Swiftech Apoggee review by Robotech systemcooling

[gone_fishin]

Originally Posted by Annirak

Well, no shit. His argument was based on the evidence that the internal TIM is NOT consistent.

What evidence?

As BillA pointed out, IF any variation is observed in IHS tim from cpu to cpu it is indistingishable from variations in the cpu's themselves. Any observation made would quantify the two together, so how do you seperate them?

Edit to clarify point.

[Albigger]

Originally Posted by gone_fishin

What evidence?

As BillA pointed out, IF any variation is observed in IHS tim from cpu to cpu it is indistingishable from variations in the cpu's themselves. Any observation made would quantify the two together, so how do you seperate them?

Edit to clarify point.

Buy some cpu's, test with IHS on, pop IHS off all cpus and test the bare dies themselves.

Who is going to lay out for that one - probably no one.

[Annirak]

Originally Posted by gone_fishin

What evidence?

Oh, good grief. You're arguing because I didn't post evidence? The whole point is that Cathar stated that he had anecdotal evidence that IHS interfaces were highly variable.

Originally Posted by gone_fishin

As BillA pointed out, IF any variation is observed in IHS tim from cpu to cpu it is indistingishable from variations in the cpu's themselves. Any observation made would quantify the two together, so how do you seperate them?

Edit to clarify point.

Nothing in there has any relationship to me pointing out that BillA's argument is running in a logical circle.
And besides. I was the one demanding evidence. Stop asking me to provide it.

[Marci]

*/me points at the chocolate biscuits and wiggles eyebrows...*

Circles... as in "we are going round in"... and have been for days.
Anyone else been watching Swiftech's shares values since this all began...?

Pass me some skii's - there's a good downhill slope to fly down...

Bedtime - n' me woman's waiting... much more interesting now...

[gone_fishin]

Originally Posted by Albigger

Buy some cpu's, test with IHS on, pop IHS off all cpus and test the bare dies themselves.

Who is going to lay out for that one - probably no one.

That would be one way to get a number. So you get 10 cpus (a very small group 100 would be more of what we would want) you do 5 trials IHS on 5 trials off....that is 50 sets of data for 10 cpu's. Now do the same for AMD....then the same for each type of AMD and Intel cpu on the market today..... that is over 400 test runs and 80 cpus to buy not to mention all the mobos for different socket types. I doubt Cathar or anyone did this so how could anyone (other than the manufacturers themselves) have evidence?

[Cathar]

The results are in disagreement. Something is wrong.

Surely the first point of call would be to presume that there is an unaccounted variable, being introduced by the IHS->CPU die interface, that is been ignored.

Surely, in light of the increasing amount of data that shows that this is highly likely to be the case, that it is then up to prove that the IHS->CPU die interface is invariant, and not the other way around.

The assertion is that the IHS->CPU interface is variant. There is already sufficient data to show that this is highly likely given the rather obvious existing discrepencies. Where, then, is there any shred of evidence to prove the opposite, other than merely taking someone's word for it?

The evidence is mounting to presume that the interface is variant, but we are still arguing that it is invariant with absolutely no evidence? Please. When did logic take a leap out the window?

[Joe]

I for one will believe that Intel TIM interfaces are invariant once someone produces PROOF showing it that way(real test, real numbers) Including a TTV. And from that I assume any TIM interface on any CPU is going to be the source of dynamic readings and shouldnt be trusted for exacting tests.

Real world or no, if you show that every block has the same rating with an IHS but vastly different without it, you will see people ripping IHS's off their CPU's. as I said earlier... why spend money on a cooling system if you arent willing to get the best cooling you can for the money or the purpose. (ie: do something free and remove that IHS)

[RoboTech]

Originally Posted by nikhsub1

I just wish that Lee did the CPU test with a known dog, IE maze 3 or 4, I bet either of those would have performed in that test as well as the storm or the apogee. Then what? All blocks are suddenly equal?

My old Maze3 is on the CPU and testing now...

I contacted Swiftech and expressed my concerns about the shavings and thickness of the top housing. I was not surprised to receive an immediate reply and hear they were already aware of the concerns, had investigated them in-house, and have implemented corrective action where necessary. Here is a direct quote from Gabe:

Copper shavings: Nothing is loose shavings when delivered. Each base plate is inspected in production, and cleaned up. Some might have escaped scrutiny. After all, these pins are really tiny. The lady doing this is now using a big magnifying glass to see well. In any case, you really have to scrape them with a sharp tool to remove them, which is what she is doing. Also note that the bigger pieces (on the outside perimeter) that you showed on your site are actually left over pins from the machining not “shavings” or burrs. These are normally solidly attached to the base. Finally, we have also taken steps in production to completely eliminate any shavings using a new process.

Gabe is correct about the few tiny shavings I found. Too small to see until I hunted them down with a magnifying glass, and they did have to be pried out with a fine dental tool.

I inspected the “thin” areas as you mentioned, and they completely meet our specs. Pressure tests at 40 PSI resulted in no problems on 100’s of blocks (the pressure we test at). We also did some destructive testing this morning on 20 of these housings, using a heavy hammer and hitting the top of the housing (inlet & outlet), which places a direct stress on the alleged “weak” spots, and NONE failed. The entire housing will eventually disintegrate when hit hard enough, but we found absolutely no signs of stress related failures.

