What is the source of 5.Findings in the Core2 Temp Guide?

[Mondoman]

Specifically, I'm interested in Finding (A) from the Guide:

Section 5: Findings

(A) Depending upon Stepping, the Delta between Tcase and Tjunction is 10c +/- 3 (or) 15c +/- 3.

Are the delta values experimental or theoretical? How were they determined? Does the delta vary:
1) As Tcase varies?
2) Among different CPUs within the same stepping (e.g. between M0 CPUs e2160 and e2200, or e2160 and e4600)?
3) With different CPU coolers?
4) With different motherboards?
5) As CPU core speed varies? (If so, this would imply that (2) is also "yes")
6) As CPU Vcore is varied? (If so, this would imply that the delta is different between two otherwise identical CPUs that have different VIDs)

 

[CompuTronix]

Mondoman, it's always a pleasure to see a valued Forum contributor such as yourself, and thanks for asking some excellent questions. The Delta values were acquired from empirical data collected from multiple combinations of motherboards and Core 2 processor variants using stock to high-end coolers.

Yes, you're quite correct; the Delta does vary, as detailed in the Guide:


"Section 6: Scale

Although the Delta between Tcase and Tjunction is constant, these temperatures do not always scale in a precisely linear manner with respect to one another, due to Variables such as Vcore, clock speed, Stepping and Load. Idle at very low Vcore and clock on G0 Stepping may cause Tcase to Tjunction Delta hottest Core to indicate less than 7c, while 100% Workload at very high Vcore and overclock on B2 Stepping at may cause Tcase to Tjunction Delta hottest Core to exceed 18c."


Coolers will affect Deltas, but within the constraints described above. Motherboards will also affect Deltas due to erroneous BIOS versions, which directly affects Tcase (CPU temperature), but have no affect on Tjunction (Core temperatures).

Additionally, since no two processors have identical dynamic operational characteristics such as overclocking ceiling, Vcore tolerance or thermal behavior, there are also mean deviations in Deltas among variants of the same Stepping Revisions. However, the majority of processors will typically fall within their respective +/- 3c tolerance.

If you have any suggestions, I'm always interested in improving the Guide. Thanks again for your questions,

Comp

 

[Mondoman]

Thanks! I guess what initially puzzled me was "Although the Delta between Tcase and Tjunction is constant..."; perhaps taking that out of the Guide or changing it to something like "Although the Delta between between Tcase and Tjunction is generally in the range described in section 5 above, ..." would help clarify things.

 

[Mondoman]

Comp, I seem to be getting Tcase/Tjunction deltas that are substantially smaller than the 10-15C values. As I test different systems, I will add posts to this thread.

System 1:
Asus F3Jp notebook running 32-bit Vista Home Premium
T7200 CPU (2GHz core) B2 stepping VID: 1.225V
3GB RAM
Tcase measurement software: Notebook Hardware Control 2.0 pre-release 6, CoreTemp 0.96.1
Tjunction measurement software: NHC, CoreTemp as above
Ambient temp: 20C

At idle (0-5% load)
Tcase: 50C
Tj1: 54C
Tj2: 54C

At load (100% load via Orthos)
Tcase: 84C
Tj1: 85C
Tj2: 85C

 

[red hook]

Thank you both very much!

Comp, as a first-time oc'er i found it very useful to read your article before trying anything, then trying an oc, then rereading parts of your article, then trying to modify my oc, then rereading...etc.

Mondoman, I had the same question as you re Tdelta value, as I get these temps in core temp, speed fan and cpu hardware monitor (within 1 degree C), running my e4600 @ 3.3, vcore at 1.41875 (Bios), 1.376 (CPU-Z):

idle: 22/25-28
load: 57/60-62

I was wondering if the apparently small Tdelta should be worrisome, but now feel OK about it.

 

[Dunkel]

Hopefully CompuTronix will correct me if I'm wrong.

@red hook - If you haven't calibrated speedfan and/or you're not using the right version of CPU-Z with the correct Tjunction max, your temps might actually be load: 72/75-77.

I finished the whole guide before I started OCing. It really helped.

 

[red hook]

Thank you, Dunkel. Re the correct version of CPU-Z, my reading of various threads suggests that perhaps you mean the correct version of Core Temp, which reports Tjunction max as 85 rather than 100. I have version 0.96.1 of Core Temp, and version 1.43 of CPU-Z (which doesn't report Tjunction max).

