Hi Mark, Mark M. Hoffman wrote: >Hi Trevor: > >Please, send only plain text email! > > > Sorry about that. >* TN <tnuro at yahoo.com.au> [2003-10-31 10:20:49 +1100]: > > > >> The temperature readings I am getting are definately too low, despite >> variances like bios loading vs linux loading while observing the >> readings, as I take the readings as soon as the OS or bios has booted, >> and so I don't believe the CPU can change temperatures by such a large >> gap in such a short time. >> >> > >I have a power measuring device which tells me that one of my computers >uses ~55 watts at idle but ~80 watts when in the BIOS screen. Cup your >hand over a 25 watt light bulb and turn it on, then you'll believe... > > > well that 25 watts is dissipated over the entire machine of course, not just the cpu, gpus run damn hot too - BUT- I take your point. I have a 12 watt soldering iron which gets reasonably warm ;) so I do know what 10 watts can do. I still think the readings are incorrect though, as my remote server is sitting in a tropical location with an ambient room temperature of about 30celsius, so for the CPU to claim only a few C above that just isn't believable to me. >> lm-sensors on my P4 2.8 says about 30c. Bios says about 40-42, which I >> believe is more like it, maybe even then a little low based on my >> experience with all our other machines. >> I've had a quick rummage around in the w83627hf.c code, and I can't >> see anywhere where the CPUTIN offset register is added into the raw >> reading from CPUTIN, and neither can I see the other offset registers >> for SYSTIN & VTIN being added in to SYSIN, VTIN. Am I blind & haven't >> read the code right ? >> >> > >The way I read the datasheet, the value of the offset registers are added to >the raw temp readings by the sensor chip itself. This is bizarre, I've never >seen another chip do that. I haven't tested this theory yet though... > > > Yep, I agree - it's vague - and I've not seen chips auto-add registers before either. "CPUTIN temperature offset value. The value in this register will be added to the monitored value so that the reading value will be the sum of the monitored value and the offset value." Having a flick through the rest of the doc, it reads to me a bit like chino-english in some minor parts, and so I think the translation is not accurate, although on the whole its good. I'm guessing the opposite case to you I suppose, I guess that they are instructing the reader to do the adding (as in "will be added" = "you shall add") >> When I do a isadump on reg 0x55, bank 4, my CPUTIN offset is 0x11. >> What has written this value into this register, is it the bios or >> somewhere in the lm-sensors modules ? >> >> > >Just checking, but you did this? > ># isadump 0x295 0x296 4 0x4e > > > yep, i did that. >Mine reads 0x11 too. But, the reported value of 38C looks ok for me. Also, >my VTIN offset is 0xfe, which is ridiculous unless it's interpreted as a >signed value (two's complement)... in which case it's -2. And furthermore, >what's the scaling? Two of the temps report 9 bits for LSB of 0.5C. So >are those offset regs 1C/bit or 0.5C/bit? > > > My VTIN offset is also 0xfe, and I interpreted it as 2's complement too, so its -2. I assume the setup is saying that the VTIN sensor is reading too high, so they pull it back by 2C. I'm also guessing that the offset is 1C increments, since all other 8 bit temp values are. Its still not exactly spelled out though in the docs. >Who knows? The datasheet is very ambiguous about all that. Some time this >weekend I'll modify the driver to poke some values in there and see what >happens. > > yes, we are guessing and things are open to interpretation, so we'll have to dig some more. I'll attempt to contact winbond and get an answer to the vague parts. >In any event, whatever programmed h/w offset is redundant anyway because >you can get the same result by modifying the compute lines of sensors.conf. > >And BTW: here is my current sensors.conf with some comments about what >I'll be working on... > ># P4C800E-Deluxe > >chip "w83627thf-*" > ># VOLTAGES > label in0 "VCORE" > #(TODO) the following compute line is b0rk... > #compute in0 (0.488 * @ / 100) + 0.69 , (@ - 0.69) * 100 / 0.488 > > label in1 "+12V" > compute in1 3.8 * @ , @ / 3.8 > set in1_min 12.0 * 0.90 > set in1_max 12.0 * 1.10 > > label in2 "+3.3V" > set in2_min 3.3 * 0.95 > set in2_max 3.3 * 1.05 > > label in3 "+5V" > compute in3 1.68 * @ , @ / 1.68 > set in3_min 5.0 * 0.95 > set in3_max 5.0 * 1.05 > > ignore in4 > ignore in5 > ignore in6 > > label in7 "5VSB" > compute in7 1.68 * @ , @ / 1.68 > set in7_min 5.0 * 0.95 > set in7_max 5.0 * 1.05 > > # (TODO) driver needs to enable this one... > label in8 "VBat" > set in8_min 3.0 * 0.80 > set in8_max 3.0 * 1.20 > ># FANS > label fan1 "Chassis Fan" > ignore fan1 > set fan1_div 8 > set fan1_min 1500 > > label fan2 "CPU Fan" > set fan2_div 4 > set fan2_min 2000 > > label fan3 "Power Fan" > set fan3_div 8 > set fan3_min 1500 > ># TEMPS > label temp1 "M/B Temp" > set temp1_over 50 > set temp1_hyst 45 > > label temp2 "CPU Temp" > set temp2_over 60 > set temp2_hyst 55 > > label temp3 "PWR Temp" > ignore temp3 > set temp1_over 50 > set temp1_hyst 45 > ># OTHER > # (TODO) vid technically not supported... > ignore vid > >Regards, > > OK, thanks for that! Its certainly interesting comparing notes & interpretations on this, so I'll try to discover more too. I'd love to afford a reasonably accurate temperature probe to check all this out, I'll see if I can borrow one. cheers Trevor.
