My computer keeps crashing

[cali]

>> Sometimes when I run CPU intensive applications  the system will
>> crash at seemingly unpredictable times, I have to hard reset the  machine 
>> as it
>> is completely unresponsive, I was running an experiment in console  mode 
>> and it showed me the kernel panic:

>With those uptimes I would say your heat sink and fan(hsf) is to blame. 
>The old idea about amds running hot is kind of crap, any cpu will run  hot 
>if not installed correctly or overclocked.  You got that white  stuff 
>between the cpu and hsf?

I think I recall putting the white stuff in.

>I checked your hsf on the net and in silent mode it does not support  your 
>cpu speed, though it does in normal higher speed fan mode.  If you  have 
>the fan make sure to keep it in normal mode or it may bake your  cpu! 
>http://www.zalman.co.kr/eng/product/view.asp?idx=33&code=005009010

Damn! I should have researched this properly when I bought the heatsink. 
Thanks for that information, I'm lucky that you noticed this.

>Try this command several times after you boot.  Then after you boot  your 
>box do it while under load.

>sysctl -a|grep thermal

>That will tell you your cpu temp, you'll need acpi on.  If you did not  put 
>the hsf on right it will go up and you get problems like after 5  minutes 
>or less.  I set my bios heat alarm to go off and set a shutdown  temp too. 
>You might want to check that stuff out in your bios too.  Go  to amd.com 
>and get that pdf on how to install the hsf, I made a mistake  a month ago 
>when I was switching out cpus and that was my problem.

>Everything else looks good, but do you have some case fans?

OK, I think I had better invest in some, or some better cooling.

> I moved a  120mm fan over near my cpu and my 100% load temp while folding 
> droped  about 10C.  I am overclocked and it was maxing out at about 58C or 
> less.  Now it hardly hits 50C, usaully 48C but it might go down to 45C  if 
> it is cool in my room.  I wonder how it will do in the summer? :)

I used that sysctl command you suggested above and it says 55C-55.5C -- this 
is for when running underclocked.

I rebooted, put the CPU speed back to normal, left the fan on its 
dangerously low setting and then ran the program again, whilst checking the 
cpu temperature every second with:

while [ 1 ]; do sysctl -a | grep thermal; sleep 1; done

I observed the CPU temperature rise from a base of 50C at an approximately 
steady rate (I should have taken periodic readings too then I could have 
made a graph or something). It slowed down at about 57C (having took about 
3-4 minutes to get there) or so but carried on rising, 
58C...58.5C...59C...59.5C... kernel trap 11m33s (unfortunately I was setting 
up another process to run on another console so I never saw the final 
temperature).

This was with CPU thermal throttling enabled and set to 50% in my bios 
(although I'm not sure at which temperature it enables as it doesn't seem to 
say)

I turned the fan up to max, rebooted and ran the program again. The 
temperature seemed to stabilise around 52C.

Given this information I think it is highly likely that the temperature 
hypothesis is correct, and the reason for the crashing.

Thanks

cali