Anandtech Socket A heatsink review

[OneEng]

thermite88,
Chill.

MikeWarrior2,
I agree that the relative difference between coolers my not be indicative of the actual difference in performance. That having been said, it would appear that the HedgeHog with the Delta fan would be the best cooler. Was that your conclusion as well?

Given the huge difference in fan performance, it would appear that the Alpha has the best thermal design; however, for an end user, I wouldn't think anyone would care why it cooled better, only that it did. A better fan makes a better package.

Given the huge fiasco around the web over the Orb products and Socket 7, I personally would not purchase one. Most reviews I have seen show it performing far below both GlobalWin and Alpha. The HedgeHog is new to me, and seems like a strong performer.

Were these pretty much the conclusions you came to Mike?

 

[Mikewarrior2]

Yes, the Hedgehog is probably the best heatsink. But I will say this.

If you're not familiar with how the hedgehog is made, it is a copper base, holes are drilled into it, and the copper pins are "soldered in". There have been numerous reports of alarming gaps in these "solder" spots, poor heatsink finish, etc. And most reviews with units purchased at stores have it only performing at alpha level.

My point is, this heatsink was most likely "hand-picked" for anandtech/heatsink guide to test it. A handpicked one can have perfectly soldered pins, perfectly lapped bottom, etc. Is this indicative of the one you can buy, probably not.

Also, you're an experienced computer user. I was not saying that the Anandtech review would confuse you. But imagine a new computer user, not knowing anything about heatsinks? WOuld he/she not have been confused by this review? I mean, that $16 quiet c-orb performs pretty well "compared" to a $30+ alpha...


Mike

 

[Laughingman]

Alot of hot air on this thread.

I have measured the C/W almost every heat sink in this review on a 1 gigahertz Athalon on multiple boards. First I drill a hole in the heat sink over the core (see Intel & AMD techdocs they tell you exactly how to do it ) set a T-Type thermocouple flush with base with expoxy, lapp flush with the base; set-up takes all of about 10 min. The manufactuers do not lie. C/W is C/W. The better sinks are cool to the touch; lesser sinks like the "Amazing" Agilent CA are hot to the touch @ .48C. From experience, Tillman's numbers do not reflect even on a hierarchical basis the real numbers.

A suggestion: Tillman Should Repeat the Tests using the Intel or AMD spec. With a Thermocouple thermometer - A reasonably good one from Omega (MOD.HHM23) can be had for aprox. $200.00 US. Alot of multimeters these days from Fluke, Sperry or Extech include a thermocouple for less than $100.00 US. That way a reader would know exactly what the relationship is between actual temp and these hardware monitor programs and select a heat sink that suits his needs.

These off the cuff reviews were OK with CPUs' @ 30W. Now that we are begining to see numbers approaching 60 -75 watts 80% of these heat sinks are not going to make the grade. A few fried processors from your readers based on your recomondation may decide for you how you test.

Kapton or polymide heaters can be purchased for about $25.00 US for a bench test. The polymide can meet the small contact area requirement if desired.

You can get a good ballpark C/W with a TEC by metering the powersupply and daming the the coldside with a closed cell foam. you do have to watch your ambients as most TECs' are on a parabola curve.

For An excellent Analysis on the accuracy of Heat sink review articals check out Burning - issues, Of 20 sites reviewed the bulk of are off 50-100%!

Burning Issues

 

[Nevin]

Amen!
Laughingman,
Amen!

Nevin

 

[Mikewarrior2]

Thanks Laughingman for posting that.

Like I have said before, i am certainly no expert, but I certainly had issues with this review based on my limited knowledge.


Mike

 

[eyor]

One thing laughingman: you said that a good heatsink would be cold, but a bad one would be hot. Wouldn't this be false, to a point, as a "good" heatsink, which conducts very well, would take more heat from the processor, while a bad one would not conduct as much heat. I can see how you might mean that the heatsink would disperse the heat and have the fan suck it off, though. Maybe you should post a heatsink showdown, with more accurate measures. I would be very interested in reading it!

 

[Laughingman]

Your Heat sink is on a gradient- If the source heat is lower;your sink to air is lower.

Showdown? - Just Keep your eyes peeled over the next couple of weeks

 

[Mikewarrior2]

Laughingman,

Also, I don't think you'll find any luck getting Tillman to change his review/comparison. I have seen/heard nothing from him, neither on the bbs or in e-mails that says anything other than "the socket thermistor is accurate enough for comparisons, period" and "repeatability is what matters".

Also, if there are problems wiht my arguments, would you let me know either on the bbs or by e-mail? I'm still trying to learn all of this stuff.


