TJunction and TCase

Toggle sidebar Toggle sidebar
P

Petrolman

Distinguished
Apr 24, 2009
20
0
18,510
I read over CompuTronix's article on temperatures for the dual and quad core processors and I had just a couple of questions to ask. They may look like statements but they are really questions.

First off, from what I understand the core temperature is mostly used by the processor to prevent damaging itself. The values are not, at least initially, useful values for the end user. I'm hoping I'm right so far because I just recently got a Q8200 and the core temperatures idle around 45-50 degrees, depending on the core, and hit around 75 degrees on the Prime95 stress test. In the meantime, the TCase value is around 22 degrees idle and 50 degrees during load. All of this is with a stock cooler and no overclocking.

If I go through CompuTronix's article the idle TCase including offset becomes 27 degrees and the load temp becomes 55 degrees. So, if I understood correctly, I should assume core temps of around 60 degrees? I'm not really planning on overclocking for a while but I just was kind of curious how my new computer was doing so far.

All of the TJunction sensors appear to respond okay to increasing temperatures but I'm not so certain about their minimum values. Do I really need to worry about this should I ever decide to overclock? From what I understand, the TJunction values are derivatives of the TCase value anyway. I think.

Anyway, I appreciate the help in advance.

PetrolMan
 
PetrolMan.

If your measured ambient was actually 22c, and you followed the Test Setup to the letter, then your results are correct.

Since Tcase Max (maximum CPU temperature) for the Q8200 is 71c, then the corresponding Tjunction (Core temperature) is 76c, which shows the proper 5c relationship between CPU temperature and Core temperatures on a calibrated system.

Although Tjunction Max (maximum Core temperature) is 100c, this is shutdown temperature, which is far too hot for sane operation.

For the Q8200, any CPU temperatures which exceed 71c, and any Core temperatures which exceed 76c should be considered as an "overtemp" condition.

Comp :sol:
 
I guess what I was really trying to nail down is that 5 degree relationship between Tcase and Tjunction. Forgive me, but there is so much conflicting information on the internet about CPU temperatures.

I read your article and from what I understand, people have found that there tends to be a roughly five degree variation between case and junction values at load. There is still a part of me that insists on rebelling when I see RealTemp or HWMonitor spit out 75 degrees for one of my core temps. But then again there is more than a give degree variance between individual cores which hardly seems right.

I guess I really just need someone to comfort me and tell me that my computer won't catch fire in the middle of the night or perhaps just stop running a bit before its time because I ignored the core temps.

On a side note, I've lowered my Vcore to sub 1 levels, since I'm not planning on overclocking any time soon and my system is apparently stable so temperatures are less of an issue. The new issue is that the Tcase temperature (no calibration) hovers in the 19 degree range now which seems unrealistic.

PetrolMan
 
The most common mistake that many users make is to assume that uncalibrated temperatures must be accurate. Your Tcase Idle is below ambient, which obviously needs calibration, and your Idle AND Load Core temperatures are too high for default stock settings, which points to an improperly installed stock CPU cooler.

The second most common mistake is failing to understand how to install Intel's stock CPU cooler, which includes removing the protective clear plastic cover, and operating the push pins correctly.

From the Guide:




Simply pushing on the top of a pin causes the tip to spread while penetrating the board, which in turn causes the tip to resist seating and latching. Properly seating the the push pins is instead best accomplished as a two-handed operation, which is not described in the directions. Try the following:

(1) Use a large flat blade screw driver to push straight down on the end of the bracket next to the push pin, so as to carefully seat the tip of the pin completely through the motherboard, without pushing on the top of the pin.

(2) While maintaining pressure on the screw driver with one hand, use your other hand to then push on the top of the pin, which should latch easily with very little forcre, indicating that the pin is latched completely through the board.

If your Tjunction sensors aren't sticking, then under the conditions of the Test Setup, your mean Core temperature should Idle in the low 30's at 22c ambient.

