Re: What's an acceptable temp increase when overclocking?

Thanks for the detailed response, Paul. Much food for thought.

-phil

"Paul" <nospam@xxxxxxxxxx> wrote in message
news:nospam-0509062257570001@xxxxxxxxxxxxxxxx
In article <fyoLg.106982$du4.1180415@xxxxxxxxxxxxxxxxxxxx>, "Phil"
<phil_essing@xxxxxxxxxxx> wrote:

Oops, I meant to imply that the thermistor on the motherboard is the
source
of the BIOS and Probe CPU temp readings, not the core temps as read by
CoreTemp. But, it would seem that even that was incorrect. I guess I had
bad
information. I was completely unaware of this "normal thermal diode" and
that it is the likely source of the BIOS and Probe CPU temp readings.
Interesting. As for the rest of your insightful reply, it went
pretty-mush
over my head. Can you explain TCaseMax and how it differs from the a
chip's
thermal specification. For example: my E6600 has a thermal specification
of
60.1C, while CoreTemp displays a TCaseMax of 85C. I take it TCaseMax and
Thermal Specification are not the same thing.

Also, since you seem to know your stuff, can you explain why I get
dramatically different temps from the normal thermal diode, and the two
core
DTS? For example:

At idle:
CPU (normal diode): 27C
CPU core1: 43C
CPU core2: 44C

At load:
CPU (normal diode): 35C
CPU core1: 54C
CPU core2: 54C

Thanks.

-phil

http://www.intel.com/design/processor/datashts/313278.htm

In the Intel datasheet, there is a "thermal profile" graph, and it
relates power in watts, on the bottom axis, to "Tcase" on the left
hand axis. To me TcaseMax is where the chart hits the side of the
graph, and that is 60.1C for the E6600 and E6700 processors (4MB
cache). As I understand it, at that temp, throttling should
occur. (And throttling is one way to partially calibrate your
other temperature sensors.

Using a tool that can detect throttling, such as ThrottleWatch
or Rmclock, you could adjust the cooling on the CPU, to drive it
to the throttling point, and then you have some idea what the
temperature is. When throttling just begins, you look at your
normal diode and digital readouts, and record what they read
at the instant that throttling begins. If 60.1 is your maximum
temp, throttling then begins in an attempt to prevent the CPU
from overheating.

On previous processors, about 20C higher than that, the processor
should shut off (the signal is called THERMTRIP). Maybe the
TCaseMax shown in the program, is the temperature that the
THERMTRIP feature uses ? I don't think the THERMTRIP value is
even stated in the datasheet. (OK, I did find 85C in the datasheet
- it is the maximum "processor storage temperature". Not exactly
the right thing to be adding to the relative measurement, if that
is what is being done.)

And this is the problem with the relative temperature
measurement scheme. The user needs to know what value to
add to the relative measurement, to make an absolute
temperature readout. At the throttle point, the relative
readout should be 0C, and when the processor is quite
cool (say at idle), the relative readout is -20C. If we
add -20C to 60.1C, then an absolute temp of 40C might make
sense. What we don't know, is what arithmetic that hacked
temperature readout program is using, and whether it is adding
the correct value.

Another way to calibrate the thermal system, is to note
that the "normal diode" works even when the processor is
unplugged. All you need is a measurement current shoved into
the two thermal diode pins on the bottom of the processor.
You could place the processor in a distilled water bath,
with a mercury thermometer, and compare the measured diode
temperature with the thermometer. Several applied temperatures
would give you a graph suitable for curve fitting. Then, dry
the processor off and fire up both measurement methods.
Since you know what the normal diode is measuring, you
can use that to determine how far off the digital measurement
method is, and also whether throttling is happening at
the expected temperature.

It seems the more methods they invent to do this stuff,
the worse the results seem to be getting. I wonder what
it would cost to just slap an NTC thermistor inside the
processor packaging ? Some of those really are calibrated
to +/- 1C.

Paul


"Paul" <nospam@xxxxxxxxxx> wrote in message
news:nospam-0509061334580001@xxxxxxxxxxxxxxxx
In article <1%dLg.89092$du4.985575@xxxxxxxxxxxxxxxxxxxx>, "Phil"
<phil_essing@xxxxxxxxxxx> wrote:

Download CoreTemp and check your core temps. Each core has its own DTS
(Digital Thermal Sensor) located as close as possible to its hottest
point.
These temps will be hotter than what you see in the BIOS or PRobe, as
they
are read from a thermistor on the mobo, likely under the CPU socket.
You'll
want to make sure your core temps stay below mid-70s.

http://www.xtremesystems.org/forums/showthread.php?t=103638

-phil

According to this, the DTS are on the silicon die of the processor.
There is no need for any thermistors in the processor socket area.


http://www.tomshardware.com/2006/07/14/core2_duo_knocks_out_athlon_64/page7.html

DTS measures relative temperature - a reading of -20C from the sensor
means the processor core is 20 degrees C away from the thermal throttle
point. A reading of 0C relative, means the processor core is so hot
that
it is about to throttle. The job of any conversion program is two-fold.

