M
Matt
Vin said:
I haven't followed the whole thread, but we should know whether you
upset the contact between the HS and CPU somehow, maybe by bumping or
removing the HS.
I haven't followed the whole thread, but we should know whether you
upset the contact between the HS and CPU somehow, maybe by bumping or
removing the HS.
T
Thomas
Strontium said:Perhaps I'm the living deadI should be dead, by all rights!
Having worked (unsafely, I might add!) with many carcinogens and
toxic solvents for the better part of 14yrs....
Whagh, our amanuensis (lab assistent) managed 30 yrs already, and still
breathing ;-) I feel a bit queer after sniffing a bit lot of Acrylonitryl
one time, but i think it's normal, hehehe
Thomas
S
Strontium
AN is some nasty stuff! Just don't get it on your skin 
I worked for
Monsanto (after they spun off into Solutia)....the makers of AN used for
Wear-dated carpet. A catalyst, in the reaction, and byproduct is liquid
HCN. I loved going to work, knowing I was going to be handling liquid HCN
(they sampled the liquid HCN tanks once a week for GC and color)...was kinda
neat! AN, depending on it's grade can contain lethal amounts of HCN.
Smells like bleach (as pure HCN does, also). Just don't take too big of a
whiff hehehe.
-
Thomas stood up at show-n-tell, in 7wM6b.36380$tK5.4322954@zonnet-reader-1,
and said:
I worked forMonsanto (after they spun off into Solutia)....the makers of AN used for
Wear-dated carpet. A catalyst, in the reaction, and byproduct is liquid
HCN. I loved going to work, knowing I was going to be handling liquid HCN
(they sampled the liquid HCN tanks once a week for GC and color)...was kinda
neat! AN, depending on it's grade can contain lethal amounts of HCN.
Smells like bleach (as pure HCN does, also). Just don't take too big of a
whiff hehehe.
-
Thomas stood up at show-n-tell, in 7wM6b.36380$tK5.4322954@zonnet-reader-1,
and said:
D
David Maynard
w_tom said:
Center of what? The heatsink? Well, close since the heatsink is typically much
larger than the CPU die, or even the IHS, but it's also not (usually) precisely
in the 'center'. The heatsink is often offset due to usually extending on over
the socket hinge cam section. Center of the CPU die? as in the case of an Athlon
XP? No. It depends on the die layout and what portions are being used at any
point in time.
The P4 IHS is flat to within 50 um. The Athlon die is essentially 'flat'. Where
is the point of "most pressure?"
The celeron PGA IHS is slightly concave with maximum height on the edges. But
then it not only has to contact the heatsink but the CPU die, down there
underneath in the center.
W
w_tom
Do the entire thermal circuit. Calculate the numbers. 9
degrees is not a serious improvement. Exception is
overclocking which means no valid numerical specifications are
available anyway. Therefore no reliable calculations can be
performed.
9 degrees must be well below what any CPU and heatsink
assembly does in a system running in a 100 degree room. Any
properly constructed system works just fine in a 100 degree F
room.
But when overclocking, then one no longer has any idea of
the heat produced by CPU, a what temperature makes internal
CPU electronic timings unstable, and other parameters. These
are not parameters that damage hardware. These are parameters
that determine CPU stability. Since no calculations can be
performed, then even those trivial 9 degrees might be
significant.
First running that system without thermal compound will
demonstrate how effective that CPU/heatsink interface really
is. More that thermal compound reduces CPU temperature, then
the more inferior that heatsink really was. Just another way
of finding which heatsinks have superior surface machining -
before improving heatsink performance with a least amount of
thermal compound.
If 9 degree C is of significance to a standard clocked
system, then the system has far more serious problems; not
thermal problems. And testing a heatsink without thermal
compound can go a long way to verifying the real integrity of
that heatsink - something that any overclocker should want to
learn.