Comp :sol:
 
Well, I've gone back and reseated the CPU cooler as per your instructions and there is no difference in the temperatures of the cores. Again, I'm using HWMonitor to get these temperatures, so they are uncalibrated. I have compared the values to RealTemp and a few others and for uncalibrated values they are all consistent. The interior plastic pin is now flush with the exterior plastic pin on the back side of the motherboard and there is roughly an eighth of an inch of the pins exposed. I would have taken a picture but I didn't have a camera at the time.

The only thing I that seemed perhaps wrong was that the thermal compound had spread out evenly but had left small areas of the copper on the heat sink either thinly covered or exposed.

I was aware that 19 degrees is too low for Tcase considering room temperature is probably around 20-21. My guesstimate places the realistic value for Tcase roughly 8-10 degrees higher for idle temperatures. So under load, the Tcase value should be roughly 42-44 degrees. The Tjunction sensors stick at a range of values between 40 and 47 so I have no way of determining the actual idle temps but those don't worry me so much. The question I have is should expect, even uncalibrated, that the Tcase and Tjunction values stay within 20 degrees of each other? The level of inaccurate that I'm dealing with seems a bit much.

My computer has been under load for the past 10 minutes and the temperatures for the cores are hovering at 60, 54, 58, 58. This is with a rather substantial drop to Vcore (1.225 to 1.000).

Another random issue: The Tcase value will occasionally jump much higher (to 55-60 degrees) for a brief moment. I have no idea if that is at all meaningful in any way. This could even be a error on the part of HWMonitor.

I appreciate the help.

PetrolMan
 
You are right to be skeptical; methods for "calibration" of Tcase (Tc) and Tjunction (Tj) are at best general guides, and at worst horribly inaccurate.

What are you trying to accomplish? If you want to know whether the part is exceeding it's design temp and potentially has a damaging effect, there are indicators specifically for that purpose (e.g., PROCHOT and THERMTRIP). Activation of TCC (thermal control circuit) can also be detected by the CPU reducing the clock rate or reducing the clock duty cycle. Newer CPU's also have a "countdown" (degress to TCC) accessible through the IA32_THERM_STATUS MSR. The reason for those is because there is no simple rule, as it varies by part, bin, stepping, load, voltage, etc. and the values are set at production.

So what's "safe"? Hard to tell. You can run with a lousy cooler and the system will simply invoke TCC and slow it down (PROCHOT) until it cools off, and if that isn't sufficient it will eventually shut down (THERMTRIP). Intel says that PROCHOT should be an exceptional condition, and that THERMTRIP should be a very exceptional condition (that is, you're risking damage by hitting it, specifially, it makes "an attempt to prevent permanent damage to the processor", but note they don't guarantee that hitting THERMTRIP won't damage it).

If you really want to get an accurate read and correlation, monitor Tc and run until you see TCC invoked: (1) you see the effective speed reduced due to a reduction in the clock duty cycle (PROCHOT); (2) until it shuts down (THERMTRIP); or (3) preferrably, if your CPU has it, by monitoring IA32_THERM_STATUS MSR. (1) Will tell you how hot you can run it before you reach the point of diminishing returns ("I'm hot and tired and slowing down"). (2) Will tell you how hot you can run it before you damage it ("It's too hot, goodbye"). (3) Will allow you to get as close as possible the limit without crossing it ("X degrees until I start slowing/shutting down").

In short, the only way to know is to correlate what the CPU is telling you via other information with what you see from Tc. If one core trips due to thermal excess and TCC is invoked, the system has, by definition, exceeded its thermal design limits and will attempt to recover by slowing down, or shutting down--unless you have a way to control cores and unlink them from the CPU thermal control, Tj for any given core is essentially irrelevant. If anything matters, it's the Tj across cores, not the Tj-Tc correlation. (Apologies to CompuTronix et. al., but the continued fascination with Tj and its correlation with Tc is a mystery to me. Intel has repeatedly stated why they don't state Tj-Tc numbers. Anyone familiar with the manufacturing processes should understand why, and why the published Tc-Tj correlations and calibration techniques are suspect.)


