CPU Core temperature, the REAL temperature?

RaptorZX3 · Post by **RaptorZX3** » Sun Feb 03, 2008 2:16 am

ok i have to ask this question i've told myself so many times before asking it here...

do the CPU Core temperature is the one i HAVE to report here? or the "CPU" temperature?

i checked in the latest version of Speedfan, and now there's 2 different CPU temperatures monitoring: CPU, and CPU Core, both have different chip name:

name: CPU, Chip: Winbond W83627EHF, Sensor: CPU, Sample: 44C

name: CPU Core, Chip: AMD K8, Sensor: Core, Sample: 34C

which one i have to report for the CPU temperature? CPU or CPU Core?

does that mean one take the temperature from somewhere else than the core itself, and 1 take the real temperature from the source directly?

cpemma · Post by **cpemma** » Mon Feb 04, 2008 6:30 am

From what I gather re my Gigabyte DS3L board, the "CPU" one is a thermistor (on the board?), the "Core" ones are diodes in the Intel core (2 in the E2180). AMD may be similar these days. The diodes should be more accurate, though Speedfan reports my cores as little different to ambient (currently on idle 19C-19C with ambient 16C 6" above floor level) so I'm not wholly convinced even with a Ninja cooler and Intel's Speedstep.

"CPU" reports 23C, "System" 30C and GPU a more believable 54C (passive XFX 7200GS card known to run warm). Stress-testing, the "CPU" reading is a steady 2-3C higher than the "Core" readings. The freeware CoreTemp utility agrees exactly with Speedfan on my system.

tehcrazybob · Post by **tehcrazybob** » Mon Feb 04, 2008 9:39 am

As cpemma said, the readings with 'core' in the name are from the CPU's internal sensors, while the other reading is a less-accurate motherboard-based reading which you might as well ignore. However, the 'core' readings on certain recent Intel CPUs read 15C too low through most monitoring software. You'd probably want to take a peek at them with the very latest version of CoreTemp, which corrects for this error. Cpemma, that's probably the same reason you read so close to ambient - try the latest version and see if the numbers seem more reasonable.

cpemma · Post by **cpemma** » Mon Feb 04, 2008 5:23 pm

I only came across (and downloaded) CoreTemp a few days ago so assumed I had the latest. I had seen reports where Quad core Intels had the reading discrepancy between Core Temp and Speedfan, Core Temp was judged to be the correct one of the two, but in that case SF temperatures below ambient were being shown.

tehcrazybob · Post by **tehcrazybob** » Mon Feb 04, 2008 6:23 pm

I'm pretty sure the discrepancy shows up in every Intel with a 100C maximum Tjunction. That means it's present in all Allendale chips (E2xx0 and E4x00) as well as all Conroe chips from the G0 stepping (E6x50 and some quads, I think). I'm using CoreTemp 0.96, and I know it correctly handles these processors.

1398342003 · Post by **1398342003** » Mon Feb 04, 2008 6:46 pm

I can confirm the Speedfan error. Speedfan reports that my core2 quad runs at CPU 21c, Core0/1/2/3 17C. Core Temp says 32C - 34C for cores 0/1/2/3.

cpemma · Post by **cpemma** » Tue Feb 05, 2008 8:16 am

cpemma wrote:I had seen reports where Quad core Intels had the reading discrepancy between Core Temp and Speedfan...

Info was here. I'm on Core Temp 0.96.1 so I'd better believe it.

cmthomson · Post by **cmthomson** » Tue Feb 05, 2008 5:53 pm

On all C2D Intel CPUs, the DTS (digital temperature sensor) reports the same thing: the delta below throttling. It's a negative number reporting the number of degrees Celsius that the current CPU junction temperature is below the throttling temperature for that particular chip.

Intel chose not to report on a per-chip basis what its throttling temperature is (a stark contrast from AMD), so it becomes a guessing game as to what a DTS value of (say) -15 means in terms of absolute temperature.

The earliest C2Ds (eg, E6600) had a throttling temperature of 85C or so, while later ones (eg, E4x00) used 100C. More recent ones apparently are 105C.

On one hand, this ambiguity drives hobbyists crazy, because they can't accurately determine how hot their CPUs are. On the other hand, it really isn't that important, since what really matters is how much headroom the cooling system is providing to the CPU.

And this is of course why Intel came up with this approach in the first place. If you are a major system vendor (think Dell), the most important measurement you want is headroom. It tells you when, and by how much, to speed up the fan on the CPU heat sink to avoid throttling or shutdown.