"Degree C per watt" is an overall number. The heatsink
without and with thermal compound tests but one aspect of that
overall heatsink performance.
degrees is not a serious improvement. Exception is
overclocking which means no valid numerical specifications are
available anyway. Therefore no reliable calculations can be
performed.
9 degrees must be well below what any CPU and heatsink
assembly does in a system running in a 100 degree room. Any
properly constructed system works just fine in a 100 degree F
room.
But when overclocking, then one no longer has any idea of
the heat produced by CPU, a what temperature makes internal
CPU electronic timings unstable, and other parameters. These
are not parameters that damage hardware. These are parameters
that determine CPU stability. Since no calculations can be
performed, then even those trivial 9 degrees might be
significant.
First running that system without thermal compound will
demonstrate how effective that CPU/heatsink interface really
is. More that thermal compound reduces CPU temperature, then
the more inferior that heatsink really was. Just another way
of finding which heatsinks have superior surface machining -
before improving heatsink performance with a least amount of
thermal compound.
If 9 degree C is of significance to a standard clocked
system, then the system has far more serious problems; not
thermal problems. And testing a heatsink without thermal
compound can go a long way to verifying the real integrity of
that heatsink - something that any overclocker should want to
learn.
"Degree C per watt" is an overall number. The heatsink
without and with thermal compound tests but one aspect of that
overall heatsink performance.
W
w_tom
We provide the simplest instructions to those who follow
instructions without knowing the full story. Experience
without understanding the underlying theoretical science means
such people are ripe for junk science reasoning. Even I would
tell the naive hobbyist to use thermal compound because many
don't want to even bother to learn the whys and why nots.
Intel and AMD do same. They tell hobbyists to use thermal
compound regardless of whether it is really necessary.
But, for example, one Intel engineering paper demonstrated
no advantage to using thermal compound on higher heat
generating semiconductors. Correct. High heat semiconductors
demonstrated no significant advantage over a bare heatsink to
CPU interface. Intel went on to discuss other superior
concepts beyond the scope of this discussion. But if a
hobbyist does not even know if his heatsink is machined; if he
only buys on price, then why tell him any of this. Better to
have him use thermal compound.
Not everyone uses thermal compound. In a previous
discussion, one found no thermal compound on his Intel CPU /
heatsink assembly direct from the factory. Not a problem. It
is rather hit or miss as to whether to apply thermal
compound. Many of our custom designs did not use it because
complications from thermal compound caused other reliability
problems. Semiconductors come both ways - with and without -
because thermal compound only provides a small additional
advantage.
These advantages become glaringly obvious once theoretical
numbers confirm what the product does.
There is a difference between theoretical science and
applied science. If both are not used, then failures are a
probability. What is unique in previous posts? Both
theoretical and applied science were used. As a result, a
number of points were made:
1) that thermal compound must be applied so sparingly that CPU
makes mostly a direct contact with heatsink. So little
thermal compound that it does not spread much into the outer
half of CPU.
2) if heatsink is properly machined, then heatsink can be
applied to CPU without any thermal compound. If properly
machined, then thermal compound would only result in single
digit temperature decreases.
3) many heatsinks are sold even without the essential "degree
C per watt" number. Many don't even know how good their
heatsink really is OR how much better it would be if properly
mated to CPU. That test first without thermal compound, then
with goes a long way to learning how good a heatsink really
is.
4) Arctic Silver is overhyped. Most thermal compounds do for
dimes what Arctic Silver does for dollars. But then Arctic
Silver also does not make numerical specifications easily
available - which should be the first indicator that Arctic
Silver is hiding something. Products sold without numerical
specs should be routinely suspect. Arctic Silver is mostly
sold on hype - engineering specs be damned when your customers
too often fear the numbers.
instructions without knowing the full story. Experience
without understanding the underlying theoretical science means
such people are ripe for junk science reasoning. Even I would
tell the naive hobbyist to use thermal compound because many
don't want to even bother to learn the whys and why nots.
