HardwareBanter - View Single Post

**Paul[_28_]** · January 27th 20, 02:59 AM posted to alt.comp.hardware.pc-homebuilt

wrote:
I have an old HP business desktop that seems to have a problem with
overheating. One has to keep the heatsink dust-free in order to survive.
Now this is an aluminium heat sink and the CPU is TDP 95 W.

Looking at some Intel stock coolers, they seem to use aliminium for
65 W CPU, but a copper core for the higher powered products.
So were HP cutting corners?

There's the size and composition of the fins.

But there's also air velocity.

Above around 800 linear feet per minute, extra air
velocity doesn't help. 200 linear feet per minute is
a more common value for hardware design. We did do
one high velocity design at work, but you had to wear
protective earphones when in the lab next to it.
Pretty kooky.

Take the Noctua cooler that is cooling my 150W processor
right now. The air through it doesn't move all that fast,
but the heatsink has a *lot* of fins. They tried to make the
heatsink as quiet as possible, by making it bigger.

And you know all the thermal domains by now anyway.

1) THERMTRIP - processor is so hot, a thermal diode
inside the processor triggers a logic signal that
gates off PS_ON# and the ATX supply shuts off.

This protects a CPU when the heatsink falls off. Sometimes
the plastic tabs snap off and the heatsink falls off.

Before this protection was invented, a CPU could shoot up
past 200C in a matter of a couple seconds, after the heatsink
fell off.

2) Throttling - the processor inserts "No OPs" in the instruction
stream. The CPU does no useful work during these cycles, and
the CPU power consumption is reduced (but not eliminated) by
"doing nothing" for a clock cycle. This offers a way for the CPU
to "improve a bad situation" by "not drawing as much power".

Occasionally a CPU enters this state, despite having a very
nice heatsink. CPUs can have the silicon die soldered to the lid
(good) or a crumbly thermal paste is used inside the package, to
mate the CPU die to the cover above it. A void in the paste can
cause the CPU to throttle when running Prime95 for example.

3) All-OK - with a massive enough heatsink, maybe the die temp is
between 40-60C no matter what. And we call that "good enough".
Even if we're running Prime95, the CPU stays within the proper
range and we're happy.

A company like HP can just sit there and try heatsinks, and see
if at 25C or 35C room temperature, the CPU stays out of throttle
country. You don't have to use fancy engineering tools, if you
"cut and try" and stop when you just barely meet the specs.

Now HP *has* made good kit. We had a machine at work, pretty
ordinary looking, that we used to take into the thermal chamber.
It was rated for 50C room temp. We ran it up to 50C... and it
didn't crash! Amazing. So HP can make good gear. You can imagine
in that case, that the CPU would have a decent heatsink.

Some of the machines, put a high static pressure fan right
next to the CPU fins. And that thing runs at reduced voltage
most of the time. The fan can give 110CFM (noisy/lots of air)
if the CPU is too hot. Usually the fan only goes to the max
when the CPU has crashed and is not available to properly
program the fan to a lower setting.

We could use a copper slug aluminum heatsink and a moderate
flow of air, or we could use a cheap pure aluminum heatsink
and a *lot* of air.

And the reason this happens, is the aluminum really doesn't
conduct heat well enough. Maybe only 1" of fin next to the
base block actually transfers heat into the air.

And that's where heatpipes come in. If you go to 2" above the
base and run heatpipes through the fins, you can backfill the
fin and fill the top part of the fin. Now, three inches of fin
does work for us.

Heatpipes are the ultimate solution. Solid copper heat slugs
are pretty and all, but they're inferior to heatpipes.
Heatpipes and heat chambers with a few drops of a phase
change fluid, do an excellent job of moving thermal flux.

The original heatpipes, some of them used to leak. Today,
you're more likely to get most all of the pipes in
working condition. At one time, they didn't do a very
good job of sealing them. And that's a good reason to
use four or five pipes. If one leaked and dried out
and became useless, the others would continue to work.

The thermal transfer mechanism in heatpipes stops working
if you pump too many watts into the heatsink. The heat
transfer is "non-linear". Whereas the plain aluminum or
copper slug aluminum ones, those are "linear" coolers.
When a heatpipe cooler is overwhelmed, the entire heatpipe
is so hot, only vapor is inside the heatpipe and liquid
no longer condenses because the entire pipe is too hot.
If you jacked up the temp high enough, the heatpipe would
burst open from the pressure.

Paul