NOTE: PROCHOT can be both an input and an output. As an output, it means "CPU is overheating, I'm slowing down"; as an input, it means "slow down, the system is overheating" (e.g., VRM or other system components are overheating and the CPU needs to reduce it's power/heat output). If you want to accurately determine that PROCHOT is due to CPU activiy, you need to ensure all other components are adequately cooled. THERMTRIP is an output; it means "I've had it, goodbye".
 
The psychology behind why I want to know this is probably more important than anything. This is the first system that I've built and I want it to work well for quite a while. For most things, the information is pretty cut and dry. I can use benchmarks to judge the relative worth of processors, hard drives, ram, graphics card, etc. These benchmarks should hold largely true across the spectrum. The problem is often in the details and what I probably need someone to do is slap me and tell me that there is a forest and not just a tree.

The worry with my processor is twofold. On one hand I've seen very little information about operating temperatures for this particular processor, though what I have seen suggests that my experience is not completely uncommon. On the other hand, there are plenty of mentions of quad core processors running in the 30-40 idle range and loading in the fifties, even when overclocked. To get temperatures even close to those I had to drop my Vcore to 1 volt flat. If I had never known about the individual core temps I would have never bothered messing with the Vcore at all and would have never worried about my processor until it did hit a heat ceiling, and then only if it happened often. But now, gifted with incomplete and often conflicting knowledge, I worry. The best advice I've been given is "don't worry about it." But I can't help the part of me that says "lower temperatures are always better, so maybe you should worry just a bit."

So, here I am. In all likelihood, had I never messed with anything, the processor would have been just fine and would have lasted until I was ready for my next upgrade. And honestly, longevity is my biggest concern; I don't want to burn up this processor in 6 months or a year.

Before I ramble on for too long I just want to add that I think of this like I do the heat gauge on a car. My car should run in the 190 range when warm. The problem is that the processor has five temperature gauges with five different readings and I can't find a consistent answer on which gauge(s) to worry about.

I think I just need to quit worrying about this so much and move on. I'm going to go ahead and invest in a decent heatsink and hope that helps allay my worries.

If you've ever seen "Bad Santa" I feel like Therman Merman when he keeps asking questions about the North Pole and Santa'a Reindeer.

Thanks again,

PetrolMan

 
Then I'd suggest you ignore Tj and ensure Tc stays in spec and you should be fine. You'll only risk long term reliability if you consistently exceed it and your system shuts down (i.e., you hit THERMTRIP, whether caused by an individual core or the CPU as a whole).

To use your analogy, it's the difference between the engine temperature and individual cylinder head temperatures. An individual cylinder that runs a little hot is fine as long as it doesn't start misfiring (i.e., it can dissipate sufficient heat through the head or adjacent cylinders, i.e, the chip/heat-spreader).

Unless there's something seriously out-of-whack with one of them (which presumably would be detected/rejected at production), you're better off paying attention to the engine temperature (Tc) rather than individual cylinder temperature (Tj)'
 
jrst,

Welcome to Tom's.

Your points are well taken, however, understand that I have invested over 2,000 hours of work in my Guide, which includes research, intensive hardwarde testing, composition, editing and updates. While much of the information contained in my Guide is empirical, which has been painstakingly acquired over the past 32 months, I have also studied hundreds of pages of Intel documents, and have made it my mission here at Tom's to help users understand how temperatures work.

Additionally, I have read each of the 3,479 posts in each of the 140 pages of the Real Temp thread at XtremeSystems - http://www.xtremesystems.org/forums/showthread.php?t=179044&page=140 - and I have collaborated with it's author, Kevin Glynn, (unclewebb), as well as the authors of Core Temp, Arthur Liberman, and SpeedFan, Alfredo Comparetti. As such, rest assured that the temperature data shown in the Scales Section of my Guide was not pulled out of thin air, but is instead compiled from Intel's Processor Spec Finder - http://processorfinder.intel.com/Default.aspx

Real Temp is the ONLY temperature monitoring utility which provides a means to calibrate Tjunction Max, as well as Tjunction Idle. Although my SpeedFan calibrations are not perfect, I've tried my best to balance the procedure between accuracy and simplicity as much as possible, as I occasionally take heat (no pun intended) for the Guide being too detailed and complex, which is why I state that it's intended for intermediate to advanced users. Nevertheless, I believe that the calibrations are accurate within 3c, provided that users don't attempt to shortcut the Test Setup.

