 | |
CPU temperature.... too hot? :o(
Hardware (General Discussions) General Hardware Discussions (alt.comp.hardware)
 |
| | Thread Tools | Search this Thread | Display Modes |
#1
 
| |||
| |||
|
CPU temperature.... too hot? :o( -
05-17-2007
, 03:39 AM
#2
| |||
| |||
|
|
Hi people I hope you can help with my dilema... I recently replaced my P4 3.2 with a P4 3.6. Firstly I used the original heat-sink and the CPU temp was 45-50c idle. I then obtained an Artic Freezer 7, and now my CPU runs at 55-60c idle??? (Idle = 0-5%) I have re-seated the sink twice, double-checking the spread of the thermal paste (Arctic Silver) which was applied following the instructions. The CPU fan seems to run faster than normal most of the time to keep the temp down  (What am I to do? Thanks. Demon. Sandra Report: Full Report in HTML Format: http://www.alienmoons.com/public/misc/sandra.htm Plain Text: snippage |
Intel processors have a thermal throttle feature. You didn't report
your load temperatures, but if while the CPU is 100% loaded, you
are hitting 70C, the processor will slow down its compute rate,
to try to keep cool. So that kind of defeats the purpose of
having a fast processor. Under load, I'd want to stay under 65C
at least. And preferably do better than that.
For a tool to watch throttling, there is this one. Throttling sets
a bit in the processor, and this program (RMclock) is supposed to read
and display what is happening.
http://cpu.rightmark.org/download.shtml
http://cpu.rightmark.org/download/rmclock_225_bin.exe
For a CPU cooler to work, the air inside the computer case
has to be cool. If you wanted to tell a friend how well your
cooling would work, you would:
1) Load the CPU to 100%. Prime95 Torture Test, Orthos SP2004, CPUBurn
are some examples. There is also a piece of software with S&M in the
name, but I don't remember its full name. Let the program run
for 10 minutes, so the temperatures in the case stabilize.
2) Record room temperature. Record the case air temperature.
Record the CPU temperature.
Now, we work an example. Say the room temperature is 25C, the
computer case air temperature is 32C, and the CPU is 64C at
100% load.
We compute (case - room). This is 32C - 25C = 7C. A difference
of only 7C to 10C is a well cooled computer case. If the
difference is 20C, you need more case fans or need to open
up the ventilation area that feeds the fans. (A fan cannot work,
unless there is a path for the air to go). I opened more
grille area on the front of my computer, by removing the
plastic trim piece from the front.
The second thing to compute, is (CPU - case). In the example,
this is 64C - 32C. The difference is huge and is 32C. Now,
say we look up the processor's rated power dissipation, and it
is 100 watts (what it draws roughly at 100% load). Divide
32C by 100 watts, gives 0.32C/W. That is the thermal resistance
of your current cooler.
(You can look up your SLxxx code off the processor box, here.
The processor power will be listed in watts.)
http://processorfinder.intel.com
If I go to the Zalman web site, they rate their products
according to their thermal resistance or theta_R. If we run the
fan on this one at 100%, the thermal resistance is 0.12C/W.
That is the spec Zalman lists.
http://www.zalman.co.kr/eng/product/...x=165&code=005
So, now we want to know how much cooler our CPU would be
with the Zalman CNPS9500 LED. Using the thermal resistance
and the power rating, we multiply 0.12C/W * 100W = 12C.
Add 12C to the case air temperature of 32C, gives 44C.
And that is the expected improvement by changing coolers.
We would go from 64C to 44C CPU temperature, with that new cooler.
Now looking at that example, our other option, would be
to reduce the computer case air temperature. The case
air in this case is 32C, and there is hardly any room
for improvement. The delta is only 7C, and even if the
case was fitted with a 12" diameter desk fan, it would
not get much cooler.
By looking at the two deltas, you can see which place to
make an improvement. If the case delta was 20C, you'd fit
some more fans and reduce the delta to 10C. If the CPU
delta was huge, you'd find a way to reduce it.
Sometimes, coolers simply are no better than the retail unit.
It happens. Most coolers lack a theta_R on their web site, so
you cannot judge them unless you can find a review on a
web site that knows how to review them properly. The
volume the heatsink occupies can give you some hint as
to its performance, as long as both use the same tech,
such as heat pipes. A small sink is a risky purchase.
If the surface of the processor is not flat, the thermal paste
spreading pattern may give you a hint. A lack of flatness,
or a crooked fit, will affect how well it cools. Surface
flatness has been a problem for some recent processors.
If the push pins don't snap properly, you may not have the
proper retention force. The nice thing about paste, is it
doesn't need a lot of pressure to work. But if the metal
is not flat (either the cooler or the CPU), then it isn't
going to work very well.