Intel and AMD do same. They tell hobbyists to use thermal
compound regardless of whether it is really necessary.
But, for example, one Intel engineering paper demonstrated
no advantage to using thermal compound on higher heat
generating semiconductors. Correct. High heat semiconductors
demonstrated no significant advantage over a bare heatsink to
CPU interface. Intel went on to discuss other superior
concepts beyond the scope of this discussion. But if a
hobbyist does not even know if his heatsink is machined; if he
only buys on price, then why tell him any of this. Better to
have him use thermal compound.
Not everyone uses thermal compound. In a previous
discussion, one found no thermal compound on his Intel CPU /
heatsink assembly direct from the factory. Not a problem. It
is rather hit or miss as to whether to apply thermal
compound. Many of our custom designs did not use it because
complications from thermal compound caused other reliability
problems. Semiconductors come both ways - with and without -
because thermal compound only provides a small additional
advantage.
These advantages become glaringly obvious once theoretical
numbers confirm what the product does.
There is a difference between theoretical science and
applied science. If both are not used, then failures are a
probability. What is unique in previous posts? Both
theoretical and applied science were used. As a result, a
number of points were made:
1) that thermal compound must be applied so sparingly that CPU
makes mostly a direct contact with heatsink. So little
thermal compound that it does not spread much into the outer
half of CPU.
2) if heatsink is properly machined, then heatsink can be
applied to CPU without any thermal compound. If properly
machined, then thermal compound would only result in single
digit temperature decreases.
3) many heatsinks are sold even without the essential "degree
C per watt" number. Many don't even know how good their
heatsink really is OR how much better it would be if properly
mated to CPU. That test first without thermal compound, then
with goes a long way to learning how good a heatsink really
is.
4) Arctic Silver is overhyped. Most thermal compounds do for
dimes what Arctic Silver does for dollars. But then Arctic
Silver also does not make numerical specifications easily
available - which should be the first indicator that Arctic
Silver is hiding something. Products sold without numerical
specs should be routinely suspect. Arctic Silver is mostly
sold on hype - engineering specs be damned when your customers
too often fear the numbers.
D
David Maynard
w_tom said:
Ok, I'll do the numbers. My CPU is sitting at 44C in a 28C room with case
ambient at 33C. That's 11C above case ambient and if the contribution of thermal
compound to that was 9C then it would represent 80% of the cooling efficiency:
certainly a "serious improvement."
I'm not saying that IS it's contribution in my system but that your casual
dismissal of 9C, without doing ANY analysis yourself, is unjustified. See below.
Simply untrue. A reasonable estimate can be derived from the clock frequency and
Vcore. Not that it makes any difference to the discussion because whether one
'knows' the heat watts has nothing to do with whether the thermal compound is a
significant improvement over a bare heatsink.
"Must be?"
100F is 37.7C. 9C is 23.8% of that. Not that it means any more than what you
said because one would need to know CPU temperature to know "what does" the
heatsink/thermal compound do. See below.
Probably. Open it up and you'll find thermal compound on the heatsink too.
Not so but also irrelevant to whether thermal compound improves heatsink
performance.
There's empirical data for that.
Such as?
If they're significant then they aren't trivial.
The typical metal to metal contact area on commercial component surfaces is 1%,
or so, of the total area. http://www.thermaflo.com/interface.shtml
That thermal compound does even more for a 'poor' heatsink says nothing about
it's contribution to a good one.
Hogwash. Room ambient is considered to be 25C. Maximum operating temp for a
desktop is often spec'd at 50C. 9C represents 36% of your temperature range.
Put another way, if your maximum component temp is 75C (P4 case temp) and you
have to operate in a 50C environment, per your system specification (like say a
Compaq Presario 5200), then 9C is 36% of your 25C thermal budget and even if you
derate that to a 40C operating ambient temp (Compaq rack servers are typically
43C, derated another 8C from room ambient by the rack) it's 26% of your thermal
budget. Heck, it's 20% in even as low as a 30C ambient.