Over the past two plus years, I have extensively tested Idle, 50% Load and 100% Load at stock settings as well as overclock settings to observe and carefully document Tcase and Tjunction thermal relationships, sensor linearity characteristics, slpoe error behaviors, and power consumption. All testing was conducted under highly controlled conditions on a variety of motherboards, chipsets and CPU cooler combinations, which yielded results that were both consistent and repeatable. The following processors were tested:

(1) E2160 L2
(1) E4500 M0
(1) E5200 R0
(1) E6400 B2
(4) E6600 B2
(1) E6850 G0
(6) Q6600 G0
(1) Q6700 G0
(1) E7200 M0
(1) E8500 E0
(3) Q9650 E0
(6) i7 920 C0
(1) i7 940 C0

Of these 28 processors, I have found the 5c relationship between CPU temperature and mean Core temperature to be relatively constant, which is shown in the following Intel document - http://arxiv.org/ftp/arxiv/papers/0709/0709.1861.pdf - and was validated during the detailed thermocouple testing performed by rge for Real Temp. This is why I don't need to use Tjunction Max values in my Guide to achieve accuracy. To put it into perspective, utilities that use Tjunction Max values for calibration are a "top-down' method, while my procedure is a "bottom-up" method.

Additionally, my findings show that when Real Temp's author, Kevin Glynn, conducted his original IR testing, he was actually closer than he thought. His testing revealed that on processors which were Tjunction Max 100c, the measured value was actually 95c, however, I suspected that he was off by about 2c due to a few low order variables. Regardless, when Intel finally disclosed Tjunction Max values at last years' IDF 2009, he was reluctantly compelled to change his tables in Real Temp to comply with Intel's values, some of which still make no sense, and continue to be called into question.

Here's where it becomes interesting in terms of accuracy and calibrations; all the processors which I've tested have shown Tjunction Max values that average 97c, with the exception of the i7's, which are spot-on at 100c. This explains Intel's introduction of their new term "Tj Target" (synonymous with Tj Max), since Tjunction Max values are round figures such as 100c or 90c, while Tcase Max values are exact to the tenth of a degree.

Although the values shown in Intel's Processor Spec Finder are Tcase Max, most users are so brainwashed on Core temperatures since the introduction of "Core Temp" a few years ago, that they unfortunately continue to dismiss CPU temperature as a valid measurement, and mistakenly assume that the Processor Spec Finder shows Core Temperature. Regardless, Intel has repeatedly stated that the Digital Thermal Sensors are designed for throttle and shutdown protection only, which become less accurate below high temperatures, may be unresponsive below 50c on 45 nanometer Core 2 processors, and are unreliable at idle.

Further, since my testing, as well as the thermocouple testing conducted by rge, has shown that the Analog Thermal Diode (CPU temperature) is typically linear from low idle temperatures thru very high load temperatures, this explains why motherboard manufacturers, as per agreement with Intel, do not support Core temperature in their monitoring utilities. Consequently, my Guide is based upon SpeedFan simply because it provides a means to calibrate CPU temperature AND Core temperatures.

Having said this, I agree with you in that this entire temperature mess is just so much overkill. 4 cylinder engines don't need 5 temperature guages, and CPU's don't need intercoolers.

Comp :sol:
 
Small world. I actually ran into UncleWebb over at HardForum and this was the advice he gave me...

After a year or so looking into temperatures, I finally concluded that they are not that important. If your Intel CPU is running reliably at full load and it is not thermal throttling then you can ignore your core temperature. These things do a great job of looking after themselves. Your core temps are well within spec so no worries. Enjoy your Quad.