Mike

 

[Laughingman]

Mike

Having never tested using any other method how would he know the difference? I believe that most of the arguments on this thread raise enough valid arguments for alternate test comparisons to be made. After all Ananund + tillman do have a responsibility to their readers if they present themselves as a tech info/review site to provide accurate data. Also if they want to grow to become the next PC mag of the internet a continual refinement of test methodology is in order. After all How many millions do the PC mags' have invested in test and evaluation?

 

[Mikewarrior2]

Laughingman,

He wouldn't know. He assumes that an under-socket thermistor is just as good as a thermistor touching the edge of the cpu, and assumes that since a=b, they are good enough to run a comparison without trying to measure core temp.

I would want something in this review changed. I don't see the point of adding something to the next review after this review is proven to be very poorly done. My comment to you was one that, I personally do not believe Tillman sees the need to correct this review, despite the overwhelming evidence against him in this matter.


Mike

 

[Laughingman]

Mike you are probably correct in you assessment. You can lead a horse to water.....

In any event the disscusion has highlighted some of the major issues as to the accuracy and that is a good thing. I guess I am more befuddled by the fact that there is an easyway to get the right numbers and these guys will take it right down to the mat defending indefensible positions. Why not just do it right to begin with? It certainly would waste a lot less of of everybody's time(although I am enjoying the disscussion).

Regards

 

[johncar]

Laughingman, Mike, et al,
Might as well give our credentials...a career acceptance testing steam turbines in central power stations using strict ASME Power Test Code procedures, including calibrated thermocouples and later platinum RTDs
hand made in our Research Lab, as we were years ahead of industry, due to our desire not only to get accurate results for commercial acceptance but also to "prove" the ongoing steam path design improvements which kept our turbines as world's most efficient. In the 1950s time and materials cost our department $1000 each TC, yes, no mistake, one thousand. Wire quality was poor and much time was spent finding 100 ft or longer lengths with minimal impurities. We used over 100 TC per test and had enough to conduct 2 tests simultaneously, plus spares. Each TC was recalibrated, at 3 immersions, after each use and its history studied very carefully for application in the process where sensitivity analysis dictated. So we know a little bit about thermal metrology. Back on subject....

We've had the TC installations that Laughingman wrote about and that Intel and AMD datasheets recommend for about 3 years...TC beads epoxied and polished flush with hs surfaces at case top center...made ref to same in earlier post. Use type E chromel constantan as has the highest emf/degC of common TCs...read with Fluke 8840A DMM with DC accuracy around .03% and resolution of 1 uv. Transfer function of type E is about 60uv/degC in range of application. Get TCs from Omega, .005" wire with teflon insulation lets us drill only .031" holes thru hs bases for minimum effect on the cooling process. Flexible wire TCs also makes it easy to install/epoxy/bend/feed them out thru hs pins. Won't detail how we handle the temp of cold junctions, CJs, but it's quite accurate, involving temp measurement of room air flowing past uninsulated bare wire CJs and appropriate compensation.

We measure typically 8 to 12C temp drop from hs surface TC to air entering hs fan which is metered room air due to ducting from inlet case fan to hs...no mixing or recirculation...(still using AT case SS7 system with current k6-3+/450 oc to 600/2.2V). So if we knew the power dissipation and the hs surface temp profile we could fairly accurately calculate the hs/fan's C/W. We've made some estimates which put us in the published C/W range of Alpha SS7 ?6035? hs.

All of which is by way of saying that we agree with Laughingman that this procedure would be much better than the cavity thermistor. But the reality of the situation is that we're fighting the apparent convenience of the installed thermistor. And do we really expect these people to start installing and using TCs??

But finally, remember the limitations of the hs surface TC method, surface temp profile may make comparison to mfrs tests invalid, depends on their procedures...and not easy to measure power dissipation. Though even "relative" C/W comparisons would be infinitely better than socket thermistors.

Post is long enough...thanks for listening.
John C.

 

[Laughingman]

Burning - Issues
Just Posted some interesting info - They are incorporating several different techniques in tandem to provide a method for calibrating these hardware monitor systems so they will be more representative of the actual temp. check it out.

Using thermocouples for measurement is not all that complicated. For those among us who test this way frequently I think there is a tendancy to over parse the details becuase that is required of us in our work. But in this situation with the review sites I believe most of the readers would be happy to to have a measurement within a degree or two of the absolute value. Ten or fifteen degrees off the absolute between sinks is where the problem arises it is valueless for any comparison whatsoever.

Bench test. com shows shows how to hack a radio shack thermometer for thermal testing. I think most of the review sites could do better than that, but it is probably good enough for most review purposes if money was the issue. Being 90% accurate is better than 100% error.

I am going to bed

regards

 

[johncar]

Agreed, but still not optimistic that most reviewers will give up the convenient, cheap, but dirty cavity thermistor.

