Swiftech Apoggee review by Robotech systemcooling

[GlassMan]

Some possible fuel
All side measurements are appxs from photo
Amd 90nm 1meg cache 8.5x13.5mm=114mm*mm
Amd x2 2meg cache 17mmx13.5= 199mm*mm diagonal appx 21.7mm

AMD San Diego, 3700+
Contact patch diameter of TT cam lock (silent boost) on heat spreader 24mm
The heat sink was lapped flat (I hope) on a 3/8" glass plate, down to 2000 grit (w/water) and contact patch appeared to be perfect. Around circle was excess TIM (Thermal right compound) which did not make good contact.
Judging by result of new Storm based wc kit the heat spreader nor it's TIM were damaged by a the HS's hand lapped surface.

[Brians256]

Originally Posted by BillA

Scott
please stop whipping yourself with test data, for now drop it unles you are seeking the best testing wb
look only at the CPU results (thoughts about core and edges fly out the window)

Lee has data as does Swiftech and there are net 'reports' springing up -> but it is NOT the dog described by Lee's bench data ??
to me this a problem in logic, which is moot if CPU results only are considered
test bench winners eh ?
kinda like the dyno racers

Bill, can I restate what you said and make sure that I'm understanding this correctly? Let me know if this is correct.

1. Different DieSims show differing results for Storm vs. Apogee etc...
2. DieSims are hard to correlate to actual CPU die temps and the resulting improved overclockability.
3. Lee's tests with die temps of an A64 show that the Apogee, Storm and MCW6002 perform similarly (within level of any user's ability to detect) even when using high voltage to show deltas that might be compressed on low-end systems.
4. DieSims (because of the difficulty in correlating their data to on-CPU-die temps) are less and less useful.
5. We are getting all bent out of shape over what appears to be a meaningless difference in performance (as seen on Lee's test).
6. Anecdotal evidence is interesting but useless precisely because we see unsubstantiated claims on both sides of the performance issue.
7. Bill is NOT trying to defend the poor QC which has led to the reports of metal shavings in the blocks.
8. Bill is not a liar because he accurately reported test data taken on his test system (and his Swiftech replacement isn't either).

My personal conclusion from Lee's data is that Storm is a better performing block than Apogee but (when QC issues are resolved) may not perform significantly different than Storm on current blocks. Future larger blocks may narrow and even reverse the small difference in perfomance if die sizes and power outputs greatly increase. But, this is not guaranteed. Apogee is certainly cheaper.

Here's my obligatory NiceGuy Moderator statement:
Please guys, stop telling each other that we're all full of fecal matter. As long as we are all open and honest, I sincerely believe that we can politely have differences of opinion because of different assumptions of what is important.

By all means, poke holes in each other's assumptions, but do it politely and remember that repetition seldom makes better penetration into someone else's thick skull. I know you aren't going to start a hug fest, but at least stop pouring napalm on the campfire. It's already burning JUST FINE.

[GlassMan]

A possible defect in Robotech's analysis of the cpu testing?
In know way to demean his hard work*, but the Apogee is cooling to the edge of the heat spreader, while the Storm consentrates it's work in the center. Wouldn't the equal temps at the Heat Spreader edge imply that the Storm is cooling over the cpu better?
* Yes, resolution problems, and he was using his best data for his analysis.

[gone_fishin]

Sure the storm does cartwheels on the die sims but people don't buy waterblocks to cool dies sims all day do they? The measurable number that counts is the overclock. If they provide the same overclock to joe sixpack then why should the manufacturer choose to make the more expensive product?
It boils down to simple business decisions.

Problem is they have to convince joe sixpack overclocker that they do not need a gold medal olympic thermal die performer to get that same overclock.

[BillA]

backwards, I too think the Storm is better - hence my objection to replacing it in the Apex kit
but from a price/maint perspective, if the performance is nominally equal then for most users it is probably a better choice

1 ok
2 no; I don't think actual die temps are known, just a 'measurement' named such
while much discussed over the years, the measure of overclockability is more than any single temp, thermal capacitance is an obvious contributor which we ignore in our testing as we have no dynamic evaluation capability
- this is quite doable, but if my primitive setup is considered high-end then the idea of thermal transient analysis is laughable
3-8 ok

do you have any silicon temp limits you can talk about ? (I do not)
I think most here are applying case thermal design temps to silicon; the whole die temp thing might be of interest to designers, cannot see its need for OCers
(but I'm not an OCer either)

[ricecrispi]

Swiftech made it sound like the apoggee was a clear cut one temp degree improvement over the Storm on TTV and RL tesbed data to back it up.
I would love to have robotech use an apex kit like swiftech did on a cpu testbed to compare results. We should throw in mcw55 and that would add more heat into the loop.
=====
According to Swiftech

"Here it is apparent that the Apogee has the lowest thermal resistance at all flow rates and more importantly features a wider performance advantage at "real life" flow rates between .3 and 1.5 GPM."