I don't recommend lapping, unless you have some clue about
the consequences. Depending on how you do it, you could
round the top of the processor, void the Intel warranty,
and end up worse off than when you started. Practice on
something first, rather than ruin your good processor.
Generally speaking, I wouldn't bother, because I'd
probably mess it up.
Personally, I would sooner just put a few dollars into a
better cooler. Everyone has their own favorite solution,
and I'm sure others will contribute their favorite method.
The above gives you a framework to judge how things are
going. If the processor is getting throttled, you should
at least improve it enough to stop that from happening.
Now, since I don't know the HP model number, I cannot
comment on the custom design of the cooling in your
computer case. An idea situation would be if the CPU cooler
could blow its exhaust towards the rear exhaust fan on the
computer case. There are a few tricks like that you
can use, to help ease the hot air out the back of
the case. But it all depends on how fancy the cooling
solution is on the computer, as to how easy it is to
retrofit an improved cooling solution.
Paul
#3
| |||
| |||
|
|
Hi people I hope you can help with my dilema... I recently replaced my P4 3.2 with a P4 3.6. Firstly I used the original heat-sink and the CPU temp was 45-50c idle. I then obtained an Artic Freezer 7, and now my CPU runs at 55-60c idle??? (Idle = 0-5%) I have re-seated the sink twice, double-checking the spread of the thermal paste (Arctic Silver) which was applied following the instructions. |
as extraction makes all the difference.
what happens when you take the side of the case off and run the pc?
#4
| |||
| |||
|
|
Hi people I hope you can help with my dilema... I recently replaced my P4 3.2 with a P4 3.6. Firstly I used the original heat-sink and the CPU temp was 45-50c idle. I then obtained an Artic Freezer 7, and now my CPU runs at 55-60c idle??? (Idle = 0-5%) I have re-seated the sink twice, double-checking the spread of the thermal paste (Arctic Silver) which was applied following the instructions. The CPU fan seems to run faster than normal most of the time to keep the temp down (What am I to do? Thanks. Demon. |
IMO, the Arctic Freezer 7 might not be the ideal HSF for the P4 3.6 CPU.
While the concept is good of using heat tubes and large, finned cooling
areas, there are issues with drawing cooling air through them. Both the
Asus Silent Square and the new Thermaltake V1, for example, have large
fans inserted between the banks of the radiator. We have found them ideal
for the Pentium-D 950's, which are notorious for hot running.
#5
| |||
| |||
|
#6
| |||
| |||
|
|
Firstly. Thanks very much for some excellent replies, they are very much appreciated. Ok. I have done some tests. I used Sandra first of all to test LOAD temp of the CPU, but Sandra just stops all real-time monitoring, which is useless. I used the rightmark software to run some tests: Running RMMT (Thread test) which loaded the CPU 100% for 2 minutes, the max temp the CPU reached was 66c, with the fan picking up speed at about 60-61c Sometimes SensorView reports the CPU running at 255c for a few seconds, but when this happens the clock speed is not stepped-down so is this a false "blip" reading? Case temp is about 32c, GPU 45c Room temp is 23c I will try another test without the case on. I need to reboot as SensorView is refusing to find my sensors, which it sometimes does. SensorViews fault, or the m/board? .............. I tried to use Rightmark Memory stability test at it just "Fails to allocate 2047mb of memory" ??? Also, using the memory analyser (RMMA it says the CPU bus speed is 200mhz is this correct? Is this referring to the FSB, which should be 800mhz? Regards Demon. |
tried to read the sensors at the same time, the readings would get corrupted.
You could get a bad value that way. But more modern boards use the LPC
(low pin count) bus, and one program cannot interrupt the reading being
taken by another program, as far as I know. In any case, I would treat a
reading which is way off, as being wrong.
A temperature of 66C is still pretty warm, especially considering your
room temp is 23C. If the room gets hot in summer, you'll be in
"throttle country". I'd still want to work on the cooling a bit.
And the case delta of 9C, means the case cooling is in reasonably
good shape.
I'd try running the fan on the CPU cooler with the full 12V feeding it,
if you aren't already doing that.
If you need a RAM tester, there is memtest86+ (which tests all memory), or
you can use programs like Prime95 and its Torture Test. Microsoft also has
a memory tester with characteristics similar to memtest86+ (self booting,
no OS).
http://www.memtest.org/ memtest86+
http://www.mersenne.org/freesoft.htm prime95
http://oca.microsoft.com/en/windiag.asp Microsoft memory tester
A Pentium 4 processor has a "quad-pumped" bus. Four pieces of data are
passed per clock cycle. FSB800 thus comes from a 200MHz clock, so the reported
results make sense. A 200MHz signal passes from the clock generator
chip, to the CPU. The CPU does bus transfers 800 million times a second
in response. The P4 bus is 64 bits wide or 8 bytes. 8 bytes times 800 gives
6400MB/sec theoretical max transfer rate. A couple PC3200 DIMMs in dual
channel fits that perfectly.
Paul
#7
| |||
| |||
|
|
DemonTraitor wrote: Firstly. Thanks very much for some excellent replies, they are very much appreciated. Ok. I have done some tests. I used Sandra first of all to test LOAD temp of the CPU, but Sandra just stops all real-time monitoring, which is useless. I used the rightmark software to run some tests: Running RMMT (Thread test) which loaded the CPU 100% for 2 minutes, the max temp the CPU reached was 66c, with the fan picking up speed at about 60-61c Sometimes SensorView reports the CPU running at 255c for a few seconds, but when this happens the clock speed is not stepped-down so is this a false "blip" reading? Case temp is about 32c, GPU 45c Room temp is 23c I will try another test without the case on. I need to reboot as SensorView is refusing to find my sensors, which it sometimes does. SensorViews fault, or the m/board? .............. I tried to use Rightmark Memory stability test at it just "Fails to allocate 2047mb of memory" ??? Also, using the memory analyser (RMMA it says the CPU bus speed is 200mhz is this correct? Is this referring to the FSB, which should be 800mhz? Regards Demon. There was a time when sensors were connected to the SMBUS. If two programs tried to read the sensors at the same time, the readings would get corrupted. You could get a bad value that way. But more modern boards use the LPC (low pin count) bus, and one program cannot interrupt the reading being taken by another program, as far as I know. In any case, I would treat a reading which is way off, as being wrong. A temperature of 66C is still pretty warm, especially considering your room temp is 23C. If the room gets hot in summer, you'll be in "throttle country". I'd still want to work on the cooling a bit. And the case delta of 9C, means the case cooling is in reasonably good shape. I'd try running the fan on the CPU cooler with the full 12V feeding it, if you aren't already doing that. If you need a RAM tester, there is memtest86+ (which tests all memory), or you can use programs like Prime95 and its Torture Test. Microsoft also has a memory tester with characteristics similar to memtest86+ (self booting, no OS). http://www.memtest.org/ memtest86+http://www.mersenne.org/freesoft.htm prime95http://oca.microsoft.com/en/windiag.asp Microsoft memory tester A Pentium 4 processor has a "quad-pumped" bus. Four pieces of data are passed per clock cycle. FSB800 thus comes from a 200MHz clock, so the reported results make sense. A 200MHz signal passes from the clock generator chip, to the CPU. The CPU does bus transfers 800 million times a second in response. The P4 bus is 64 bits wide or 8 bytes. 8 bytes times 800 gives 6400MB/sec theoretical max transfer rate. A couple PC3200 DIMMs in dual channel fits that perfectly. Paul- Hide quoted text - - Show quoted text - |
A mate of mine has his own PC cooling business, I will get in contact
with him and get him to see what he can do to cool "me" down a bit.
You are right - my old 3.2 set-up used to have the fan on full during
the summer, and if this is "fanning" already then I need to get
something sorted. Now that I know the CPU is definitely running warm
etc, I can maybe get it looked into.
Quote:
|
I'd try running the fan on the CPU cooler with the full 12V feeding it, if you aren't already doing that. |
BTW: this is a pic of my naked PC... (x-rated):
http://www.alienmoons.com/public/images/_MG_0001.JPG
Regards
Darren
#8
| |||
| |||
|
|
Thanks again for your advice - very very much appreciated. A mate of mine has his own PC cooling business, I will get in contact with him and get him to see what he can do to cool "me" down a bit. You are right - my old 3.2 set-up used to have the fan on full during the summer, and if this is "fanning" already then I need to get something sorted. Now that I know the CPU is definitely running warm etc, I can maybe get it looked into. I'd try running the fan on the CPU cooler with the full 12V feeding it, if you aren't already doing that. ?? uh? :$ Do you mean having it on FULL all the time? BTW: this is a pic of my naked PC... (x-rated): http://www.alienmoons.com/public/images/_MG_0001.JPG Regards Darren |
with it, to turn it down.
If your friend is in the PC cooling business, he probably has a
$50 chunk of copper and heatpipes, that will fix you up.
Paul
|
« Previous Thread
|
Next Thread »
| Thread Tools | Search this Thread |
| Display Modes | |
Linear Mode
| |
Powered by vBulletin Version 3.5.3
Copyright ©2000 - 2008, Jelsoft Enterprises Ltd.
Copyright ©2000 - 2008, Jelsoft Enterprises Ltd.