1) Find the hardware support on the motherboard for DTS. I haven't
figured out yet, which chip holds this info. (At least the
Winbond chip that interfaces to DTS via the PECI pin, uses
the SMBUS.)
2) Compute Tcase_Max plus the DTS relative value, to get a true
die temperature. In reading the Intel datasheet, I don't see
an indication of how accurate this digital on-die sensor scheme
is. The Intel datasheet claims the absolute temp measurement
capability only goes down to 30C, meaning people with phase change
cooling will be getting the DTS error code 8002 hex instead of a
temperature reading. For air or water cooled systems, DTS should
still work.

Conroe also has the normal thermal diode on its silicon die, which
a motherboard maker can connect to the hardware monitor interface
on the Super I/O chip. The Intel datasheet says Conroe processors
do not include a correction factor for the thermal diode, like
previous processors did.

I would think any software tool reading the DTS, would be better
off just reporting the relative reading straight from the hardware.
A reading of -20C relative tells you that your processor is well
cooled. A reading of 0C relative tells you the processor is about
to throttle. So even the relative value, without any attempts
at calculation, would be better than nothing.

On this page, I can see Winbond has a hardware monitor chip
(W83793G) that sits on the SMBUS (yuck). It has a PECI input
on the chip, to get DTS readings from the processor.
Unfortunately, the datasheet is not available for download.

http://www.winbond-usa.com/mambo/content/view/142/271/

ITE also makes a couple of chips that connect to the PECI
pin, and they claim +/- 1C accuracy. Again, no datasheet,
and no way to verify what claims they are making. The +/-
1C could be for the old thermal diode on the silicon die.

http://www.iteusa.com/ite_news/press2_20060512-1.asp

It is also possible a motherboard may not even have a PECI
readout chip, but just have something to fake out the
interface. To understand whether that is feasible, I'd need to
find more detail on PECI - the Intel processor datasheet is
not too helpful.

Intel also keeps track of whether PECI works on a given
processor. Here is an example from the processorfinder
site, and says "These parts are PECI enabled":

http://processorfinder.intel.com/details.aspx?sspec=sl9s7

I think sticking a finger on the processor heatsink,
gives you more accurate info than all the fancy sensors :-)
All this extra complexity and for what purpose...
I guess the old diode sensor on the silicon die,
just wasn't enough fun.

Paul



"greysky" <greysky@xxxxxxxxxxxxx> wrote in message
news:Ga8Lg.24259$gY6.12800@xxxxxxxxxxxxxxxxxxxxxxxxxxxxx
I have a new system running a 2.4 GHz core 2 duo. Normally my cpu
temperature is under 39 C when running at stock speeds. When I run
the
FSB
at 420 MHz my cpu temp increases to 50 C. My question is what is
considered an acceptable temperature increase for these new Intel
chips?
Should I back things down or is 50 C ok? TIA.




.

Relevant Pages

  • Re: Whats an acceptable temp increase when overclocking?
    ... that it is the likely source of the BIOS and Probe CPU temp readings. ... other temperature sensors. ... DTS measures relative temperature - a reading of -20C from the sensor ...
    (alt.comp.periphs.mainboard.asus)
  • Re: Whats an acceptable temp increase when overclocking?
    ... that it is the likely source of the BIOS and Probe CPU temp readings. ... the DTS are on the silicon die of the processor. ... DTS measures relative temperature - a reading of -20C from the sensor ...
    (alt.comp.periphs.mainboard.asus)
  • Re: System lock ups possibly due to heat
    ... Do a Google for Rightmark CPU Clock Utility ... My core 2 duo with an aftermarket Thermalright Cooler and 120mm fan never goes above 55 running stress test for hours. ... The room temp also comes into play as the Northbridge could be getting too hot and causing your shutdown...try taking the ... Your CPU temperature under load seems high to me, so consider removing, then ...
    (microsoft.public.windows.vista.hardware_devices)
  • Re: Many problems with a P5GD2-X board
    ... > I tested the temp in the case with a little thermometer. ... > BOINC (to always use 100% of CPU) ... > All these temperature seems pretty good to me. ... Your board is probably too new, to expect Motherboard Monitor ...
    (alt.comp.periphs.mainboard.asus)
  • Re: Board runs 44C all the time, is this TOO HOT? please help me....
    ... MB temp at 44C is nothing to be alarm (I would assume that you use PC ... The important thing is what your the CPU ... temperature - probable much hotter than your MB temp? ... I have one front intake fan + PS exhaust fan + one exhaust ...
    (alt.comp.periphs.mainboard.asus)