. Even using the Intel Burn Test (Which got my cpu 5c hotter than Prime95) the difference (50(Tj)-45(Tc)=5)

Stock settings for this CPU (2.66GHz/1.296Vcore) gives me the same exact temp gradient as this overclock (3.2GHz/ 1.248Vcore)

m
0
l
November 1, 2009 3:50:19 PM

Yes I see the consistent 5c difference and that is great. The only thing I question is :

"Tcase is reading 23c. The ambient temp is ~20c." That is the same 3c difference as it was during the Test Setup at 1.1v and I would expect more like a 6c difference.

So your CPU temp might be set a couple of degrees too low - what was the SpeedFan offeset you entered? Did you do the Test Setup with the doors open?

If you did the calibration correctly, more likely you have great temps and you can believe it.

m
0
l
November 1, 2009 4:36:13 PM

Both side panels were removed during the Tcase calibration. I have a 120mm intake fan on the side and the case has very good airflow (NZXT Apollo. 1 side intake, 1 front intake, 1 rear exhaust. All120mm 1500rpm fans. 57.67CFM Airflow):

http://www.newegg.com/Product/Product.aspx?Item=N82E168...

The Tcase temp was already correct at 25 in speedfan during Tcase calibration. I didn't need to offset it.
m
0
l
November 1, 2009 5:41:00 PM

broketechjunkie said:
Both side panels were removed during the Tcase calibration. I have a 120mm intake fan on the side and the case has very good airflow (NZXT Apollo. 1 side intake, 1 front intake, 1 rear exhaust. All120mm 1500rpm fans. 57.67CFM Airflow):

http://www.newegg.com/Product/Product.aspx?Item=N82E168...

The Tcase temp was already correct at 25 in speedfan during Tcase calibration. I didn't need to offset it.


What adjustments did you have to enter for SpeedFan and RealTemp? I found SpeedFan 4.4 and RealTemp to be pretty close and the average of all adjustments for each setting was 0.

Sounds like you've got some good temps.
m
0
l
November 1, 2009 6:00:44 PM

The Tjmax is 94c for my E6750 if these temps are accurate. Realtemp and speedfan use a Tjmax of 100c so I offset the core temps -6c Also, both cores vary about 1c in temps. Is this good? I've seen people with core temps varying 3-6c between cores. Roght now my temps are 25/29-29. I played crysis for two hours and it got to 34/39-39! My gpu temp never went above 42c with the Arctic Cooling Accelero/92mm fan! :sol: 

I just did a realtemp sensor movement test and I got movement of 5 for both cores. the closest the temps came to Tjmax were 50 (Tjmax = 94) That means the highest temp was 44! :bounce: 

m
0
l
November 4, 2009 3:48:27 AM

CompuTronix said:
Asus P6T BIOS does not allow auto Vdimm to exceed Intel's 1.65 volt specification, regardless of whatever memory is plugged into the slots, so rest easy ... your rig is OK.


Here are some posts I have read:

Quote:
A couple of days later I installed TurboV and it showed my DRAM Bus Voltage as being at 1.9v, when the manual says that the voltages over 1.65v may permanently damage my CPU according to the Intel spec. If that is the case, why does [Auto] seem to risk damage to my CPU? How much damage will actually have been done by this?


Quote:
I've seen the P6T set 1.9V for the memory voltage while still at stock BCLK. It's best to manually enter VTT and memory voltages.


Quote:
... and the voltage at auto. I just realized after having my computer on for about 2 hours that the RAM voltage was 1.80v the whole time.

etc...

Yes my rig is ok - it was only a couple of hours. But it is amazing that Asus takes all the trouble to put a yellow sticker on the memory slots warning what may happen over 1.65v and then automatically sets it to 1.8 or 1.9. After reading the following post, I calmed down: Here.
m
0
l
a b K Overclocking
November 4, 2009 4:24:32 AM

Thanks, I stand corrected. I was under the impression that recent BIOS releases prevented destructive Vdimm settings. I'll look into this further.
m
0
l
March 4, 2010 1:36:04 PM

not to resurrect an old thread for no damn good reason, but the "how to calibrate a c2d (OR) i7 for speedfan" articles are all now nonexistent. you click on the link at the top of this page and it says "page does not exist," and the same for the one on the c2d wherever i last found that link. is there a reason for this? logically i would assume that means speedfan may have fixed the problem in a recent release? however speedfan shows 25c at idle for my cores while real temp shows 32c, so obviously there is a descrepancy somewhere.

again sorry for bringing back an old thread, but any help is appreciated!
m
0
l
a b K Overclocking
March 5, 2010 1:05:30 AM

^ CompuTronix, the author of the guide took it down for personal reasons.
m
0
l
!