Here's the latest version of RealTemp for those that like to keep an eye on the numbers:
http://www.fileden.com/files/2008/3/...alTempBeta.zip

Run your CPU as cool as possible but don't worry too much about specific numbers since most sensors, core and Tcase, are far from 100% accurate.

Thermal throttling doesn't start until your core temperature is up to 97C. You have plenty of head room. If Intel thought you were going to damage your CPU at this temperature then obviously they would lower the thermal throttling point. Intel did the opposite. After the original 65nm chips came out, they raised the temperature by 10C before thermal throttling begins with the second generation 65nm processors.
Click to expand...

CompuTronix: You've actually thrown me for the biggest loop by suggesting that my temperatures were so abnormal as to suggest that I clearly had not seated the heatsink correctly. I appreciate all of the work you have done but at times you seem to suggest that the core temperatures aren't a particularly useful value and should be instead be linked directly to the Tcase value. Well, that is fine for me but then you've also told me that my Tjunction values were high, which seems to suggest some importance to even their uncalibrated values.

From some kind of a general consensus from people who seem to know what they're talking about I'm going to pay the most attention to a Tcase value, even though I know I need to add a few degrees to that. Still, I'm going to invest in a decent heatsink because that can't ever hurt.

My last question is short of watching CPU voltage and speeds, how would I know if I've ever reached the the slow down and stop points?

Thanks,

PetrolMan
 
PetrolMan,

Don't let me throw you for a loop, but don't misunderstand me. As I stated above, since most users are so brainwashed on Core temperatures, the vast majority will tell you to just ignore CPU temperature. While Real Temp monitors Core temperatures only, it is clearly my contention that CPU temperature AND Core temperatures are both relevant, which is why my Guide is based on SpeedFan. Off the record, 5 temperatures is overkill. On the record, since we have 5 temperatures, the more information, the better.

I agree with unclewebb in most respects, however, it always makes me cringe to see any processor temperatures above 75c. Temperatures are seldom an issue at stock settings, but when overclocking, it's all about understanding Intel's voltage and temperature specifications, as well as the importance of standardized test methods and calibrations. Since I'm a serious overclocker, I'm much more stringent with respect to advocating temperatures which don't exceed specifications during Prime, as I fully appreciate that stability is directly affected.

For example, if you look at my signature, you'll see that I've specified the following relevant items:

Core i7 920 @ 4.1 | Vcore Load 1.360
67c CPU | 72c Core @ 22c Ambient

The i7 VID Voltage Range Specification for Vcore Max is 1.375, so I'm within spec.
The i7 Thermal Specification for Tcase Max is 68c, which is CPU temperature, so I'm within spec.
The corresponding value for Core temperature would be 5c higher, or 73c, so I'm within the 100c spec.
The Ambient was 22c, which is Standard, and the Load was Prime95 Small FFT's, which is Standard.

Keep in mind that the Tjunction Max 100c spec is shutdown temperature, and is far too hot for sane operation, so Tcase Max (CPU temperature) is always the limiting specification, which is the value shown in Intel's Processor Spec Finder. Also keep in mind that even the most grueling games or apps will never cause your processor to approach the steady-state 100% workload of Prime95 Small FFT's. The highest peak workloads your processor will typically see is about 70% to 85%.

Thermal testing was conducted according to the Test Setup shown in my Guide, which is with case covers removed, and all fans at 100% RPM. As you can see, with the Xigmatek HDT-SD964 using dual 68 CFM push pull fans, which is the largest and most powerful cooler that will fit in my case, I've pushed my i7 as high at it will go without actually exceeding specs, although it's damned close. It's all about understanding the specs. After that, it's a balancing act between CPU cooler, clock speed, Vcore and temperatures, which is why calibrations become critical when overclocking.

Many users read my Guide but don't bother with the calibrations because they find that 4th grade math is too difficult, just as many users read the Real Temp documentation, but don't bother with those calibrations either. My time is usually limited, so I don't frequently offer to walk anyone through the calibrations. Regardless, I can see that your temperatures are inaccurate beyond a simple Tcase offset.

