Re: Which temp are right.

"Phil Weldon" <> wrote in message
news m...
> 'Dumbo' wrote:
>> Has anyone got a sure way of monitoring the CPU temperature accurately.
>> When idle in BIOS my CPU displays as 45C but `Core Temp` shows it as 50C
>> and real temp as 36C. SANDRA shows a third different temperature and a
>> fourth from Everest.
>>
>> How do I know which is right and are the temp sensors on the motherboard
>> that accurate anyway.
> _____
>
> What to consider:
>
> A. actual temperature
> 1. sensor somewhere in a core (some areas tend to produce more
> heat that others)
> 2. sensor somewhere in the CPU chip (but not in a core)
> 3. sensor on the CPU chip in contact with the CPU heat spreader
> 4. sensor not part of the CPU chip, but in contact with the CPU
> chip or heat spreader (yes some CPUs have not had heat spreaders)
>
> [all of the above locations have been used in the past, but only 1.
> and 2. are used for current x86 CPUs]
>
> B. type and precision of sensor
>
> C. convertor (A to D) used in the motherboard system monitoring chip
> (or Super I/O chip) to translate sensor voltage or current signal from the
> sensor to decimal digits ( precision of the reference voltages )
>
> D. offsets (fudge factors) applied to measured temperatures to get
> something closer to actual temperatures
>
> E. monitoring program used (some monitoring programs may not recognize
> the monitoring chip or Super I/O chip [ there are dozens of
> possibilities ] used by your motherboard, some monitoring programs may not
> know the type sensors (thermistor, diode, transistor, active devices) used
> other than on CPU chip sensor
>
> [all of the above affect the ACCURACY of the temperature reported]
>
> F. PRECISION errors
>
> 1. the number of bits the temperature A to D convertors use;
> an 8 bit A to D convertor can only represent one integer steps from - 127
> to + 127, with a plus or minus one error ( i.e. 45, 46, and 47 might all
> represent the same analog signal )
>
> 2. if the monitoring program is set to display in Fahrenheit
> then, since only integers are displayed, the reported temperature has an
> even larger PRECISION error.
>
> The monitoring program supplied by the motherboard manufacturer SHOULD be,
> but is not always, the best. After two years, EVGA still has not provided
> a monitoring program that will report all the values reported in the BIOS
> for my 680i SLI motherboard.) MotherBoard Monitor is the most universal
> motherboard monitoring program ever issued, but it hasn't been updated for
> 5 or more years. If you haven't already purchased your motherboard make
> sure you get one that comes with a good motherboard monitoring system that
> works correctly with your specific motherboard model AND revision
>
> Then the question is, what is the most USEFUL CPU temperature? My feeling
> is that the temperature most useful is that at the outer surface of the
> CPU heatsink. YOU have control of heat transport from this surface
> outwards; the CPU manufacturer has control of heat transport from this
> surface inwards ( unless you remove the heat spreader ). But to get this
> temperature requires modifying your heatsink and adding a temperature
> sensor in contact with the heat spreader, but insulated from the heat
> sink. I can suggest three ways of doing this, one of which I have tried.
>
> 1. Back in the days of the Pentium Celeron 333a I drilled a hole through
> the center of the heatsink bottom surface and glued it in place with
> Silicone which held the sensor in contact with the heat spreader, but
> insulated from the heat sink. Since this was a temporary setup I used a
> DVOM ( 4 1/2 digit ) to measure the thermistor resistance and manually
> converted to temperature. I tested extensively with the following
> variables; CPU voltage, CPU clock speed, and thermal compound. The
> temperatures reported by this sensor was stable, and tracked the
> temperature reported by MBM using the on-CPU-die thermal diode. This
> temporary system could be made permanent and more convenient but replacing
> the thermistor with an active device that produces a pulse width modulated
> signal that can be digitized and presented to a serial port. The sensor
> device and digitizer/RS-232 Serial interface with a monitoring program
> are available off-the-shelf.
>
> 2. Fiber optic bundle with non-contact digital temperature readout, less
> than $10 US for both.
>
> 3. Small diameter heatpipe with non-contact digital temperature readout,
> less than $10 US for both if you get a fee sample heatpipe swizzle stick.
>
> On the other hand, why worry about your CPU temperature? The on-CPU-die
> thermal protection device will prevent heat damage (this is a fail-safe
> device that can't be bypassed ). Assuming you want the best possible
> reliable performance and not just bragging rights for the fastest possible
> CPU without regards to stability, then
> 1. check for proper thermal compound application ( the brand and type
> make almost no differences )
> 2. check for proper heatsink installation
> 3. check or proper case ventilation
> 4. check room ambient temperature ( each degree drop in room ambient
> temperature will result in almost a degree drop in CPU temperature )
> After the above checks, consider if a better cooling solution is worth the
> expense and how that expense compares to the expense of spreader plate
> temperature monitoring.
>
> Phil Weldon
>
On the new i7 (Nehalem) processors, I am not sure about a few things in
regard to temperatures. The tjmax is 100C. That is about where the processor
begins to throttle down. I backed down my a very stable 3.7Ghz OC just
because of my older views on what is too hot.....:-). I don't know if the
fact that the memory controller is on chip make these run hotter, but there
is a significant rise in the temperatures compared to what I have become
used to over the years. If these temps are actually the max temps inside
each core, then I might not be quite as concerned since that would be about
what I would expect since previous sensors usually just read the surface
temps of each core. I have ran the i7 920 at over 85C for hours on end and
it stays perfectly stable, but I don't........:-)


Ed

OT--- BTW, how is your astronomy project going Phil?

Ed Medlin wrote:
> "Phil Weldon" <> wrote in message
> news m...
>> 'Dumbo' wrote:
>>> Has anyone got a sure way of monitoring the CPU temperature accurately.
>>> When idle in BIOS my CPU displays as 45C but `Core Temp` shows it as 50C
>>> and real temp as 36C. SANDRA shows a third different temperature and a
>>> fourth from Everest.
>>>
>>> How do I know which is right and are the temp sensors on the motherboard
>>> that accurate anyway.
>> _____
>>
>> What to consider:
>>
>> A. actual temperature
>> 1. sensor somewhere in a core (some areas tend to produce more
>> heat that others)
>> 2. sensor somewhere in the CPU chip (but not in a core)
>> 3. sensor on the CPU chip in contact with the CPU heat spreader
>> 4. sensor not part of the CPU chip, but in contact with the CPU
>> chip or heat spreader (yes some CPUs have not had heat spreaders)
>>
>> [all of the above locations have been used in the past, but only 1.
>> and 2. are used for current x86 CPUs]
>>
>> B. type and precision of sensor
>>
>> C. convertor (A to D) used in the motherboard system monitoring chip
>> (or Super I/O chip) to translate sensor voltage or current signal from the
>> sensor to decimal digits ( precision of the reference voltages )
>>
>> D. offsets (fudge factors) applied to measured temperatures to get
>> something closer to actual temperatures
>>
>> E. monitoring program used (some monitoring programs may not recognize
>> the monitoring chip or Super I/O chip [ there are dozens of
>> possibilities ] used by your motherboard, some monitoring programs may not
>> know the type sensors (thermistor, diode, transistor, active devices) used
>> other than on CPU chip sensor
>>
>> [all of the above affect the ACCURACY of the temperature reported]
>>
>> F. PRECISION errors
>>
>> 1. the number of bits the temperature A to D convertors use;
>> an 8 bit A to D convertor can only represent one integer steps from - 127
>> to + 127, with a plus or minus one error ( i.e. 45, 46, and 47 might all
>> represent the same analog signal )
>>
>> 2. if the monitoring program is set to display in Fahrenheit
>> then, since only integers are displayed, the reported temperature has an
>> even larger PRECISION error.
>>
>> The monitoring program supplied by the motherboard manufacturer SHOULD be,
>> but is not always, the best. After two years, EVGA still has not provided
>> a monitoring program that will report all the values reported in the BIOS
>> for my 680i SLI motherboard.) MotherBoard Monitor is the most universal
>> motherboard monitoring program ever issued, but it hasn't been updated for
>> 5 or more years. If you haven't already purchased your motherboard make
>> sure you get one that comes with a good motherboard monitoring system that
>> works correctly with your specific motherboard model AND revision
>>
>> Then the question is, what is the most USEFUL CPU temperature? My feeling
>> is that the temperature most useful is that at the outer surface of the
>> CPU heatsink. YOU have control of heat transport from this surface
>> outwards; the CPU manufacturer has control of heat transport from this
>> surface inwards ( unless you remove the heat spreader ). But to get this
>> temperature requires modifying your heatsink and adding a temperature
>> sensor in contact with the heat spreader, but insulated from the heat
>> sink. I can suggest three ways of doing this, one of which I have tried.
>>
>> 1. Back in the days of the Pentium Celeron 333a I drilled a hole through
>> the center of the heatsink bottom surface and glued it in place with
>> Silicone which held the sensor in contact with the heat spreader, but
>> insulated from the heat sink. Since this was a temporary setup I used a
>> DVOM ( 4 1/2 digit ) to measure the thermistor resistance and manually
>> converted to temperature. I tested extensively with the following
>> variables; CPU voltage, CPU clock speed, and thermal compound. The
>> temperatures reported by this sensor was stable, and tracked the
>> temperature reported by MBM using the on-CPU-die thermal diode. This
>> temporary system could be made permanent and more convenient but replacing
>> the thermistor with an active device that produces a pulse width modulated
>> signal that can be digitized and presented to a serial port. The sensor
>> device and digitizer/RS-232 Serial interface with a monitoring program
>> are available off-the-shelf.
>>
>> 2. Fiber optic bundle with non-contact digital temperature readout, less
>> than $10 US for both.
>>
>> 3. Small diameter heatpipe with non-contact digital temperature readout,
>> less than $10 US for both if you get a fee sample heatpipe swizzle stick.
>>
>> On the other hand, why worry about your CPU temperature? The on-CPU-die
>> thermal protection device will prevent heat damage (this is a fail-safe
>> device that can't be bypassed ). Assuming you want the best possible
>> reliable performance and not just bragging rights for the fastest possible
>> CPU without regards to stability, then
>> 1. check for proper thermal compound application ( the brand and type
>> make almost no differences )
>> 2. check for proper heatsink installation
>> 3. check or proper case ventilation
>> 4. check room ambient temperature ( each degree drop in room ambient
>> temperature will result in almost a degree drop in CPU temperature )
>> After the above checks, consider if a better cooling solution is worth the
>> expense and how that expense compares to the expense of spreader plate
>> temperature monitoring.
>>
>> Phil Weldon
>>
> On the new i7 (Nehalem) processors, I am not sure about a few things in
> regard to temperatures. The tjmax is 100C. That is about where the processor
> begins to throttle down. I backed down my a very stable 3.7Ghz OC just
> because of my older views on what is too hot.....:-). I don't know if the
> fact that the memory controller is on chip make these run hotter, but there
> is a significant rise in the temperatures compared to what I have become
> used to over the years. If these temps are actually the max temps inside
> each core, then I might not be quite as concerned since that would be about
> what I would expect since previous sensors usually just read the surface
> temps of each core. I have ran the i7 920 at over 85C for hours on end and
> it stays perfectly stable, but I don't........:-)
>
>
> Ed
>
> OT--- BTW, how is your astronomy project going Phil?
>
>
Might this be part of the industry move allowing for hotter computing
rooms? There is a move to do away with costly air conditioners in server
rooms. Going to use more ambient air; collect the cooler night air, etc.
But also getting the electronics to run at higher temperatures for long
periods of time, without harm.

>> On the new i7 (Nehalem) processors, I am not sure about a few things in
>> regard to temperatures. The tjmax is 100C. That is about where the
>> processor begins to throttle down. I backed down my a very stable 3.7Ghz
>> OC just because of my older views on what is too hot.....:-). I don't
>> know if the fact that the memory controller is on chip make these run
>> hotter, but there is a significant rise in the temperatures compared to
>> what I have become used to over the years. If these temps are actually
>> the max temps inside each core, then I might not be quite as concerned
>> since that would be about what I would expect since previous sensors
>> usually just read the surface temps of each core. I have ran the i7 920
>> at over 85C for hours on end and it stays perfectly stable, but I
>> don't........:-)
>>
>>
>> Ed
>>
>> OT--- BTW, how is your astronomy project going Phil?
> Might this be part of the industry move allowing for hotter computing
> rooms? There is a move to do away with costly air conditioners in server
> rooms. Going to use more ambient air; collect the cooler night air, etc.
> But also getting the electronics to run at higher temperatures for long
> periods of time, without harm.