You do bring up an issue that I don't feel I know the answer to, even though I have read many posts on the subject: the question of the need for offsets in Speedfan (and Core Temp?) for C2D chips. As I reported above, Core Temp, Speedfan and CPU hardware monitor all give me consistent temps at idle and load. So are they all off by 15, or are they all correct? I also note that in Graysky's article on oc'ing, he says that there is a need to offset 15 in Speedfan, but for C2Q's only. The definitive answer (if there is one) is important to everyone. For me, I was just trying for 3.4 on my oc, and load temps were 69. Acceptable for a stability test, but if that's actually 84, I want to back off!

 

[Mondoman]

System 2:
Abit NI8-SLI running 32-bit Win XP Pro
Pentium D 945 CPU (3.4GHz core) C1 stepping 1.23V / Freezer 7 Pro CPU cooler
3GB RAM
Tcase measurement software: Abit uGuru chip/"clock" custom hardware reader
Tjunction measurement software: none, no Tj sensor on Pentium D
ambient temp: 24C

At idle (0-5% load)
"system" sensor: 27C
Tcase: 30C

At load (100% load via Orthos)
"system" sensor: 28C
Tcase: 52C

 

[CompuTronix]

red hook, regarding SpeedFan 4.33, a few of graysky's temperature explanations are slightly misleading and somewhat less than informed, no disrespect intended toward my next-door Sticky neighbor. The temperature issue that you describe is mentioned in my Temperature Guide no less than 3 times:


Section 8: Tools

Note 3: SpeedFan... CPU`s with Steppings which are Tjunction Max 100c typically require +15c Core Offsets. See Section 11.


Section 9: Calibrations

Part 4: Tjunction Offset Correction

Note: CPU`s with Steppings which are Tjunction Max 100c typically require +15c Core Offsets. See Section 11.


Section 15: Troubleshooting

(G) SpeedFan 4.33 typically requires +15c Core Offsets for CPU`s with Tjunction Max 100c Steppings.


There are no special or qualifying considerations that differentiate Quad's from Duo's. The issue is confined to processors which are Tjunction Max 100c only. SpeedFan 4.34 BETA 38 correctly interprets Tjunction Max 85c processors as well as Tjunction Max 100c processors.

Comp

 

[red hook]

Comp: as I mentioned above, I read your Guide more than once and found it very informative and helpful. With each reading I get something more, so thank you again.

My question perhaps shows my ignorance, but here it is: why do SpeedFan, CoreTemp and other apps, such as CPU Hardware Monitor, all give me consistent results (without offsets)? I read your offset advice as relating specifically to an anomoly in SpeedFan 4.33. Is this then equally applicable to other apps using the same sensors?

Interestingly, and more confusing to me given your post, I downloaded SpeedFan 4.34 BETA 38 yesterday, and got the same readings as with SpeedFan 4.33!

So perhaps the anomoly is with my particular rig, or of course I may just be going crazy (as some have suggested.)

 

[Dunkel]

Don't feel bad. Have you seen my thread on SpeedFan questions?

I would have given a bit of cheese to see my various apps. post equal temps.

 

[red hook]

HaHa Dunkel, I have! Thanks.

UREKA!!!!!!!!!!!!!!????

Just read the latest posts to my vcore and vdroop thread and see what might be the answer, for me. My e4600 has MO stepping and according to Comp this may mean 85 Tjmax, not 100 as previously thought, ans so no offset would be required in SpeedFan, and consistency among all apps is explained!

 

[Dunkel]

By that rational, I have the same issue with M0stepping in CoreTemp. My CoreTemp matches SpeedFan!
That's why I had so much trouble.

What is the other app you are using for reading temps? If I can get a third opinion on my temps I will be very happy.

I'll still go through the motions of calibrating again, but I surely won't stress over it.

 

[red hook]

It's called CPU Hardware Monitor, and you can get it at the same site as CPU-Z (they have a number of apps available.)

I would appreciate your post if you calibrate again as it hopefully will verify our answer.

 

[Dunkel]

@redhook

Note: From the temp guide, Comp states that Tjunction is always higher than Tcase.

I reset my SpeedFan to default and my Tcase (Temp2) is 4C higher then my Tjunctions (Core0 and Core1).

So in my case, I still need to re-calibrate. I'm not doing anything until Sat.

 

[red hook]

ooops, I meant Comp's latest post to your SpeedFan questions thread.