There are many questiuons that I could ask in order to identify the issue, but since you're running stock clock at 1.000 Vcore, temperatures are a non-issue, just as unclewebb pointed out. Also, like many users, you're much too concerned regarding electromigration and processor longevity. I've been overclocking for more than 15 years, and I haven't toasted one yet. Most of these overclocked rigs have long since been retired without ever having exploded into a mushroom cloud, and the remaining rigs are still running just fine 3 to 5 years into their obsolescence. So your stock cooler is fine until you actually begin overclocking. Again, it's all about understanding your processor's specs, and not exceeding them.

Q8200 M1 & R0:
Vcore Max 1.3625
Tcase Max 71c
Tjunction 76c

To answer your last question, if you want a utility capable of showing throttle activation, then Goolge for RM Clock, but you'll have to overtemp your processor to see it happen.

Comp :sol:
 
Thank you. That's quite helpful to me from both a comfort standpoint and from a knowledge standpoint. I realize I have very little to worry about running the my processor at the current voltage and speed. That was never so much an issue. What had me worried is that the motherboard was feeding the processor 1.225 V which is well within specs but was pushing core temps up to 75 degrees even though the Tcase was only around 55 degrees calibrated. That seemed hot for stocks speeds a voltage. Back to car my car analogy, the engine has a red line limiter to prevent excessive RPMs but I prefer to never reach the point where I even get close to it.

I know I've been worrying a bit too much about all of this and thanks for bearing with me but I always figure that a little extra knowledge never hurts. And I do plan on eventually overclocking to some degree even if I just bump the FSB up to 400.

Anyway, on a quick side not, how is your Xigmatek cooler? I've been looking around and I could probably squeeze it's cousin the 1283 in my case but I'm one of those unfortunate shoppers who has trouble making a final decision.

Thanks again,

PetrolMan
 
I couldn't be more pleased with the 964. It's one of only a few coolers that provides for a push-pull fan configuration. It's absolutely the best 92mm fan-class cooler available. It does a great job on my overclocked i7, which is a 130 Watt processor, and also did a great job on my overclocked Q9650, which is a 95 Watt processor.

Nevertheless, since I live aboard my sailboat, my case type and size was determined by the compartment into which the case must fit. Consequently, at 133mm, the 964 was the best possible choice to squeeze into a case that only has 131mm of clearance. As such, it's a bit of a trick to install the case cover, since the tips of the heat pipes actuall touch the inside of the cover, which makes it convex by 2mm, but isn't noticable.

The 1283 is awsome. Stock idle temperatures are no less than the 964, but overclock load temperatures are a few degrees lower. Check out the cooler test rankings at FrostyTech - http://www.frostytech.com/articleview.cfm?articleid=2383&page=5

If I could somehow cram a 1283 into my case, my 964 would certainly be history. I've installed several 1283's in new overclock builds, as well as upgrades to replace less efficient coolers on overclocked rigs. I highly recommend the 1283. It's outstanding. Make sure that you also purchase the 775 / 1336 Retention Bracket - http://www.newegg.com/Product/Product.aspx?Item=N82E16835233027

For thermal compound, I recommend Arctic Cooling MX-2 - http://www.newegg.com/Product/Product.aspx?Item=N82E16835186020

Comp :sol:
 
CompuTronix -- Thanks and apologies as I didn't mean to in any way disparage or minimize your work and effort, which should be obvious to anyone reading your how-to and subsequent posts. However...

As you said, "4 cylinder engines don't need 5 temperature guages, and CPU's don't need intercoolers." (Wonderful quote BTW which I expect to use frequently. :)

I think it would be great if you took that quote, and a short explanation of why (cut-and-paste from your response to me), and prefaced your guide/how-to with it.
 
jrst

No offense taken, and please feel free to use my quote. I came up with that one last year, and it seemed to be a hit.

As per your suggestion, believe me, I've thought about it, and have been very tempted to do exactly that.

Again, welcome to Tom's. It's always a pleasure to see someone new who brings clear and present wigage to the Forums.

Best Regards,

Comp :sol:
 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom