rtsai wrote:I have an Soltek SL-K8TPro-939 motherboard, A64-3500+ Winchester CPU, and Thermalright XP-120 heatsink...
..the motherboard BIOS (e.g., not even booted into an operating system yet) reports CPU temperatures of 50C, at what should be completely idle. But other threads in this forum (such as
this one) imply that temperatures should be more in the 30C range. Any ideas on why my temperatures are much higher?
There's a sample of that Soltek board here in the lab that I've been exploring, and it appears the board accesses the 2nd thermal sensor in the A64 -- and maybe the first as well. I can't recall exactly because it's been a couple weeks since I looked at this.
Both the P4 and the A64 have TWO thermal diodes in them. One is used for casual monitoring that is accepted by most of us as being the CPU core temp. The other one is used by the internal thermal protection circuit which shuts down the the CPU from working at all when a part of the die reaches 125C in the A64 -- not sure of the exact temp in the P4, but it's around there. Just like a circuit breaker.
My Soltek SL-K8TPro-939 sample's "CPU temp" sensor appears to read this second diode. This may or may not have changed; it did arrivea couple months ago, and boards are tweaked constantly by the mfgs.
To be more certain, while in the BIOS, just feel the top of the base of the HS where it meets the CPU. If it feels REALLY hot, then that 50C is in fact a casing temp and not the hottest core temp reading.
---------
All the above brings to mind the question:
Why 2 sensors? and why is this more deeply embedded hotter position diode not used for temp monitoring?
The answer is
Marketing, I believe: Intel made a decision somewhere along the way to hide real core temp. To keep the geek masses calm, happy. "
What they don't know..." Kind of goes along with specifying Thermal Design Power instead of Max Power.
Why would AMD duplicate this in the A64? Following the precedent set by Intel. If they just showed the true hottest temp in the core, they'd look too hot compared to the P4. Keep it simple, continue the CPU temp "protocol."
All this is speculation, of course.
The other possibility is that before thermal diodes appeared in CPUs, the standard procedure for establishing safe CPU cooling was to embed a thermal probe in the center of the base of the heatsink, mount it, then use the output from this thermistor to establish the outer CPU casing temp and track it under a controlled burn of the CPU. With a decent sampling of CPUs, you could establish the min casing temp at which the core would actually burn. For safety's sake, define the max recommend temp to be 20C (or whatever) below that.
The casing temp is still referred to in both AMD and Intel docs when they talk about max safe CPU temps, and the method to measure casing temp is as described above. (Go ahead and check in any of their thermal reference tech docs.) For CPU cooling developers (read: HS makers), this is the reference they are supposed to test with.
I wrote a long time ago how the P4's thermal monitoring diode is placed in a corner of the die far away from the core which can be 15~20C higher and that this seems like a kind of deception. This may well be true, but it also seems to me now that the corner placement of the diode may have been an attempt to replicate the external casing temp readings obtained by the HS-base embedded thermal sensor technique., which remains in use as the primary experimental reference for both Intel and AMD.