Those are NOT 'trivial' numbers.
D
David Maynard
w_tom said:
Not putting on thermal compound is simpler.
And just what is there to learn about "don't bother with thermal compound?" In
fact, why even mention or provide thermal compound? Just how much 'simpler' can
it get than to not even MENtion it?
Your argument makes no sense.
I suppose they get chuckles from buying, stocking, and providing it for no purpose.
Again you provide absolutely nothing to support your claims however it's quite
obvious that your synopsis is inadequate, if not entirely incorrect, as "higher
heat generating" means absolutely nothing without knowing all the other
parameters, such as area of the thermal interface, what cooling method being
used, etc.
We're not talking about some 'random' third party advice to a user buying an
'unknown' heatsink. We're talking about the heatsink manufacturer, who knows
darn good and well what their heatsink is, providing and recommending the use of
thermal compound. And that includes the ones provided by the makers of the
processor as well.
Anecdotal stories don't mean much but ones with no information mean even less.
Like, what processor? What 'factory'? Ever consider it might have been a defect?
For all I know the person you mention might not have even known what 'thermal
compound" was. "Oh, this tape looking thingie?"
I bought a 'reconditioned' (I wondered how you 'recondition' a processor)
Tualatin from Newegg. Looked brand new. Boxed and in it's cute little plastic
carrier and all. Except for one thing: the thermal pad had been scraped off. It
had obviously been mounted once. Ah, now I know what 'reconditioned' means but
you can't draw any 'factory' conclusions from it.
Then why was he looking?
He ran a system test on it at maximum rated ambient? Or it still 'works' in a
nice cool air conditioned room that's 20C lower than the system is supposed to
be able to operate in?
Is that a precision mathematics term?
"Kits" may come with thermal compound but semiconductors are semiconductors.
Well, come to think of it, a CPU might have thermal compound under the lid so,
ok, some 'come with it'.
I've already posted real numbers in the previous message that disproves that
assertion.
I haven't seen you use a number yet.
You haven't used either form of science in your posts yet. All you've done is
'claim' that science is involved.
The problem is those points are misleading.
Yes, you don't want to use so much thermal compound that you prevent what
contact can be made from being made but no matter HOW 'good' you think the
contact is, the fact of the matter is that MOST of the surface is NOT in
contact. The thermal compound fills those gaps and it take little because
they're small gaps if, as you say, things are 'good'.
I have no idea where you come up with the notion that properly applied thermal
compound doesn't get on the "outer half of the CPU."
The problem here is that even your "single digit" improvement IS significant.
(numbers provided in the previous message).
Whether one 'knows' the 'numbers' has nothing to do with the effectiveness of
thermal compound. Most people don't know the first thing about nuclear physics
either but the sun's temperature is not affected because of it, and neither is
thermal compound.
And just how are they going to mate it any 'better' than orienting it properly,
per the instructions, and applying the included spring clip, or whatever
mounting method is provided? The user is not there to 'engineer' a heatsink.
He's there to install it and YOU'RE irresponsibly telling the poor soul to leave
the blasted thermal compound off.
An example of a 'good' heatsinks might be helpful, although you've ignored my
request to provide one, but telling someone to not use the included thermal
compound on the heatsink they've GOT is NOT.
Even if your opinion of Arctic Silver were true it is irrelevant to this
discussion of thermal interfaces in general.
E
Ed Medlin
2) if heatsink is properly machined, then heatsink can be
know what "properly machined" is. Even if a HS were machined to 5X what
current heatsinks are, the best molecular contact would be 5% or so. That
would probably work out to us paying several hundred dollars for a very
average HS. The problem is molecular in nature. There is no way for us to
shave off part of a molecule........ Thermal compounds only help to
transfer heat for that other 95-99% of the HS that isn't 'molecularly' in
contact with the CPU. Very simple indeed. Without the compound, we depend on
air to transfer that heat. Air doesn't do that well in most cases. (actually
in ALL cases if you believe the papers written on the subject) Thermal
compound only provides a better medium than air to transfer heat. Nothing
more, nothing less.