"As in the preceding graph, the Apogee outperforms all previous solutions whether a small aquarium pump or a high pressure industrial pump are used."
===
I know I wont trust Swiftech test results and their statesment from now on. Fool me once, shame on you. Fool me twice, shame on me.

If the data was still good, I wouldn't buy the block anyways. Who cares about the performance of the block if it breaks apart, leaks, and has shavings that might damage a pump. Have we forgetten to talk about reliability? If a car in MaxSaleen's example broken down each run do you think that would matter? The nail in the coffin for the apogge is not performance data, it's the Qc and reliability issues.

[nikhsub1]

Originally Posted by Brians256

Here's my obligatory NiceGuy Moderator statement:
Please guys, stop telling each other that we're all full of fecal matter. As long as we are all open and honest, I sincerely believe that we can politely have differences of opinion because of different assumptions of what is important.

By all means, poke holes in each other's assumptions, but do it politely and remember that repetition seldom makes better penetration into someone else's thick skull. I know you aren't going to start a hug fest, but at least stop pouring napalm on the campfire. It's already burning JUST FINE.

I hope I'm not guilty I am just trying to understand the how's and why's here. Seems a complete 180 has happened it terms of assessing a WB's performance and I can't figure out the reasonings why.

[gone_fishin]

Wow, you guys really want to put that sucker in the grave. It's introduction has caused such angst it has led to full rebellion.

Maybe they should have been honest and pushed it as,

"Here is a cheapass version of a waterblock that will give you the same overclock even though it has shitty thermal die numbers. Now go out and buy it because we refuse to make expensive products with the high overhead anymore for zero detectable real world gains."

[GlassMan]

Wasn't my point, and I agree the proof is in the pudding.
I have no plans to get an Apgee when I get to the limit of the Storm, and most people will feel the same and vice versa. So we compare 2 (insert #) rigs nothing is proved either. If Robotech had done that I would give the result limited credence (since I have the same cpu), but I know if I had the same equipement I could easily have different results. What's the difference in C? When does that amount result in a better overclock. Stew brought up the same point with his G7
Should have gotten a 6000?
I am happy my Storm only had a thumb print inside, and if I screw up sometime putting the tubing on I might break the fitting, but won't break the housing. That's worth the $10.00 buck extra I payed, out of the several hundred I layed out. Thank you.

[Joe]

Originally Posted by nikhsub1

I hope I'm not guilty I am just trying to understand the how's and why's here. Seems a complete 180 has happened it terms of assessing a WB's performance and I can't figure out the reasonings why.

Yeh there seems to be a nice whiplash from this change that has everyones hair up...

[BillA]

Stew backed me into that corner, how can I say that artificial test results are more meaningful than actual performance data ?

testing is/can be a guide, but the the goal is a performance increase
if the test yields data that is found to be useful, good
if not, revise the test or the analysis

until now it seems, the various heat sources produced different results but not in extreme conflict; but now the Apogee results 'don't fit'
such is something for investigation by those with an interest, but there is a very handy arbiter, actual performance
(and if we say that we cannot discern based on performance, then what is going on ?)

[bigben2k]

Originally Posted by BillA

...
until now it seems, the various heat sources produced different results but not in extreme conflict; but now the Apogee results 'don't fit'
such is something for investigation by those with an interest, but there is a very handy arbiter, actual performance
(and if we say that we cannot discern based on performance, then what is going on ?)

Am still puzzled. Ref post#19.

The largest difference noted was > 3 deg C, under 14 by 14 @ 80W (simulator)

Why would there be no discernable difference under (roughly) the same contact area, at 100W? 125W? (real CPU)

[Cathar]

I am not going to add additional shit to the fight, but will only repeat user observations that have been reported to me over the years, and not specifically relating to the Storm at all. Correlate them at your will/peril.

In all non-IHS scenarios there has consistently appeared to be a very strong correlation with artificial die performance, and the ability to overclock and/or sustain lower reported CPU temps. A trawl through the various OC forums over the years shows that this is a strong general trend. Not a given fact.

Users of IHS capped CPU's have fairly consistently reported a "numbness" effect with respect to reported CPU temperatures. There is always very little to separate different blocks from each other, if anything at all, unless there is a truly vast difference in measured artificial testbed performance.

Users of IHS capped CPU's who have subsequently removed their IHS's have trended to observe cooling performance and overclocking performance consistent with projected artificial test-bed performance. i.e. lower reported temps and better overclocks. As always, temperature reporting differences are numbed/dulled by the practises of temperature "fiddling" and compression that occurs by mobo makers in the industry.

