Re: Computer turns on and then just dies

On Fri, 12/6/2024 11:43 AM, s|b wrote:
> On Wed, 4 Dec 2024 14:14:07 -0600, VanguardLH wrote:
> 
>> CoreTemp shows you temps when you manually instigate investigation.  It
>> doesn't remain resident to issue alerts when temps get too high.  Be
>> careful during installation, though, as I've read it defaults to
>> installing bloatware.  Deselect some checkboxes during installation.
> 
> That can't be right?
> 
> <https://ibb.co/mvRw8Yn>
> 
> (BIOS said something around 26°C.)
> 

The developer claims to have made a fix for the 4C thing in the year 2012.

https://www.alcpu.com/forums/viewtopic.php?t=2297

In this CPU-Z for your processor, you can see that at idle,
and with a decent heatsink, the CPU can pull a low temperature
reading (29.3C) at idle.

http://valid.x86.fr/p8hcd8

At best, a temperature reading shouldn't even bother with digits
after the decimal. The uncertainty is from +/- 1C to +/- 10C,
depending on measurement method. The best method is the
"two current method", which involved two constant current
sources alternately switched into the measurement circuit.
This allows correcting for some offset term in one of
the sensor types. While this technical detail is known,
no hardware seems to be designed to do it that way.

Measurement methodology could also be calibrated for each
hardware instance, but nobody does that in a comprehensive way.
If there is a calibration done, it is a single data point,
and does not have sufficient information for a polynomial
best fit. Nobody takes, like, 10 readings while the silicon
die is in the test head at the fab, and records those in
some flash cells.

The purpose of temperature readings is mostly do determine
when there is grossly insufficient cooling. It's not an
attempt to make a weather station out of the PC hardware :-)
This is why, on an Intel CPU at idle, you can get a 15C
core reading, when the room is 23C. The reading could even
be 13C, and that is within the known behavior of those
circuits. The AMD measurement does not have the "high bias"
effect, like on Intel. On Intel, the measurement is
very accurate at 90C (1C uncertainty), and poor at room
temp where it does not matter (10C uncertainty).

*******

I still think, for contrast, you should run a CPU-Z and
click the About tab, and save a report. On my modern AMD,
I get a ton of temperature measurements (from the Infinity
Fabric), and these are not normally displayed in other
software. On AMD Ryzen Master, only the "die temperature"
(which is also in CPU-Z) is displayed, and they do not
bother showing individual core temperatures, as that
would be information overload.

My processor that I'm typing on, routinely has low
temperatures just above ambient at idle. And if you look at
the PWM controlled cooling fan on the CPU, the
cooling control also thinks the CPU is cool, as the
revolutions per minute is very low (maybe 400 RPM right now).
The CPU is basically running on one core, the min clock
is maybe 500MHz, and the CPU might be running at 800MHz
or so.

Like Van says, a load test will indicate how good your
cooling is. Even CPU-Z, in the Bench tab, it has a
Stress Test button, for leaving the benchmark running
and warming up the CPU. That raises my CPU to 60C die
temperature, and that is sufficiently below 90C for
me to be happy about it.

   Paul