Seasidepeter wrote:


With no overclocking, the cpu seems to idle at around 42/43C. Under the
torture test it rises to 62, at which point I stopped the test. Call me
a chicken, but I don't want to blow it up just yet!


The processor has automated protection features. The
absolute cutoff is provided by THERMTRIP, which will turn
off the power to the PC instantly. (That will cause
a dirty shutdown on your file system, and then CHKDSK
could start running on the next startup.)

So the processor should be protected against any
short term problems. I don't know what AMD uses for
throttling, and whether they change multipliers
if the processor gets too hot. But I believe both
AMD and Intel have THERMTRIP. And the trip point is
established purely in hardware, so a software crash
won't prevent the protection feature from working.

*******

You have to go out and do some data mining, to get a
reasonable safe value. Remembering that some laptops,
their hardware runs at close to 100C... :-) That would be
the silicon die temperature. The silicon itself, can
withstand 135C or so without parameter drift, and lower
temperature limits are caused by damage to the package
that holds the silicon die. If the processor came in
multi-layer ceramic (MLC), it might have been good up to the
135C number. The logic simulation of the CPU (i.e. proof
the design works), stops at around 105C or 110C.
Nobody does design verification at 135C or anything.
It would be too hard to close timing that high up.
The logic is proved to work at around 105-110C. And the
organic packaging used on chips now, probably can't take
that temperature, and a little less than that is safe
for the chip.

http://www.overclock.net/t/635091/ph...re-for-prime95

*******

There is a difference between silicon die temperature, and
socket temperature. On older hardware setups, the temperature
being monitored, was measured underneath the socket of the CPU.
Let's pretend that was 65C, just to make a numeric example.

The silicon die has a temperature differential to the outside
world. It could be around 25-35C higher than the socket
temperature. So if you were using a silicon die based measurement,
it would be 90-100C die temp.

So when you're looking for the "stable max", keep in mind
that the guys could be talking about socket or die temps
and could be implying different things. Socket temperature
may also be referred to in the documentation as Tcase_max.

I'm not going to do the data mining for you... because I'm
"lazy" :-) And it's getting hot here again.

Paul