Those few users who have persisted with assessing waterblock performance between both different waterblocks and different IHS capped CPU's, and as such perhaps provide a clearer independent picture of the extent of the muddiness of the IHS->wb relationship, have found that reported waterblock performance is wholly dependent on the individual CPU+IHS being used. On one individual CPU, broad cooling patterned wb A can be found to be better than focused cooling patterned wb B. Switch out the CPU for another of the exact type, and the observable wb performances will trade places, as will achievable overclock.

In all of this, it is quite clear to me that no-one is lying, and no one is at fault. The only issue is this:

The use of IHS's provides a variant and inconsistent mechanism around which cooling device performance may be assessed. Where any individual gets "tripped up" in all of this is choosing to believe otherwise, or worse, choosing to base expensive business decisions on the results provided by a variant and inconsistent testing mechanism.

Are bare-die-sims better than "real world IHS die sims". Yes, and no. They are because they are consistent. No, because they do not reflect the real world directly. The problem here though is much like chaos theory. The introduction of the IHS provides for a large set of variables that absolutely defy consistent measurement. We could sit here all day and construct hypotheticals for why the use of IHS's are providing varying results as they are, and the more we do so, the more we will come to understand how futile it is to attempt to use them categorise cooling design performance.

So where does that leave us? Between a rock and a hard place. Damned if we do (use IHS's for measurement) and damned if we don't (cue cries of "NOT REAL WORLD!").

The real world sucks (IHS use). In a world where IHS's pollute results, the best any of use can do is fall back to the one thing that is providing consistent results, and that is bare-die simulation testbeds. It may not simulate the real world, but I'll be damned if someone can convince me that using a mechanism that introduces a multitude of variables for purposes of assessing wb design performance can ever be described as "a good step forwards".

[Althornin]

Originally Posted by gone_fishin

Wow, you guys really want to put that sucker in the grave. It's introduction has caused such angst it has led to full rebellion.

Maybe they should have been honest and pushed it as,

"Here is a cheapass version of a waterblock that will give you the same overclock even though it has shitty thermal die numbers. Now go out and buy it because we refuse to make expensive products with the high overhead anymore for zero detectable real world gains."

Um, where do you get the data that it provides the same overclocks?
Both of your posts in this thread assume it - where is evidence?

[Albigger]

Originally Posted by bigben2k

Am still puzzled. Ref post#19.

The largest difference noted was > 3 deg C, under 14 by 14 @ 80W (simulator)

Why would there be no discernable difference under (roughly) the same contact area, at 100W? 125W? (real CPU)

Exactly. Either we're now saying the CPU/mobo system test is the real deal (even though everything I've read up till now suggests not to trust those sensors for crap - and a temp probe on the side of the IHS - not sure about its correlation to core/center IHS temps) and then the TTV does a rather good job of simulating the real thing. That leaves us with the heat dies which don't produce similar results.


OR

Still say the temp readings from the mobo/cpu are not to be trusted but take the heat die ones as gospel. Then the TTV/Swiftech results are still in disagreement....


EDIT: would REALLY like to see a heat die with an IHS (either built-in or removable). May possibly help bridge the gap?

[Joe]

Originally Posted by Cathar

So where does that leave us? Between a rock and a hard place. Damned if we do (use IHS's for measurement) and damned if we don't (queue cries of "NOT REAL WORLD!").

The real world sucks (IHS use). In a world where IHS's pollute results, the best any of use can do is fall back to the one thing that is providing consistent results, and that is bare-die simulation testbeds. It may not simulate the real world, but I'll be damned if someone can convince me that using a mechanism that introduces a multitude of variables for purposes of assessing wb design performance can ever be described as "a good step forwards".

Cathar that was a great post.

While I know there will never be a uniform agreement on anything... as you said we are damned either way... I think you touched on many areas I was wondering about.

I mean its simple to figure out that an IHS hurts cooling... there are at least 2 more layers of material to go through for an IHS so it makes very good sense that an IHS will always be a handicap for the best cooling. I couldn't imagine dropping $300.00 on a cooling system and keeping the IHS on the CPU.

From my point of view... and this is just me talking out my butt I guess... But for REAL results in my eyes I would want one that came from a bare die than results obscured by a IHS.

[csimon]

No reason why they couldn't publish both results w/ & w/o the IHS. To me that would lend more credibility to the test.

[bigben2k]

Argh! If only I had my testbench running!

I would do a comparison of a free floating IHS, mounted with old clamping pressures. Then re-run the series without an IHS.

Cathar makes a good point; IHS add a variable that is hard if not impossible to measure/replicate. Here's a bit of info that might help clarify it, for some:

When the core was bare, some people reported using mounting pressures way above specifications, with varying results, but has anyone tried to do the same with an IHS capped processor? I see none. (!)