Ed Medlin
I am jumping in here and then just shut up on the subject. I would like to
know what "properly machined" is. Even if a HS were machined to 5X what
current heatsinks are, the best molecular contact would be 5% or so. That
would probably work out to us paying several hundred dollars for a very
average HS. The problem is molecular in nature. There is no way for us to
shave off part of a molecule.......
. Thermal compounds only help totransfer heat for that other 95-99% of the HS that isn't 'molecularly' in
contact with the CPU. Very simple indeed. Without the compound, we depend on
air to transfer that heat. Air doesn't do that well in most cases. (actually
in ALL cases if you believe the papers written on the subject) Thermal
compound only provides a better medium than air to transfer heat. Nothing
more, nothing less.
Ed Medlin
W
w_tom
An exotic polished surface is not necessary for a properly
machined heatsink. And 'flat' is not necessarily good.
Application notes from serious heatsink manufacturers discuss
how heatsink applies or conforms to the semiconductor
surface. Too little pressure in the wrong place and even too
much pressure can distort a heat transfer.
Properly machined is a function of that heatsink design.
However some heatsinks don't bother doing any of this when
customers don't even look for the "degree C per watt" number.
Provided is a test, first without thermal compound; then with,
to learn how superior or inferior that surface really is.
Some will flatten a surface or do lapping in a belief that
it makes a better surface. Not necessarily a superior
solution. How that heatsink contacts CPU is but one of many
factors that can lower a heatsink's overall "degree C per
watt"
number.
machined heatsink. And 'flat' is not necessarily good.
Application notes from serious heatsink manufacturers discuss
how heatsink applies or conforms to the semiconductor
surface. Too little pressure in the wrong place and even too
much pressure can distort a heat transfer.
Properly machined is a function of that heatsink design.
However some heatsinks don't bother doing any of this when
customers don't even look for the "degree C per watt" number.
Provided is a test, first without thermal compound; then with,
to learn how superior or inferior that surface really is.
Some will flatten a surface or do lapping in a belief that
it makes a better surface. Not necessarily a superior
solution. How that heatsink contacts CPU is but one of many
factors that can lower a heatsink's overall "degree C per
watt"
number.
D
David Maynard
w_tom said:
Define "exotic polished surface."
Define "properly machined heatsink."
Which semiconductor? Bare? Plastic case? Metal IHS? They all do this?
What's it got to do with the use of thermal compound?
It can bend the heatsink, crush the semiconductor package, break the PCB, crack
a CPU core, and other nasty things. What's improper heatsink installation got to
do with the use of thermal compound?
No kidding.
Claiming the heatsink engineer didn't do a proper design because the sales
department doesn't publish specifications they figure the buyer cares nothing
about, as you yourself assert, is presumptuous.
For the typical buyer, what makes "0.26 °C/W" 'better' than "Intel P4 478 FMB2,
3.2 GHz and higher?" Which do you think most people are more likely to understand?
Btw, that's for an all copper, thermal controlled, Thermaltake P4 Spark 7 which,
surprise, recommends "Interface Material : Thermal Grease (DOW CORNING T340).
http://www.thermaltake.com/products/spark/spark7.htm
And how is one to interpret the data from this 'test', assuming the processor
doesn't go snap crackle pop during the no thermal compound portion of it?
Is the reason a secret?
I think most people know there's more to it than just contact but that
particular portion of the contribution was the topic.
K
kony
Out of curiosity I just did some testing with a new heatsink, one
fairly highly regarded in the industry as one of the best money can
buy (not necessarily my opinion, IMHO it might be worth 1/2 it's
price, but it is a decent heatsink).
The tested heatsink was a Thermalright SLK-900U, an all-copper 'sink
that mounts via 4 spring-loaded studs, attached to the 4 points about
the socket when used on a socket A motherboard. It came with a
well-machined but unpolished base, with regular circular ridges in it.
