MBM and Heatsinks: A Discussion

Over the last couple days, some people have received Glaciators. Some have found that their MBM/thermistor readings on
the Glaciator are higher than their older heatsink.

(We use “MBM” as a generic term for any kind of temperature monitoring that involves using motherboard sensors, including
in-socket thermistors. Other monitoring programs, including those provided by mobo manufacturers, have the same problems and
inaccuracies.)

This is not surprising since our tests show that the Glaciator puts up temperatures shows MBM temperatures 5C or better higher
than actual measurement of the front of the CPU, while some other heatsinks generally show MBM temperatures at or below actual
CPU temp.

I asked Andy LeMont to explain why that is the case, and he responded with an email in which he discussed some of the design
considerations that went into the Glaciator, along with some of the ways MBM can be easily manuipulated. We would welcome similiar
comments from any other heatsink designers.

Here’s what he had to say (some editing done for clarity).


C/W ‘s vs MBM results were bound to create a controversy and you were correct
in your assumption (that people would judge solely on MBM numbers). You yourself wanted to see Joe actually do
his testing. This is a difficult sale with those who only take quick
cursory views of the available data.

Each heat sink has it’s own individual airflow personality.
Some are radial, some are bi-directional. Some blow down
on the board, some suck air off the board.

A bi-directional heatsink like the Glaciator has a
more focused thermal exhaust than a radial heat sink
like the Thermoengine, with twice the distribution area
of the Glaciator.

The air coming off the Glaciator is hotter
and slower-moving than some others due to slower fan speeds. Air coming
off heatsinks interacts with other components on the
board. That then influences the temperature on the on-board sensor.

During preliminary development tests with Joe Citrella the
initial Glaciator design was rotated 90 degrees to what
it is now, so the heated heat sink exhaust flow impacted
the caps on the ABIT board.

At that time, the MBM read 60C, or 20C higher than the test group of 6 or seven sinks we had
for comparison, even though the C/Ws were in the 0.2s, which was lower
than any of the other heatsinks. Joe
told me this was a no go, even though the C/W’s were
lower than anything else we had tested.

(Ed. note: Just based on where the air was coming off the heatsink, nothing to do with
actually cooling the CPU, a better heatsink did 20C worse in MBM than inferior heatsinks.)

After observing that other bi-directional sinks were oriented 90 degrees
opposite of what I had, I got out a hacksaw and cut
another clip channel re-sanded the base to accommodate
the cambox relief and ran another test. My MBM reading
dropped 20 degrees C despite removing 20% more surface
area, and my socket temps dropped into the range of the other
sinks.

(Ed. note: In other words, he made the heatsink worse, but got a much better MBM temperature.)

I am probably one of the few people that has had the
luxury of testing different versions of the same sink and
observing their effect on real as compared to MBM temperatures.

I can manipulate MBM at will depending on what
changes I make, and those changes in MBM temperature can be
the exact opposite of the real changes in CPU cooling performance.

Decrease the fin density from
20 to 14 fins. and you will see MBM drop 7 degrees. Why?

When you do that, the volume flow of the air is higher and cooler,
making the feedback back through secondary components
cooler and thus lowering the MBM temperature. Unfortunately,
that also lower the performance of the heatsink for what it’s actually
supposed to do.

I performed a demo for Joe on this, while C/W
was 25% higher, Motherboard Monitor was great. I even
briefly discussed a design strategy with Joe of
designing the sink for MBM review tests but was thrown
out as C/W was considered the primary determinant.

(Ed. note: In other words, the choice was between
a heatsink that actually did the job better, or one that came up with better MBM
numbers. The choice was made for a better heatsink, rather than one
that was actually worse but provided “better numbers.”)

Is a higher back temperature in itself a problem?

Our testing found some stability issues when the difference between
the front and back temperature differed by more than 10C
with high case temps. That will make your
system crash faster no matter what your C/W is. Joe and
I ran some preliminary tests on this and concluded that
any thing less than 10 C was acceptable although we
think it deserves more research time permitting.

Your test comparison if you do not have additional test equipment should be how high you can overclock and
compared to other sinks in the same format.

(Ed.note: We’ve seen people report that although they get higher MBM numbers, they can overclock more.
Others don’t. There comes a point with any CPU where it isn’t going to go any further within this temperature range.)

I prefer static testing with a heater or a bench test because the distortions
introduced by motherboards are removed. I can see the power(watts) going in
and the Temp (watts) coming out. You have a controlled test you can duplicate
with other CPUs and motherboards and any problems or variations with the heatsink
itself rather than some motherboard effect are
instantly identifiable.

But you also need other reference
points to get a clear picture of where it is you stand.

So you test on different motherboards, different CPUs
etc. and the data will exhibit patterns that are common
to each individual motherboard, so if there happens to be a
big variation due to motherboard effects, it can be noted.

But even on-die C/W s on motherboards aren’t perfect. They can
be fine for comparative purposes, but they understate true C/W.
Why?

Different board effects can leave some motherboards running cooler.
CPU stress programs do not fully stress the cpu (they generally
only push the CPU to 85-90% of maximum wattage; (Benchtest.com
has a some good info on this.) Again, see how high you can overclock should be
the final determination, if it is really hotter you will
not get there from here.

If I relied on the MBM for all my design criteria I
would have ended up with low MBM results but a higher
core temp. In short, better numbers, but a worse overclocking heatsink.

It is the full range of testing that will
give you the total picture. Using any one test as the only test
is like leaning on a reed.

You have a mountain a mile high. 10 sinks are 1
foot from the top 3 of them are 10 in. from the top. On
a given set of tests you are looking +/- 10%. Change fan
speed a little bit and 3 move up an inch. Change test
conditions and two move down an inch. This is all
turning on a degree or two and is rather silly from the
who is best standpoint.

For the range of temperatures we’re talking about, MBM probably has an error
rate of +/- 50%. A while back, Burning Issues had a couple articles (here and here) which reviewed the
issue of accuracy of sites that use MBM; some were off by 100%.

(Ed. note: The site may not be in existence any longer, but a graph based on some of
the data can be found here.

My design goals on the Glaciator were high performance,
ease of use with low noise. MBM not withstanding I
believe that is what I ended up with.

–Andy Lemont

Email Ed

Be the first to comment

Leave a Reply