Question regarding TjMax, CoreTemp, Speedfan

[Silver Fox]

From my understanding, the TjMax is a fixed value that doesn't change. Is there a way to find out definitely what the value is? Like CoreTemp 0.95 assumes it's 100C and SpeedFan 4.32 assumes it's 85C. How can I tell who is actually correct? Do all CPUs share the same TjMax? I can read the Delta value just fine, but without knowing the proper TjMax, it's kinda hard to figure out exactly what the true temperature is.

This is mainly concerning the Q6600 B3.

[Silver Fox]

Bump, anyone?

[chavalcito]

I subscribed but don't have the exact answer, sorry. Core Temp says my Tjunction is 85C. That is much different than yours. I haven't run speedfan.

Where do you even see the Tjunction in SpeedFan? I couldn't see it. My CPU temps in SpeedFan were accurate according to TAT and Core Temp, also.

[Silver Fox]

You can't see it, but you can calculate the value by reading the Delta value directly using CPUID (0x19C). If you add that value with the temp you're reading from SpeedFan, you'll see that it adds up to 85. Adding the delta with CoreTemp's, you get 100.

[Bail_w]

From what i known so far about coretemp and speedfan, either one of them can be true.

[Ad1tya]

Q6600 B3 is 100C TJunction.

Speedfan, reads it as 85C, so its very wrong. 15C Lower than your actual temp's.

Although, most people are unsure about the Q6600 G0, which may just be 110C TJunction

[unclewebb]

Ad1tya: Q6600 B3 is 100C TJunction.

Can you post some documentation from Intel to back that up. From what I've read, TjMax is not documented by Intel for any of the desktop processors. It might be 85C or it might be 100C or it might be some number completely different.

Temperature monitoring software like CoreTemp and SpeedFan guess at a TjMax which is then used to calculate an absolute core temperature but it is not documented to be accurate.

Here's the forumula they use.
reported core temp = TjMax - DTS

where DTS is the reading directly from the digital thermal sensors built into all Core based processors. If software guesses wrong at TjMax then the reported core temperature is just a number.

There is no Intel documented way to determine the true TjMax of a processor. TAT came out long before the Core desktop processors did and hasn't been updated since. It is a great laptop baking utility but I wouldn't trust it for temperature measurement.

Best bet is to go with CoreTemp 0.95 and use it to report the DTS directly.

[Ad1tya]

Intel does not specify the TJunction of their Desktop processors.

They only do of their Mobile and Server like.

X3220, was 100C TJunction, thus we believe the Q6600 is also 100C.

[Spanki]

See uncleweb's excellent explanation above ^^ but to answer your questions directly:

Is there a way to find out definitely what the value is?

No. Intel has not disclosed this information.

Like CoreTemp 0.95 assumes it's 100C and SpeedFan 4.32 assumes it's 85C. How can I tell who is actually correct?

You can't.

Do all CPUs share the same TjMax?

No. While we don't know what the numbers actually are, we can infer that they are at least different between some processor types (C2D vs C2Q, for example).

I can read the Delta value just fine, but without knowing the proper TjMax, it's kinda hard to figure out exactly what the true temperature is.

...yep! But don't feel pregnant - currently noone else knows either. For conversational/discussion purposes, it's probably best to use CoreTemp 0.95 numbers, since they are probably closest to being correct (in the right ballpark - or at the very least, a common point of reference for others).

But, as uncleweb suggests above, you can basically forget about absolute temperature numbers alltogether - these cpus are designed to take care of themselves if things get out of hand... by either throttling back or shutting down completely if it gets serious, but your system will become 'unstable' well before either of those things happen.

So, basically, as long as:

a. Your system is stable at some given settings and...
b. You're not juicing the cpu with outrageous/dangerous/life-shorteningly high voltages

...you have nothing to worry about. I personally wouldn't run my system 24/7 @1.6v+, because regardless of the temperatures insuing from that, the voltage itself will decrease the life-span of your parts. It might be fine for benching or achieving some new personal goals, but there's usually not much real reason to run clocked to the edge for every day operation.

Anyway, the only real meaningful/valid temperature-related info you can get currently is by setting the "Show Delta to Tjunction" option in CoreTemp 0.95...



...that will show you the DTS value, which is how much "temperature headroom" you have left (it counts down as temperatures rise).

[astaris]

About the Tjunction issue, The Coolest, author of Core temp, wrote:

Regarding the Tjunction... Well I wish I had a friend at Intel, heh. But I do have a friend who does, so he sends me some interesting bits of information, also I spoke to someone who decompiled TAT and found out that it was looking at the same MSR as I am and checking it for 85C or 100C.

Here the original post.

[Silver Fox]

Thanks for all the replies. I read the delta value directly using CPUID (46 on idle), so I knew I was fine. On full load however, it can decrease as much as to 26. My B3 Q6600 is running @ 3.2Ghz 1.36v (457x7) w/ Thermalright Ultra 120 Extreme w/ Scythe "F" 120mm fan on a P5K Deluxe, fully prime blend stable.

It's really annoying I have to reboot and disable driver signing just to get coretemp/CPUID working though (Vista 64). This is why I prefer to use SpeedFan most of the time as I can run it without F8ing.

[boogle]

In SpeedFan you can offset the sensors to get a 'correct' reading. Since it's widely believed that Speedfan is off by -15c, you need to offset all the core temp readings by 15c.