Hope you got a good rest.;-)
John C.

 

[Mikewarrior2]

JohnCar,

Unfortunately, I agree that most, if not all reviewers will not abandon the easy(inaccurate) way out.


Mike

 

[Mikewarrior2]

Tillman,

Since you've not replied, so therefore i assume you do not care about your $hit review, i am fully prepared to go to anand with this thread.

Also, i have found yet another farce about your "review/comparison". You claim that using thermal grease on the thermistor in the socket increases accuracy. However, looking at the hardocp SuperOrb review, and your own results, both tests get 16C over ambient case temp for the reported CPU temp. If anything, his cpu runs a bit warmer in comparison to ambient by about .3C.

So grease doesn't help... hmm, i wonder why? What more proof do you need that this backside is a very unreliable secondary heat source! If adding grease to facilitate PCB to thermistor contact does nothing, then the heat source the thermistor is reading is clearly unreliable. And if these "algorithms" that you claim which the bios uses to interprate this data existed, then even a bit of heatsink grease should increase this "reported" temp significantly.


Mike

 

[VladTrishkin]

Mikewarrior2 Calm down, no need for flaming now. I'll see if I can see what?s behind Mr. Tillmann's theory, I?ll let you know.

I also wanted to mention that a lot of good points and examples have been mentioned here, and I agree with some of them. Most of my background comes from the Cryogenics field of work, but I will have to see what?s the best testing method in our case.

 

[Mikewarrior2]

VladTrishkin,

I think I have been calm. I am still calm. But this review is a disgrace to typical good quality of Anandtech reviews.

He has simply stopped caring. There is more than overwhelming evidence in this thread, especially from people who are knowledgeable in the field(ie JohnCar, Laughingman), that this comparison is done incorrectly.

And when he did respond, all his did was "claim" that the thermistor solution worked, and that it was close enough to external probe to not matter. Well, since his temps are in line with HardOCP temps(case to cpu reading wise), and External probes, properly mounted, certainly report higher temps, one can easily see how the under-socket thermistor is wrong.

And he never managed to discuss the problems of measuring from a secondary heat source.

And he doesn't mention that the c-orb is only rated at 1ghz. As one can easily see from the review and this thread, even though the reported temp is fairly low, the actual core temp hits very near 60C(hmm.. maybe this is why it is only rated to 1ghz). How many new Anandtech views may damage their cpu's thinking that, because of this review, the c-orb is capable of use at 1.1ghz or higher.


Mike

 

[shoarthing]

MikeWarrior - Hi - I'm afraid you're missing the point re:
backside sensors. The signal from these is handled [like all
mobo temp-sensors] through lookup tables supplied by the IC/chipset
manufacturers & more-or-less tweaked for the specific mobo.

These tables - especially for VIA-chipset boards - are utter rubbish.
I have a dual MSI board with two sensors-onna-wire things similar to
the review board.

I can also simultaneously monitor true temps from each CPU-base.

The mobo reports [if set accurate at HLT idle] 6C low 19C later
- when the CPU is under 100% load. This non-linearity is unaffected
by either baring the tip of these sensors &/or applying h/s grease.

The problem is non-linearity of the look-up tables - as Tillmann
attempted to explain.

I have discussed this with two makers of mobo monitoring utilities -
they just have to accept this pre-digested input. There is no realistic
way of providing correction-tables for each mobo/CPU combo inside
a monitoring utility.

Tillmann is perfectly logical [& IMHO right] to believe that case-base
readings are more relevant to all of us - this is how real-world readings
are taken on real mobos.

Where he & other testers have been remiss is in not checking to
see if there's a simple way of getting reliable base-case readings -
a way that doesn't rely upon the mobo sensors & their unreliable
lookup tables.

We [burningissues.net] have bothered to do this & have found - to
our surprise - such analogous but accurate readings permit C/W calculations
to similar levels of accuracy as the Intel-approved technique at
negligible cost.

Your criticism of Tillmann's review is fair on these grounds:
1) the results bear little relation to true temperature - mainly
because this non-linearity bunches the readings.
2) bunched readings render delicate better/worse comparison
meaningless
3) the high wattages of current & upcoming CPU's demand a
decent standard of accuracy before there's a rash of fried
fingers & components.

Constructive debate - which your outraged criticism has inspired
here - is great: the enemy is not Tillmann; but his belonging in &
with the public acceptance of the shabby nonsense published as "tests"
by h/w sites on the web, where a couple of uncalibrated readings from
MBM are presented as evidence.

IMHO the way forward is accurate C/W s-a measures - however taken;
but there are serious practical & theoretical issues with the Intel
measurement technique: there's a piece at burningissues for us few
interested in this sad stuff.