 

[Mondoman]

Comp: nice job on the very recent update to the Guide!

 

[CompuTronix]

Thank you Mondoman.

My appologies to everyone in our O/C and Temp community for not sooner providing this very needed February 7th update to the Temp Guide. The M0 processors have been a major source of controvery, since Intel shuffled the deck several months ago. During this time, "some" M0 processors appeared in the global install-base which did not thermally conform to other variants.

Intel does not provide Tjunction Max specifications for desktops, so I chose to err on the side of safety by using 100c in the Guide, until such time that 85c could be positively verified. Over the past many weeks, I have discussed this issue with Arthur Liberman, the author of Core Temp, who confirmed that "some" M0's were being detected as 100c, while "some" M0's were being detected as 85c, not just by Core Temp, but also by an assortment of other utilities.

It is suspected that "some" M0's are actually recycled E6xx0 variants which have half their L2 cache disabled due to fabrication defects, and are repackaged rather than being scrapped. These supect units, which popped up soon after the release of the M0's, obviously did not have correct factory DTS calibrations, which added to everyone's confusion. It's also likely that "some" E2xx0 units are repackaged E4x00 units. The question of counterfeit processors was also raised.

This factory DTS calibration issue is evidenced by the number of threads which detailed Core temperatures below ambient, as well as Idle and Load temperatures well above other variants, and at the same Vcore and clock. If you recall, we went through a somewhat similar Tjunction Max mess over a year ago when the L2's were released, which increased Tjunction Max from 85c to 100c, and made no sense at the time. And now with the M0's reverting back to 85c, this underscores the need for Intel to provide vital temperature information to it's customers by including Tjunction Max in their specifications.

AMD is no less secretive than Intel, but this Tjunction Max mess reminds me of "Spy Vs Spy". Intel must wake up and realize that we're not constrained to rely strictly upon monitoring Tcase (CPU temp), which is their only published thermal specification for desktops. Since we've all been monitoring Tjunction (Core temps) for quite some time anyway, it's long overdue for Intel to give a little more consideration to their customers. As long as Intel rediculously continues to regard Tjunction Max and Core temperature specifications as corporate secrets, I will feel obligated to continue to maintain the Temp Guide, and to do my best to provide some insight and answers for our readers here on Tom's Forums.

Thanks again Mondoman,

Comp

 

[Dunkel]

I'd like some credit too. At least I knew something wasn't right. :lol:

 

[CompuTronix]

Dunkel, you're doing a great job with the O/C and Temp learning curves. It's a pleasure to have you here at Tom's.

 

[Mondoman]

System 3:
Gigabyte GA-P35-DS3P V2.0 running 32-bit Win XP Pro
Pentium Dual-core e2160 CPU M0 stepping, VID 1.325V / stock Intel CPU cooler
Overclocked to 3GHz core/333MHz FSB (CPU voltage set to 1.4V in BIOS, read as 1.36V in CPU-Z V1.43)
2GB RAM
Tcase measurement software: Gigabyte EasyTune5
Tjunction measurement software: CoreTemp 0.96.1
ambient temp: 20C

At idle (0-5% load)
"system" sensor: 31C
Tcase: 21C
Tj1: 23C
Tj2: 25C

At load (100% load via Orthos "blend")
"system" sensor: 32C
Tcase: 58C
Tj1: 62C
Tj2: 63C

 

[CompuTronix]

Mondoman, your temperatures for "System 3" are correct, however, as I'm sure you're aware, "Blend" doesn't provide a steady or continuous 100% workload, although it does yield a more realistic representation of real world maximum loads, which is probably why you chose that particulat test. If we watch the temperatures, or more ideally, if we watch SpeedFan "Charts", then we can observe a graphical representation of the thermal fluctuation which result from load fluctuations during "Blend". Small FFT's will instead give us a very smooth and consistent 100% workload while yielding the highest temperatures, which again, can be observed in SpeedFan "Charts". Heat saturation typically occurs within 7 to 8 minutes, so a test duration of 10 minutes is adequate.

 

[Mondoman]

Thanks for your comments, Comp! Yes, I use the Blend to get a roughly real-life stress of CPU+RAM, and let it run 20-30 minutes. I often also try running the small FFT for 3-5 minutes just to make sure the temps don't get alarmingly higher than Blend. As you point out, this set of tests is not meant to be complete or exhaustive, but rather allows me to make a relatively quick comparison between my different systems and new builds and make sure conditions aren't out of line.