Click on 'Configure', then the 'Advanced' tab. Select 'INTEL CORE' from the dropdown list, now select each temperature in turn and choose an offset of 15. Tell Speedfan to remember it.



Your temps will now display correctly

[Monstru]

E6600 704A - Coretemp Tjunction = 85 Celsius....

[Spanki]

About the Tjunction issue, The Coolest, author of Core temp, wrote:

Regarding the Tjunction... Well I wish I had a friend at Intel, heh. But I do have a friend who does, so he sends me some interesting bits of information, also I spoke to someone who decompiled TAT and found out that it was looking at the same MSR as I am and checking it for 85C or 100C.

Here the original post.

Interesting... and I also went and read his article on overclockers.com, but I think his quote above is misleading, as well as his article. He says that there is a MSR that can be read on Intel cpus to "checking it for 85C or 100C", but I don't think it's actually telling him the number value to use... because if it was, there wouldn't have been a problem with Allendale cpus between CoreTemp 0.94 and CoreTemp 0.95. In other words, he might be reading some MSR and based on the value he finds there (which is somehow relative to the cpu in use), he just picks one value or the other (either 85C or 100C, but both are 'hard-coded', not 'read' from the MSR).

Also note that he talks in terms of "this, or that" (one value, or the other), and not "read the Tjunction value from xxx MSR". Again, this leads me to believe that the prrgram only chooses between two values, which may or may not be correct (the behaivior was based on how TAT did things, but TAT was written for Mobile cpus and hasn't been updated).

And finally, in his various release notes, he talks about 'recognizing' or adjusting for new Intel cpus.... if it was all register-based, he could just read the values - no need to add new deciphering code.

[bing]

Yep, you're right, as long your rig is "stable and not throttling", consider it is safe, ignore those temp reading.

With current publicly and commercial Intel CPU, no one in this earth can read the cpu temp accurately using software "ONLY" method, yes no one, not even Intel engineer. Yeah, I know, you're now yelling wtf, I repeat not even Intel CPU engineers.

Just read Intel's own technical whitepapers at their site on how they measure the CPU temp accurately, trust me, you will end up with this method below using "calibrated" temp probe if you really want to know the CPU temp .... again ... "accurately".

8052.jpg
Courtesy of www.overclockers.com

[astaris]

...
Also note that he talks in terms of "this, or that" (one value, or the other), and not "read the Tjunction value from xxx MSR". Again, this leads me to believe that the prrgram only chooses between two values, which may or may not be correct (the behaivior was based on how TAT did things, but TAT was written for Mobile cpus and hasn't been updated).
....

I know, we atm are not able to find the Tjmax for these cpu's, but at least we know that there is a bit in an unknown MSR that tell us something about the Tjmax and that the Coolest has his sources...better than nothing

[unclewebb]

I know, we atm are not able to find the Tjmax for these cpu's, but at least we know that there is a bit in an unknown MSR that tell us something about the Tjmax and that the Coolest has his sources...better than nothing

That is absolutely WRONG!

There is a bit in the mobile processors that can be read which will tell you if TjMax is 85C or 100C but a representative from Intel has already publicly stated that reading that bit in desktop processors and jumping to a conclusion about TjMax is not valid.

Check the last reply before Intel closed this thread.
http://softwarecommunity.intel.com/i...howThread.aspx

"Use of bit 30 in MSR 0EEh is *not* valid for desktop, workstation or server processors based on the Intel(R) Core(TM) microarchitecture. However, for mobile processors, this assumption *is* valid. "

[astaris]

That is absolutely WRONG!

There is a bit in the mobile processors that can be read which will tell you if TjMax is 85C or 100C but a representative from Intel has already publicly stated that reading that bit in desktop processors and jumping to a conclusion about TjMax is not valid.

Check the last reply before Intel closed this thread.
http://softwarecommunity.intel.com/i...howThread.aspx

"Use of bit 30 in MSR 0EEh is *not* valid for desktop, workstation or server processors based on the Intel(R) Core(TM) microarchitecture. However, for mobile processors, this assumption *is* valid. "

Thanx for the link, really good information.
But on the same thread, we have:
question:

However, there is a problem. That reading although available in °C is _relative_ to the Tjunction which is CPU specific.
For some models it is 100°C for others it is 85°C

answer (by an Intel engineer):

It is true, some are calibrated to 100C, some are calibrated to 85C.
We should call that value as Tjmax for Conroe/Woodcrest/Clovertown instead of Tjunction.

Well, for what i know Conroe is not a mobile procesor...so we have 2 official versions in 2 pages of that thread.

[unclewebb]

The Intel rep Lexi S. went back and forth during that thread but the final paragraph contained his conclusion which I quoted previously:

"Use of bit 30 in MSR 0EEh is *not* valid for desktop, workstation or server processors based on the Intel(R) Core(TM) microarchitecture. However, for mobile processors, this assumption *is* valid. "

For the desktop processors, TjMax is 85C, 100C or ???C. It is not documented by Intel and there is no valid bit that software can read to determine what it is.

My personal opinion is that all Core 2 Duo desktop processors are exactly 85C. The mobile Core processors are all documented as 100C and I think the new revision G0 Q6600 is also 100C but as hard as I've looked, there isn't any official Intel documentation to back that opinion up.

[Silver Fox]

So what do you guys think? 85 or 100 to be more accurate? I personally feel that 85 might be closer than 100, as when I'm idling, it sits around 43C. If we go by 100, it should be 58C, which sounds really high for a processor that isn't doing anything.