The following links to an article with a few pictures of one:
http://www.pcabusers.com/reviews/thermalright/slk900u/p2.html
I did not feel that the factory finish on the base of the 'sink was
adequate so it was lapped thoroughly, producing a near-mirror finish,
not quite perfect because of minor pitting from the earlier stages of
sanding, but overall pretty flat and good enough to use to read 12 pt
text on a monitor at 4 ft away without any observable distortion,
except of course for the text being backwards.
The CPU was an Athlon XP2100, further it was lapped very lightly with
fine polishing compound but more extensively on the corner region
where there is (was) a protrusion from the factory laser-etching. I
wouldn've used a more valuable, higher-heat CPU except for my
skepticism about the results, not wanting to risk damage to a more
expensive CPU. Plus, it was available for testing.
This heatsink-CPU combination achieved the same temps without
compound, with generic white silicone compound supplied with the
'sink, and with Arctic Silver 3. I did first try the 'sink bare,
before ever applying any compounds, then twice with each compound,
alternating with wiping the compound off and running it "bare" again.
In all tests the temp deviated by only 1C, but not consistently in
favor of any one interface. The thermal compound was applied quite
thinly and evenly, yet made no difference.
I was a little surprized at these results, partially because even
though I had taken great pains to get the heatsink base as flat and
smooth as possible, after a certain point it was too near to perfectly
flat to be able to tell whether further lapping was making any
progress at all... at one point I tried lapping it on a sheet of
printer paper and it was scratched up enough that it took significant
effort to get it back to it's state prior to contact with the paper.
Point being, I am satisfied that it's certainly possible to achieve
the same temps without compound, but that it's not worthwhile to do...
it took MUCH effort and time (relatively speaking) to achieve what
could've been done in just a couple of minutes, the time it would take
for a quick and minor lap job plus $0.004 worth of thermal compound.
Dave
D
David Maynard
kony said:
What did you use to bring the CPU to full power during the test? I'd also be
interested to know what the actual temperatures were: CPU, room ambient, and case.
D
Dave
It may be true that two nearly-perfect parts (CPU & 'sink) will have
surface by contact with the CPU core?
Dave
Or the base of the heat sink warps slightly due to uneven heating of its
surface by contact with the CPU core?
Dave
M
~misfit~
Dave said:
That too. Although a good thermal conductor like copper should spread the
heat enough to prevent that (?)
K
kony
The short answer is:
CPUBurn and Prime95
41C
~90C (see below)
not applicable
The longer answer is:
It's temp was reported as 41C on an Asus A7N266-VM, which I believe
registers slightly high but at the moment I can't find exactly what
the differential is. It is the "AA" version, uses the CPU thermal
diode, no longer has an onboard temp sensor in the socket-well. The
CPU defaulted at 1.6V but the board was overvolting to 1.63V still
while under load.
The heatsink did not have the Delta FFB0912EHE (92x92x38 & quite loud)
on it. Instead it was fitted with an 80mm Panaflo M1A, spec'd at 2450
RPM @ 12V, 32 CFM.
There was no case temp relevent because the motherboard was sitting
out on a desk, with almost no ambient airflow except for a ceiling fan
on low, a very light, almost unnoticeable breeze from ~15' away.
There was a lit swing-arm lamp about 12-14" directly above it and it
sat next to the wall, half of the exhaust out of the 'sink was about
3" away, towards the wall... hardly optimal. The board (system) temp
reported by the bios was 40C. I am under the impression that this
temp is still taken in the southbridge on an nForce board, partially
because earlier testing showed no difference in system temp when I
unplugged a fan I had installed on the Northbridge. The southbridge
also had a heatsink but tiny, passive. I was forced to remove the fan
from the Northbridge as my replacement NB 'sink fan interfered with
mounting the SLK-900U. It is too big to be compatible with many
boards, but I happend to have the A7N-266-VM out and it fit... then
again I don't know if it would even fit that board with the original
passive 'sink on it. I know it won't fit my A7N8X, or at least I'd
have to pull the board and mod it, hardly worthwhile.
