TJ stands for thermal junction, and represents a measure of the temperature of a circuit or electronic package at the point where the part radiates the most heat (usually from the top surface). TJmax stands for the maximum temperature that such an integrated circuit, or more correctly a chip or chipset in its socketable package (a usable electronic part, in other words), can sustain before it fails.
The more expensive a part is, the more likely it is that electrical engineers will design it to include built-in temperature sensors and related chip logic to make sure it never run hotter than whatever value engineering analysis determines to represent its TJmax (see Wave Technology’s fascinating, if fractured English, “Introduction to the thermal characteristics simulation” for a more detailed discussion of the engineering terminology, measurements, and equations involved). Suffice it to say here that tracking the difference between a chip’s current temperature (or the temperature for some core on that chip) and TJmax is an important part of making sure the chip is behaving properly. For really expensive parts, such as CPUs (which is where this value really comes into play for PCs in general, and for this discussion in particular), this means keeping an eye on what’s called “the delta”—a value that represents the difference between TJmax for the part and a current, accurate measurement of its temperature.
For most CPUs, temperature readings that converge on TJmax cause the part to throttle back (which drastically slows processing, reduces electrical consumption, and thus also, attendant heat output). Readings within 10 degrees Celsius (°C) of TJMax are usually sufficient to trigger some kind of slowdown in the chip, which becomes progressively more drastic as temperatures get closer to TJMax, at which point the chip will usually shut itself down completely.
What makes measuring CPU temperatures interesting is that different programs set different values for TJmax for such parts, and it’s unusual for the vendor (Intel or AMD) to provide precise TJmax values for individual parts, either as part of their datasheets and specficiations, or in the set of values that is stored within a chip to describe its operating parameters and characteristics.
I learned all this because I run a “hot” QX6800 65 nm quad core processor on my production desktop, and it regularly runs a lot warmer (58-64 ° C) than I would really like it to. My AMD Athlon 64 X2 dual core CPUs seldom hit even 45 ° C, and other Intel dual core parts I’ve worked with have seldom edged much over 55 ° C. Although I’ve got a fabulous cooler (a Zalman CNPS 9500: great review of same at http://www.techpowerup.com/
printreview.php?id=/Zalman/CNPS-9500LED) and a well-ventilated case (two 120mm fans, plus an extra drive cooler in my build, in a ThermalTake Elite 330 case; for some more marvelous fractured English, read the product description at http://www.case-mod.com/cooler-master-rc330kkn1gp-elite-330-
elite-series-middle-tower-atx-chassis-black-p-3575.html), my old production rig used to regularly reach temperatures over 70 ° (even as high as 80 °) at peak loads.
Over the years, I have discovered four very good PC temperature monitoring utilities, all of which I recently tried out on this PC:
- SpeedFan, by Alfredo Comparetti, www.almico.com/speedfan.php
- CPUID Hardware Monitor, aka HW Monitor, by Franck Delattre, www.cpuid.com/hwmonitor.php
- Core Temp, by Arthur Liberman, www.alcpu.com/CoreTemp/
- Real Temp, by Kevin Glynn, www.techpowerup.com/realtemp/
If you run these programs, you can’t help but notice that the temperatures they report vary somewhat. Because I know Franck Delattre, I emailed him with some screenshots, and asked him why this might be so. Here’s a quote from his reply:
We all read the core temperature from the same place, but … none of us uses the same TJmax, that is the reference temperature (reported temperature = TJmax -read value).
Intel does not provide clear information about Tjmax, and my guess is that each processor has its own. That may explain why Intel did not continue the TAT tool introduced for Core “1” Solo & Duo.
In the simplest of terms, this means that each programmer assumes a value for TJmax (these programs make it 95 ° C, 100 ° C, or 105 ° C, depending on the programmer and the software release). Because some of these values are higher than others, they report lower temperatures (because a bigger difference value read from the chip indicates a lower temperature); because some are lower, they report higher temperatures. In fact, they’re all reading the same values from the chip (which represents the difference between TJmax, whatever it may actually be, and the temperature read) but calculating the “real temperature” a little bit differently.
According to the experts in this area, the delta between the read value and TJmax (or the value that the chip reports) is what’s really important. As long as it’s over 30 ° C, you’re in fine shape; as long as it’s over 20 ° C, you won’t see much if any performance impact. But it does go to show that keeping cool has a value above and beyond the style points that accrue thereby, particularly for expensive PC chips and chipsets. These readings also prompted me to rebuild my QX6800 system around a QX9650 CPU instead, simply because the latter’s 45 nm layout allows it to run 20-30 degrees cooler (° C, that is) than the former’s usual temps.