I really think it is mostly due to the complete redesigning of the i7 from
the previous Intel processors. They may be measuring the temps at a
different point too since I have seen some overclocking reviews where even
with liquid cooling temps are only 3-4 degC lower and sometimes not even
that. If temps were measured at the hottest point in the core of even a C2D
I would bet it would be much higher than what we see with peripheral core
readings. The only times that those extremely high temperatures are seen are
with "burn-in" apps that max out all the cores to the extreme. Even during
intense gaming the temperatures are nowhere near 85C and stay in the 60-65C
range which is acceptable for any processor. That is 35-40C under the tjmax
of 100C of the processor so that should be just fine.


Ed

On Thu, 8 Jan 2009 20:31:38 -0000, "Dumbo" <>
wrote:

>At POST my BIOS reported 39C for CPU and 34C for the motherboard yet booting
>into Vista even on idle the temps range from 46C-49C with Core Temp and
>31C-34C with Real Temp but I just want to know which one is most accurate so
>I can do a bit of overclocking without the worry of burning anything out.
>Real Temp gives a table with a `Distance to TJ Max on it but 100C seems to
>boiling temp and only asking for bother.
>With a bit of gaming and returning to Vista I get a temp of around 69C. Is
>this a safe enough temp or do I need to cool things a bit more (lower speed
>or increased airflow).
>I am trying to figure out the best most stable overclock but this seems
>inherently hot! I try to determine what is causing the BSOD or crash. Does
>the memory, CPU, NB or SB need more voltage and how hot is it going to get.
>I have a reasonably stable 3GHz Q6600 B3 but still get crashes but recently
>I have found the same crashes at default speeds of 2.4GHz so I assume its a
>bit of bad data on the drive or graphics card problem even though I haven't
>tried the card until I get 100% stable CPU.
>Thanks for your input.
>
>Robin

I get identical temps from Core Temp and Real Temp. I have noticed
that most of the people showing screen shots of values on the
overclocking forums seem to use Real Temp. Similarly, most of them use
Memtest86+2.11 rather than Memtest86 3.?. They also use Everest, but
that's to show other parameters rather than the temperatures. I don't
know that that's a recommendation, but in any event, start by checking
your memory for at least four passes or 8 hours. You may need to
detune your CPU back to 2.4 GHz until you are sure your memory is
stable.

Jim

"Phil Weldon" <> wrote in message
news: m...
> 'Dumbo' wrote, in part:
>>... I just want to know which one is most accurate so I can do a bit of
>>overclocking without the worry of burning anything out.
> _____
>
> It isn't possible to "burn anything out" by overclocking. As has been
> pointed out, the CPU will lock up and cool off long before any damage is
> done. Moreover, the 'fail-safe' on-CPU-die thermal diode is always
> functional. Raising the CPU core voltage too high can cause immediate
> destruction of the CPU, but damage from overheating just isn't possible.
>
> Phil Weldon
>
In addition to Phil's advise, errors and crashes show up long before you
will burn anything up. I have been overclocking since my first desktops
(wife says since dirt) and have never destroyed a cpu that I can remember.
Since the Pentium (maybe Pro??) there has been the protection from
overheating on Intel processors. I think AMD began after the TBirds with the
Athlon/XP series. I have also never seen the useful lifetime of a processor
shortened by overclocking. I intentionally ran this i7 at extremely high
temps mainly to see if it would throttle back or cause errors. In trying to
get to 3.8Ghz, the cpu would not begin throttling until it was at or just
below 100C. When it would throttle, the temps would drop to safer levels. I
could not get it to completely shut down or even cause errors. It would just
slow down a little. I was pushing the vcore a bit and like Phil said, that
is what can kill the processor and I wouldn't advise using voltages that
bring the temps up like that. I always try and read up on what vcore
voltages others have safely used and then go from there. Every processor can
be different and I have seen many that would run stable at much lower
voltages than others. Excessive heat may cause instability, but don't worry
about killing the processor.

Ed
>