joanne said:
Hi I Know I asked a similar question before but I now have info
I am changing a 90nM AMD 64 3200+ Winchester core to a AMD X2 4200+ 90nM
Manchester dual core. The 3200 runs at 38 deg and has a heat load of 67
watts the 4200 is rated at 89 watts.
Will my current heatsink be enough for this CPU?
It probably runs at 38 degrees at idle.
To characterize cooling solutions, you load the CPU up with software
to run. Prime95 can run the processor at 100% load. The processor
could run hotter that way, and that is how you determine whether
it is in trouble or not, while it is at 100% loading.
Heatsinks are rated according to a number of things. The simplest
characterization is called "theta_R" and would be valid at
one fan speed (like full speed).
To work an example:
Say the room temperature is 25C.
Now, say the inside of the computer case is 32C.
(The computer case temperature is the important one, because
the heatsink is trying to dump the heat into that ambient
working temperature inside the case.)
OK. Say I buy myself a heatsink with theta_R = 0.20C/W (degrees C per watt)
Further, say the CPU is 67W. I run my 100% loading software, such as
Prime95 or CPUBurn or an equivalent piece of software. Now I want
to calculate the expected CPU temperature.
Expected = case_ambient + ( theta_R * CPU_Power_watts )
= 32C + ( 0.20C/W * 67W )
= 45.4C
Now, say I use a higher power processor, like an 89W one.
= 32C + ( 0.20C/W * 89W )
= 49.8C
It didn't make too much difference in that case, because
the heatsink is a pretty good one. Some of the retail heatsinks
that ship with processors, have been poorer, at about 0.33C/W
as an example.
*If* you had run the CPU at 100% loading on your 3200+, then it
would be possible to measure the ambient case temperature (with
something like Speedfan), and solve for theta_R. (You know both
the case temp and the CPU temp, and the book value for the
processor TDP watts.) Then, you could substitute the new power
number (89W) and get an estimate of what to expect with a
different processor.
For a flexible version of Prime95, try this one. It will work
with single or multiple core processors, and run one thread on
each core. The first time you run it, it will ask if you want
to "join GIMPS". Say no. Then, a custom dialog will pop up.
Examine the total memory to be tested. On my 1GB computer,
it will recommend a number like 767MB of memory to test.
Change the number to something smaller than the recommended
value, as then there will be a bit of memory left for your
other software to be run. Say, try 200MB for example.
The left-most menu has "stop" and "exit" options, to get
rid of the program when you're done thermal testing.
http://www.mersenne.org/gimps/p95v255a.zip
The above characterization exercise is only as good, as the
real power dissipated, matches the "book" value of 67W. I've
measured my processor, and it happens to come pretty close to
the book value. But it doesn't have to. So working out
your theta_R this way, will not be particularly accurate,
but it should allow you to work out a ballpark figure for the
impact of 30% more heat. But to do the measurement, you
need a means to measure the case air temperature, the CPU
temperature, and know the number of watts the CPU is
using at the time. Then you can get the magic theta_R value
in degrees C per watt.
Paul