[BillA]

“Those few users who have persisted with assessing waterblock performance between both different waterblocks and different IHS capped CPU's, and as such perhaps provide a clearer independent picture of the extent of the muddiness of the IHS->wb relationship, have found that reported waterblock performance is wholly dependent on the individual CPU+IHS being used. On one individual CPU, broad cooling patterned wb A can be found to be better than focused cooling patterned wb B. Switch out the CPU for another of the exact type, and the observable wb performances will trade places, as will achievable overclock.”

Stew, the above is the basis for our disagreement. I do not believe that it is accurate. It is beyond my understanding how such IHS performance variability could be seen in products so different, the TIM joints in AMD and Intel CPUs. And we are to accept that this variability applies to all past and future products having an IHS ?

I do not wish to pillory you about someone else’s data, but do not believe it correctly characterizes present products. Here is the problem in assessing the internal TIM joint variability; how does one distinguish between differences in the TIM joint and differences in the CPU ?
You would posit that all CPUs are the same, such that all differences are due to the TIM joint only ?

You have not supported your proposition that all factory TIM joints are different (apart from the absolute sense), but you are stating that any wb could equal any other depending on the CPU it was mounted on.
The random walk theory of wb cooling ?

“The use of IHS's provides a variant and inconsistent mechanism around which cooling device performance may be assessed. Where any individual gets "tripped up" in all of this is choosing to believe otherwise, or worse, choosing to base expensive business decisions on the results provided by a variant and inconsistent testing mechanism.”

DO note that you are saying that IHS users must expect inconsistent cooling performance.
I do not believe this is true.

As an intellectual exercise Stew, let us posit that the Intel internal TIM joint is consistent.
Your whole argument collapses, yes the IHS will compress the range of the apparent temps – but so what ?

You have not supported your assertion that Swiftech used “a variant and inconsistent testing mechanism” as you have not shown that it was inconsistent. In fact the opposite is true, the mechanism provides vastly more consistent results than my old heat die. Nor, using consistent in a different sense, should it be said that the testing mechanism was inconsistent between wbs; similar types of wbs had similar results (MCW55 and Apogee).

For those not aware of the larger issue, it is this:
If Stew is successful in characterizing ALL ”test mechanisms” using an IHS as being “inconsistent”, then no CPUs could be used as a “Stew approved” test mechanism.
I believe that a means can be devised to use a CPU as a so-so heat source and am attempting identify such.
bare die sim vs. CPU

[Cathar]

Whatever Bill.

Didn't come here to throw shit. Seems to be just one person doing it after all.

Later.

[BillA]

Originally Posted by Cathar

Whatever Bill.

Didn't come here to throw shit. Seems to be just one person doing it after all.

Later.

I spent 45 min carefully wording a response (slow typer) and have it dismissed in 3min as shit ?

[Cathar]

You're just repeating the same thing that has been said dozens of times before. Why continue? We fundamentally disagree. There is no point. Period.

At this stage I really could care any less about waterblocks in general. This is far beyond enthusiast material, and well into grubby business-based politics. A game that I have absolutely no desire to play as I have stressed many times before.

For me to continue on posturing around the same point that we fundamentally disagree on will just destroy any last shred of enthusiasm I have left. In case people hadn't noticed, I haven't posted much of anything of late, and really, I couldn't be happier.

To me it's pretty obvious what's going on. I refuse to waste further energy on it.

As I've stated many times before, this is not about money for me. This is about the joy of discovery. I've seen grubby business politics destroy that joy about as well as a bullet to the head. I am truly sorry that I ever broke my promise to myself to remain wholly independent of the greater commercial mire.

[bigben2k]

FWIW, I think it's worth investigating.

[Annirak]

Originally Posted by Cathar

The use of IHS's provides a variant and inconsistent mechanism around which cooling device performance may be assessed.

Originally Posted by BillA

As an intellectual exercise Stew, let us posit that the Intel internal TIM joint is consistent.
Your whole argument collapses, yes the IHS will compress the range of the apparent temps – but so what ?

Well, no shit. His argument was based on the evidence that the internal TIM is NOT consistent. If you take away one of the presumptions of an argument, is it any surprise that it collapses? Thanks for an introduction to logical reasoning that is completely pointless. If you can provide data to show that the Intel TIM is consistant, then you might have an argument.

This is the most ridiculous assertion I've seen you make yet. It's like this completely fake scenario:

Cathar: Suppose A=>B and B=>C then A=>C
BillA: Suppose A does not imply B and B=>C Then A does not imply C!!! Therefore you must be wrong!!

[Albigger]

What really needs to be done to set this issue to rest is (gasp) statistical analysis and loads of testing. Buy 5-10 of the same procs and test them in the same mobo each at least 5 mounts with the same waterblock and see the variation (or lack thereof). Grooving the IHS for a temp probe may be helpful (to see (in)consistency of temp differential to core temp).