Margin of error on temp sensors

Merlin5 · 12 Sep 2009 at 04:20

I was discussing this with a friend but I'd like to throw it out and discuss it more. What kind of error margin should I consider for temp sensor accuracy? My core temp is always idle at 31 to 33C. With load under orthos or prime85, core temp maxes out at 44C. That seems very good for stock cooler almost too good. Which made me wonder how accurate the temp is. If I were to allow 10C margin of error would that be about right or too much? For instance, the max for my processor is 70C. If I saw my core temp reading 55C, I could assume it might be at 65C, but might also be less. Would that be about the right way to do it?

bifday2k · 12 Sep 2009 at 04:44

Your mixing the temps up, any official temp a maker states is the overall cpu temp, not core temps, so as you said ur max safe limit is 70c, then using something like hwmonitor or the like and go by overall cpu temp and not core temps.

JonJ678 · 12 Sep 2009 at 13:03

The reading is taken from a diode intended to shut the processor down at tj_max, say 100 degrees. So it can be reasonably expected to be accurate to within a degree or so at 100 degrees.

However, it has no interest whatsoever in any readings below this, and they are extrapolated in software. As such, it isn't calibrated for anything below 100 and so readings are not reliable. I can't even estimate how far off it swings, you'd have to ask the authors of real temp and the like for how they're extrapolating.

I would say that a reading of 50 degrees will always be hotter than 45 degrees, but there may well be more or less than 5 degrees in it. The 50 degrees might also correspond to 60 from a thermometer and most of us would be none the wiser. Just another reason to ignore idle temperatures really, there's no reliability to them whatsoever.

Merlin5 · 12 Sep 2009 at 13:13

bifday2k said:
Your mixing the temps up, any official temp a maker states is the overall cpu temp, not core temps, so as you said ur max safe limit is 70c, then using something like hwmonitor or the like and go by overall cpu temp and not core temps.

Ah! That's interesting, because I've been searching high and low for the answer to that. I've been trying to figure out which is true max operating temp, CPU v core. One person told me core. You say overall CPU temp as reported in bios. At last I know, thanks

In that case, when my core was at 44C with orthos, my cpu temp went up from 46C and stabilised at 58C, 12C away from max temp.

But why are so many people telling me to be more concerned about the core temp then? If as you say max temp is the cpu temp?

bifday2k · 12 Sep 2009 at 13:33

Iv asked myself the same thing over and over, but guess they are just to focused on the cores.

I do look at core temps too, but the main one i check when stress testing is the overal, as if thats well under safe then you have more headroom to play with, and if u have thermal control turned on, then you don't have to worry, as if the cpu hits that max temp, it will either stop the stuff stressing it, or reboot.

Im sure someone will come and say something about cores now, but ahwell, all i can say is i know my cpu's max temp, and i stay below it, and more or less ignore core temps lol.

Merlin5 · 12 Sep 2009 at 13:37

bifday2k said:
Iv asked myself the same thing over and over, but guess they are just to focused on the cores.

I do look at core temps too, but the main one i check when stress testing is the overal, as if thats well under safe then you have more headroom to play with, and if u have thermal control turned on, then you don't have to worry, as if the cpu hits that max temp, it will either stop the stuff stressing it, or reboot.

Im sure someone will come and say something about cores now, but ahwell, all i can say is i know my cpu's max temp, and i stay below it, and more or less ignore core temps lol.

Yeah, my thermal hardware control is enabled, so that's good to know it'll protect it.

Thing is, if I can get my dual core unlocked, I'll lose the ability to read core temperatures. But if as you say the bios reading of overall cpu temp is the important one, then I presume I don't need to see the core temp.

Merlin5 · 12 Sep 2009 at 14:19

Well this is an interesting article. From what this guy has discovered, it seems the cpu temp reported is inaccurate by 10C.

http://www.alcpu.com/forums/viewtopic.php?p=2844&sid=7bf2b2546a3e2ca32ed516404f8454c2

DIABLO · 12 Sep 2009 at 18:25

The temp you see using Real Temp, Core Temp is far more accurate because these programs get the temps directly from the cores the temps given to you by the bios, easytune 6 etc get the temps a from a sensor sat in the cpu socket and sometimes the bios is bugged and do not read the temps from this sensor accuratly thats why people perfer to use Real Temp, Core Temp etc.

Merlin5 · 12 Sep 2009 at 18:40

DIABLO said:
The temp you see using Real Temp, Core Temp is far more accurate because these programs get the temps directly from the cores the temps given to you by the bios, easytune 6 etc get the temps a from a sensor sat in the cpu socket and sometimes the bios is bugged and do not read the temps from this sensor accuratly thats why people perfer to use Real Temp, Core Temp etc.

Well real temp doesn't work with AMD anyway, I've tried it.
I've posted a thread at the AMD forum asking for a definitive answer to whether I need to worry about core temp going to 70c max or cpu temp. I've absolutely no idea how I'd possibly monitor an unlocked processor with no core reading visible and an inaccurate cpu temp, lol.

edit:
http://forums.amd.com/forum/messageview.cfm?catid=318&threadid=118772&enterthread=y