Questions: tcase temps on the i7 920

Mr McC said:

As Intel only officially release maximum tcase temperatures for their i7 processors, I would be much obliged if our resident experts could shed some light on the issues raised below. I have been reading various articles and forums and rather than clear information I have encountered a lot of contradictions and I currently feel more confused.

a) Tcase temp is the cpu max temp measured at the centre of the chip, between the cores. Is this correct? Almost........ The case temperature measured at the geometric center of the topside of the IHS.

b) We have absolutely no means of ascertaining the exact tcase temp, unless we are willing to drill into the chip and place a sensor. Is this correct and, if so, why do Intel only release a temperature figure that is practically impossible to accurately measure? Assuming that I cannot accurately measure tcase, should I simply forget about it? Depends on how accurate is accurate but your probably right yes, sensor readings are actually taken from the TCC (Thermal Control Circuit) which is in the centre of the IHS BUT not Topside I think. In addition to that, the TCC only kicks in at a certain temp depending on heat buildup shown in (c) below

c) The Bios and programs such as CPUID’s Hardware Monitor provide approximations of the figure for current tcase temperatures, whilst most programs provide tjunction or core temperatures. Is this correct? Again kind of yes, Intel use a formula for TCase thermal design, it goes like this .....[CA = (TTV TCASE – TAMBIENT) / Power, it is at this point that thermal control can begin to take place within the chip, in part controlled by the BIOS but the measurement is probably not completely accurate.

d) Can core or tjunction temps be estimated on the basis of the tcase temp provided by Intel? Can I assume that when I read core temperatures of X, that tcase temperature will be approximately Y degrees below those temperatures? No i don't think so, there are too many variables such as current, speed etc but I might be wrong here.

Thank you all in advance.

Click to expand...

As I understand it, there is no need to worry about TCase, if you have your thermal monitoring enabled then you theoretically should not be able to reach TCase max as you would have already reached TJunction max etc beforehand (these are read by the DTS (Digital Thermal Sensor), more accurate to read) and thermal monitoring would have pegged back the CPU.... again not an expert on this so not entirely sure.

I would suggest you got to the realtemp thread and pose your question there, the author (Unclewebb) will know a lot more about this than me Post 680 may help a little!

http://forums.techpowerup.com/showthread.php?t=64185

The Tcase specification is intended to be used by professional system builders. This number is a guideline so a manufacturer can create and size a heatsink if they need to and determine an appropriate sized CPU fan as well as manage case air flow. The core temperature that RealTemp and Core Temp measures is coming from sensors located at the hottest spots on the core and is usually referred to as the Tjunction temperature.

There is no magic formula to convert Tcase to Tjunction. The difference between these two temperatures can be substantial or it can be next to nothing depending on if the CPU is idle or what kind of stress testing program you are running. The temperature gradient between these two sensors will vary.

Thermal throttling in Intel CPUs is 100% controlled by the Tjunction (peak core temperature). Without carving up your CPU, you can't accurately measure Tcase and the CPU is not controlled by this temperature so for the end user, Tcase is not an important number. The bios can report a temperature from a sensor located in the center of the CPU but that sensor is not the same as a true Tcase measurement.

The theory is that if you design a computer and it does not exceed the measured Tcase temperature specification; in normal use, this CPU should rarely trigger thermal throttling.

The RealTemp Distance to TJMax is a direct reading of the sensors within an Intel CPU. When this number counts down to zero, thermal throttling begins. As long as your CPU is not thermal throttling, you are running it within the Intel design specification and it should have no problem running reliably for its intended life span.

The Thermal Status area of RealTemp directly monitors the status of the thermal throttling bit. If your CPU on any core starts to thermal throttle for even a millisecond, RealTemp will change OK to LOG to show that a thermal throttling episode has been logged within the CPU. When LOG changes to HOT, that shows that thermal throttling is still in progress. That's all you really need to be concerned about when it comes to temperature monitoring. OK in RealTemp means everything is OK. Your CPU is not presently thermal throttling and it has not thermal throttled since you booted up so you have been running it within the Intel specifications.

To simulate a worst case scenario, this CPU has been over clocked and over volted and I disconnected the CPU fan to create some serious heat. After 3 hours of this insanity, there were zero errors in Prime95 at which point I decided to stop this test. Only once in 3 hours did one core hit 100C. A hundred times a second, as the CPU gets too hot, it slows down a little. When the temperature is back within spec, it goes back up to full speed. It infinitely cycles like this which maintains the CPU core temperature within the Intel specification.

That's the day when I learned that people worry way too much about their CPU temperature. As long as your CPU is 100% stable and it is not thermal throttling, there's no need to worry about its core temperature. If you like to overclock then I also found out that the cooler you can run your core, the more you will be able to reliably overclock your CPU. This same CPU can run fine at 4000 MHz but it can't run that speed at 100C. At 4000 MHz, you have to keep the core temperature under about 70C to maintain long term Prime stability.

That's it. If the CPU is not throttling then it is running within the Intel temperature specification. That's all anyone needs to be concerned about.

The exact temperature that a core starts to throttle varies. The sensors that Intel uses are not high dollar, ultra precise temperature monitoring devices. These sensors are only designed by Intel to be used to control thermal throttling and thermal shutdown and are not recommended for any other purpose. On most CPUs, Intel saved a few pennies by using sensors that are not 100% accurate temperature monitoring devices. The good news is that the 45nm Core i7-900 series have some of the best sensors that Intel has ever used.

The other problem is that Intel might not set the same TJMax value for each core of a Quad CPU. Based on testing, my opinion is that they likely deliberately offset the TJMax temperature slightly to prevent all 4 cores reaching the thermal throttling point at the exact same time. The Core 2 Quad consists of two separate Core 2 Duo CPUs internally so during a thermal throttling episode, you can throttle one Core 2 Duo CPU while the other 2 cores on the second CPU can still run at full speed as long as they haven't triggered thermal throttling yet. By setting the actual TJMax 5C or 10C higher on the second set of cores, there is a lot less chance of all 4 cores throttling at the exact same time. Throttling 2 cores is usually enough to keep the heat under control. This allows for a much more gradual reduction in processing power and a better user experience. No one wants their computer to instantly slow down to a crawl when there is a minor temperature problem.

Intel does not document any of this so it's just my best guess. I believe they do the same sort of thing on the Core i7 CPUs like your 920. I find that there is frequently a spread of about 5C from Core 0 to Core 3 that has nothing to do with a difference in actual core temperature. Run Prime95 Small FFTs on all 8 threads and you will likely see this difference that I am talking about. If TJMax for Core 0 is set to 100C then Core 3 is typically set 5C higher so 105C with the two center cores set somewhere in the middle; usually close to core 0.

There is a value written into each core of the Core i CPUs but this value is not TJMax. Intel lists the value as TJ Target. The Core i7-920 is typically set to 100C. TJ Target allows Intel some flexibility. Intel is shooting for 100C but if they cheap out and use some crappy sensors and the value is slightly different than 100C, that's OK. If they decide to secretly offset TJMax a few degrees from core to core for whatever purpose they decide upon, that's OK too. Intel has never documented their exact calibration procedure or the accuracy of that procedure or the accuracy of the sensors they use which varies from one CPU line to the next. Based on that, there will always be more questions than answers.