If I had to guess what kind of ambient temp that resulted in, the
guess would be around 90C, but possibly effectively higher... I can't
say for certain since it was a unique environment, with no other
source of significant airflow except the heatsink fan and being right
next to a wall, it's inevitable that a far larger (than typical)
amount of airflow was recirculated though the heatsink, affecting the
results. The power supply was pretty far away, as far as the harness
would reach which is probably about 18" and with quiet fan, so it also
had minimal effect on airflow.
It was a pretty casual test rather than recording every possible
variable, but at least the variables should have minimal impact on the
results, almost entirely limited to the heatsink interface. The
testing was all done during a relatively short period of time, about
30 minutes per run and just long enough to change the interface
between runs, and it's doubtful that the room temp changed otherwise.
Prior to the test the system had already been running for several
hours with a different heatsink on it. The downside to this is that
neither thermal compound had a chance to "settle-in", so perhaps if
either had been used for a period of days then the temp might lower a
bit more, but I wouldn't expect much of a drop, maybe a couple C, and
I can't keep the thing running there indefinitely, or even a few days.
I'm also uncertain of the strength of the heatsink mounting springs on
the studs, if vertical board mounting might decrease the pressure on
the upper portion of the CPU enough that heatsink compound might be
beneficial again.
Dave
K
kony
LOL
Yeah, it's HOT here.
Make that 32C
Dave
W
w_tom
Similar test was performed with a known standard heat
source. Authour provides numbers with his results for
heatsink bare verses heatsink with thermal compounds:
http://www.dansdata.com/goop.htm
His copper heatsink also had concentric ridges which were
not removed. And yet thermal compound resulted in only single
digit temperature improvements. BTW, he discovered that
toothpaste was thermally more conductive than Arctic Silver.
He also tested with Vegamite.
Also would have been interesting to see temperatures for
heatsink applied bare before and after lapping its surface.
Again, more information about how a bottleneck of heatsink
operation is improved by changing its surface.
source. Authour provides numbers with his results for
heatsink bare verses heatsink with thermal compounds:
http://www.dansdata.com/goop.htm
His copper heatsink also had concentric ridges which were
not removed. And yet thermal compound resulted in only single
digit temperature improvements. BTW, he discovered that
toothpaste was thermally more conductive than Arctic Silver.
He also tested with Vegamite.
Also would have been interesting to see temperatures for
heatsink applied bare before and after lapping its surface.
Again, more information about how a bottleneck of heatsink
operation is improved by changing its surface.
D
David Maynard
w_tom said:
You misrepresent the data. The bare heatsink came in at 0.66°C/W and with
thermal compound down to 0.48°C/W but the article provided no watt numbers for
his test setup from which to derive any actual temperatures.
If we plug a 70 watt CPU into those numbers the temperature improvement would be
12.6C and not your claim of "only single digit temperature improvements." One
has to get below 55.5 watts to be under a 10C improvement. A 35 watt CPU would
get a 6.3C improvement but, as I've shown by actual thermal budget calculations,
"single digit temperature improvements" ARE significant. And I reiterate one of
them here: My current CPU is running at 44C in a 33C case ambient. That's a
total 11C rise and even 6.3 would represent more than HALF (57.2%) the
heatsink's job. Again, I'm not saying those are the numbers in my setup but that
something simply being a "single digit number" does not make it insignificant.
His data and conclusions had nothing to do with claiming thermal compound was
unnecessary. To the contrary, he opened with an explanation of why it is. His
conclusion was that one would not see a dramatic improvement when going from
'plain ole' thermal compound to expensive, supposedly 'hi tech', compounds as
the difference between THEM was not 'large'. With that I agree.
