Confusion over CPU Temps

Deanodarlo

Senior member
Dec 14, 2000
680
0
76
Is it just me, or do people seem confused about CPU temps these days ever since programs started reporting CORE temperatures from the internal diode?

They seem to get worried about seeing CORE load temps of 60C, when in fact that's absolutely fine.

I may be wrong, but my take on it is that CPU manufacturers give the Tcasemax value for their CPUs, NOT the Tjunction max.

Tcasemax is the maximum temp the heatspreader or SURFACE of the CPU should reach before throttling sets in to save the CPU from meltdown. CPU manufacturers give you this in the specs, because it can be verified via an external heat probe by third party's without having to drill into a CPU. Typical value might be 65C.

Tcasemax sould also be the temp that is reported by your BIOS or given in a program like Speedfan as CPU TEMP. It's usually about 10-15C lower than CORE temp, probably calculated from the internal diode via software, and this is the value you should be comparing to the CPU manufactures Tcasemax.

Tjunctionmax is the maximum internal CORE temp the CPU should reach. It's usually much higher and not normally given by CPU manufacturers in the specs as it's hard to verify by third parties using probes. This is the value that you should be comparing the CORE temp output to, and could be as high as 80-90C before thermal protection kicks in.

To make things more confusing, video card GPU's normally have Tjunctionmax temps in their specifications rather than CPUs tradition of Tcasemax. That's why you see much higher values like 120C as the maximum temp before throttling, but at least it means people are comparing the internal CORE diode output to the correct value of Tjunctionmax.

Finally, the diodes and software interpretations are not very accurate anyway and simply a guide.

As an example, for a CPU with a Tcasemax (surface temp) of 65C from the manufacturers, I wouldn't worry about seeing CORE temps of 60-70C which quite possible only relates to a surface temp of 45-55C, well below the specs. Anything above 70C for the CORE would be heading towards the slow down protection zone.

If I'm wrong, then someone please correct me. I'm just a little tired of seeing people freak out over high CORE temps, once they go a tad over 50C.
 

F1shF4t

Golden Member
Oct 18, 2005
1,583
1
71
Just a correction Throttling is initiated by the cpu and is a function of Tjunction temp. For example Q6600 throttle at 95C, their tjunction is 100C.
As long as your temps are below 80C on the cores it should be fine.

It also depends on the chip and the heat spreader contact. My B3 q6600 is one hot chip, its lapped and at lower clocks and voltage it still runs hotter than my G0 q6600. When the B3 is water cooled the temps are identical to the G0 one.
 

rge

Member
Feb 18, 2008
50
0
0
Core temp is from diodes on core. CPU temp is from a diode between the cores. IHS temp or Tcase, there is no sensor.

The max difference you will see between core temp and cpu temp is ~5C, because both are on die substrate where thermal conductivity is very high http://arxiv.org/ftp/arxiv/papers/0709/0709.1861.pdf . You are correct in that you might see a much larger gradient if a true Tcase sensor existed on IHS, which is across die attach (source of most of the gradient since it has a much lower thermal conductance).

So core temp = temp of core, cpu temp = temp between cores (still on die substrate), and no true Tcase sensor exists. Intel will nevertheless tell you to use CPU temp as "Tcase specs", since CPU temp will always be higher than true Tcase, which gives an added safety margin. But that gradient/safety margin is highly variable, dependent on cooling, load type ie floating point, total TDP of load, etc.

Also, some load programs, according to intel paper, http://arxiv.org/ftp/arxiv/papers/0709/0709.1861.pdf have no offset between core temp and cpu temp, so it would be equally accurate to keep both core and cpu temp below thermal specs, even though as you correctly point out, this may give a hefty safety margin, due to a variable and sometimes high gradient between core/cpu temps and actual Tcase if measured with a thermocouple on IHS.

btw...we had similar discussion here... http://forums.anandtech.com/me...id=28&threadid=2195801
 

Deanodarlo

Senior member
Dec 14, 2000
680
0
76
Thanks for the more technical info. I'll have a read of that thread. :)

So, modern Intel CPU's have this extra diode between cores, which I never knew about as I've never owned one. I don't think AMD or at least single core processors do. In that case, the "CPU temp" must be calculated via software, as there usually is no Tcase sensor.