Changing this public acceptance of nonsense may be done by offering
greater accuracy to anyone using MBM [or similar]: this means aiding
user-calibration of max CPU temp under full load - how else can this
be done except by publishing accurate C/W's together with CPU wattages
& h/s grease K's converted to C/W j for common contact areas?

- & let's not forget there's commercial interest here: while the web
allows the good, mediocre, & bad to be "tested" without fact
emerging, those who make crap will not be found out.

The 'clockers' market is a nice little earner for a lot of folk: many
of whom supply favours & more to various h/w sites. If all "tests" are
rubbish, those possibly influenced by favours present or future will
remain hidden by the general shambles.

Let's not jump to be over-righteous here; but how many sites would
you say you actually trusted? Put it another way: if, like me, you'd
just analysed 20 GOrb tests & found the average error flattered GOrb
performance by 57% & that the majority were freebies - what would
you think then?

 

[Mikewarrior2]

Thanks for posting about this Shoarthing!

The reason my outrage is directed at tillman is his unresponsiveness to respond in the forum or to e-mails. IMO, if he isn't going to try to solve this problem, then his is most definately a part of it.

And in this case, it drags Anandtech into part of the problem. Anandtech is/was one of the sites i actually trusted on the internet for their tests. I have never spoken in outrage over HardOCP tests, because I never really trusted their tests, and nor did I feel they had the amount of user base that Anandtech has.

And the point you bring out about the web allowing for bad reviews without any fact behind them is a huge problem, and is a problem with this review. How many reviews on the internet have shown that different heatsinks don't change the temp whatsoever, or that different(and better) heatsink greases exhibit ZERO improvement? I think a majority of this is due to the majority of people/reviewers taking the mb sensors readings as being perfectly acceptable for accuracy/comparison purposes.

My latest problem with tillman has been is lack of response to the posts here, by people much more knowlegdeable than myself. My other problem is when he claims that heatsink grease helps the thermistor get better readings, when in all likelyhood, the difference between a thermsitor touching the backside and a thermistor touching the backside w/grease is either non-existant or not reported by mb. I honestly don't believe he tested the thermistor+grease solution compared to an external thermistor like he claims he did.

Again, Thanks for posting here.
Mike

 

[shoarthing]

Mikewarrior - Hi - The logic in putting h/s grease on the
sprung sensor escapes me: I did it simply to see what would happen
& whether it would change longterm or under thermal stress -
but then I could actually monitor the effects [negligible BTW].

Looks to me like an unthought-out belt-'n-braces move; & in
conflict with this being "real-world" oriented or with any
prior understanding that the primary source of error would
be the look-up tables. A moderate theoretical argument for such use
would be in averaging temps from a larger-than-point area.

H/S grease testing is a separate problem: as far as I can make
out, by far the best is Circuitworks 7100 [by manufacturers
figures for K, which our tests support]. You need refined
measurement & - as Johncar pointed out - a system for even
application [& control over surface finish/flatness].

Tillmann appears to have posted several times in this thread
with pretty fair humour: his reputation is under review
here & that's tough to take. A fair few heavies have come
out from behind the curtains & I agree the debate should
continue - but perhaps leaving behind direct criticism of this
review & most certainly without any suggestion of impropriety
on Tillmann's part.

I don't normally post at this sort of place & only did so
because laughingman piled in: I'd mailed him about this test
then thread - with, it must be admitted, a . . . .

[sigh]

- but perhaps it's time to get it on & change things.

 

[Mikewarrior2]

I would personally be more than willing to stop targetting tillman for wrongdoing and attacking his review if he would respond, and or be willing to correct the mistakes of this review, or even be willing to learn from his possible mistakes. He has exhibited none of that.

The fact that he has not posted since last Wednesday is just one indication that he probably no longer cares. And I feel that since he wrote the article, he has a duty to the new user to correct it, or at least ammend it with the fact that the numbers simply are not accurate.


Mike

 

[frank828]

go mike! go!

haha, thanks for bringing this up, i completely agree with you that the average/beginner computer user would probably take everything written in Anandtech and other sites to be true, and they need to be responsible for accuracy.

sidenote: mikewarrior is probably one of the most knowledgeable ppl i have ever met on these forums and have nothing but respect for him...so listen to him!!
haha


frank

 

[Mikewarrior2]

frank828,

Thanks for the support, but I'm actually very much so not knowlegdgeable in this field. What i know is what i've picked up from JohnCar and Nevin and others. Certainly in this thread, JohnCar, Nevin, Laughingman and Shoarthing are much more knowledgeable than I am.


Mike

 

[yo2tup]

it says in that test that the taisol CGK and alpha perform exactly the same. i just replaced my alpha with a CGK, and i didn't get the same results that the article got. check out my results at the cases and cooling forum.