Air is some 3,000 times a worse thermal conductor than aluminum. Now, thermal
compound is no where near as good as aluminum either but it's a heck of a lot
better than air so the improvement over an air gap is large. However, once that
initial 'barrier' has been filled, the variation in thermal resistance from one
compound to the next isn't much in the overall picture as the lion's share of
the improvement has already been accomplished by filling the gap with almost
anything having a half way decent thermal conductance.
We are all aware that the surface characteristics affect thermal performance.
As a side note, his 'mission' to explain how statistics can lie is well and good
enough in that one can use statistics to lie but it would be better stated as
using a PROPER scale rather than his apparent affinity for '0' as the universal
and 'true' meaningful point of origin. For example, if we used Kelvin as the
temperature scale, starting at absolute zero of course, then the entire normal
operating range of a typical PC, e.g. 5C to ~40C, looks 'insignificant', using
his term of "10%" (or less), but then that's the range we typically work in.
Similarly, when looking at rise above ambient, ambient is the more appropriate
baseline, not '0', as one cannot cool below ambient (without active devices,
that is).
His '10%' rule of thumb suffers from the same lack of an 'appropriateness'
criteria. If one were to take it as a universal rule then the ATX power spec is
'insignificant' as it requires 3.3 and 5 volts to be held within 5% but clearly,
those voltages being off an 'insignificant' 9% is not 'insignificant'; which,
btw, mirrors my complaint with your constant insistence that "single digit
numbers" somehow don't ever 'count' for anything. By your reckoning, an ATX
PSU's 5v and 3.3 volt rails don't count, and neither does the processor's Vcore,
as they're all no more than a 'single digit' above the decimal point. They might
as well all be 3 volts since that's not more than an 'insignificant' "single
digit number," 2, away from any of them, right?
Buy a dozen eggs that are off by a measly, 'single digit number', 8.3 percent
and you're not going to be too happy about getting 11 eggs instead of 12.
Say the computer manufacturer specs your machine to 40C but it crashes at 31C.
Hey, that 9C difference is only a 'single digit number' so you're happy, right?
Which, speaking of appropriateness and scales, brings us to units of measurement
e.g. 9C is 16.2F. Hey, 16.2 is more that a 'single digit number so 16.2 is
significant, but 9 isn't... but 16.2 is... but 9 isn't... but
Ya know, measuring the distance between stars in inches isn't really practical
but if you're off by .1 parsecs when you get there that last step off the ladder
ain't going to be just 'one small step for a man' whether it's a "single digit
number" or not.
W
w_tom
Dansdata.com does provide a watts number. Unfortunately it
must calculated from his data. I believe it was also provided
in another article about this same test. He was using a
constant 50 watt heat source which is typically equivalent to
a 65 or 75 watt CPU.
From his data, thermal compounds resulted in a 9 degree
reduction of CPU temperature. As kony has demonstrated, this
would probably be less if the heatsink was smooth; not
concentric rings. And there was virtually no difference
between Arctic Silver and all other thermal compounds. But
demonstrated is that thermal compound only results in, at
best, single digit temperature reductions.
Even with a worse case heatsink surface of concentric rings
(not smooth), the thermal compound only resulted in single
digit temperature decrease. Thermal compound would only be
less effective had that heatsink surface been smooth.
must calculated from his data. I believe it was also provided
in another article about this same test. He was using a
constant 50 watt heat source which is typically equivalent to
a 65 or 75 watt CPU.
From his data, thermal compounds resulted in a 9 degree
reduction of CPU temperature. As kony has demonstrated, this
would probably be less if the heatsink was smooth; not
concentric rings. And there was virtually no difference
between Arctic Silver and all other thermal compounds. But
demonstrated is that thermal compound only results in, at
best, single digit temperature reductions.
Even with a worse case heatsink surface of concentric rings
(not smooth), the thermal compound only resulted in single
digit temperature decrease. Thermal compound would only be
less effective had that heatsink surface been smooth.