I read in Pro Cooling that some people are concerned about pushing the tube in and breaking the housing. Well, we “pushed the tube” on 20 blocks with repeated blows of a hammer… When you “push” hard enough, the housing will explode in pieces. I will therefore advise users who push the tubing that hard to wear protective glasses… Joke aside; I think your concerns with respect to the thickness of the wall are unfounded. The fact that ONE user reported a problem, problem that we cannot even verify since he epoxied the water-block does not mean that there is a problem with the housing. It means that either we did a poor job at QC, or that something else happened with this water-block that the user is not reporting.

I also decided to abuse the Apogee top cover I had sectioned, and found it to be very tough. Pushing, pulling, prying, etc. – in general applying a lot more force to the hose barb than I ever would under normal circumstances, without any signs of cracking or failure. I even gave it a few good whacks with a hammer and nothing.

[Cathar]

What I don't understand is this. We have literally dozens of engineering review articles detailing the variability of TIM interfaces with respect to pressure and flatness. We have an incredible amount of evidence to show that these interfaces vary for whatever reason, pressure and flatness being the two main culprits.

We have a testbed that appears to be producing odd results. This testbed introduces a second interface via a TIM + IHS. We have these odd results forming the basis of a marketing campaign for a new waterblock, strangely enough with a flexible base-plate (pressure/flatness above anyone?), and somewhat unsurprisingly for that which independent tests can't verify the claimed results.

We then have a disgruntled ex-employee who is still owed a lot of money by said company which has taken a rather large financial gamble on this new block, and he's putting his hand over his reputation and stating that we should ignore everything we've come to know about the variability of TIM interfaces, and accept that the TIM interface on the Intel TTV is now suddenly and conveniently invariable, and to believe that everything else is faulty or irrelevent.

Say what? Come again? I don't care if there is any shred of legitimacy in the apparent reasons for asking everyone to ignore established facts, but heck, it sure doesn't look good. I am sorely annoyed that Bill has chosen to label any of my calls for a quite basic level of reasoning and derisively refer to them as:

Originally Posted by BillA

a “Stew approved” test mechanism

I contend that the mechanisms need to prove themselves, before all of us, not for any one individual, especially in light of what we know about the proven variability of interfaces. If any mechanism comes along and is generating results that don't agree with established procedures, and introduces new variables that are known to cause variance, then the ABSOLUTE LAST thing we should be doing is then saying "No. The device introduces no variances" in absolute contrast to established facts about interface variances.

[maxSaleen]

ms
read the Swiftech stuff carefully
Swiftech's are now device C/Ws, previously the C/Ws were case-to-air/water and included the TIM joint
Swiftech data is comparable if on the same graph, not so between graphs (different bench in almost all cases, I know for I did it)
with offsets you can daisy-chain, but with errors probably

Thanks for the heads up, Bill, I won't bother. It won't help things, it appears.

I've never been a TIM fan. Sure it is easier for the OEMs, but it sure does make things hard over here at procooling. I'm with Stew (not that my opinion matters, a boy among the kings here) on the TIM varience issue. Here's why:

Case1:
About 18 months ago, I had an Intel 2.8e which I overclocked (great chip, hot as hell though). On air with a MCX-478 my maximum stable OC was 3.2. Temps were high (no sense in posting as I have no accurate way of measuring). When I put some of my watercooling gear on (an old BIP, MCW5000, hydor L-20) my maximum stable OC went up to 3.4. I say to myself "Strange. This chip is regularly hitting 4.2. Bad chip perhaps?" I checked out the bin number (week 23 malay if I remember) over one VR and saw that some had gone as high as 4.4 on water.

What I noticed was that my temps had not improved that much, though I was using a vastly superior thermal solution on a very hot CPU. After a couple of remounts I thought to myself "Maybe a bad IHS?" It would take to long to lap the sucker, so I took it off. Next (scary) mount of the block: major decrease in temps, maximum OC shot up to 4.0.

Case 2:

In a recent upgrade (on a different rig), I swapped out an old 2.6C for, wouldn't you know it, a 2.8E (great value for the old 478s). Checked it out on VR again, seemed like I had another good chip (week 37 malay). This time, out of the box, IHS on, the thing does 4.0. Temps are surpisingly good (using the same MCX 478). Pop off the IHS, install new WC gear (Dtek WW, MCP350, BIP3). Temps improve, but not nearly as much as the last time. OC went to 4.2, also not nearly as much as last time.

Obviously not science, but it is good anectodal evidence. The first 2.8E couldn't go anywhere near its max OC with the IHS on. The second 2.8E went almost to its max OC WITH the IHS on. This would seem to say that the IHS on the first chip was severly limiting my ability to cool the chip, while the second IHS had much less of an impact. I consider that to be inconsistancy (varience).

[plywood99]

MaxSaleen,
Iv'e experienced the same thing on the A64 platform. Ihs / tim joint just too variable. Personaly I'm convinced that bare die or die sim is the way to test.

[Annirak]

That's why this statement is so nonsensical.

Originally Posted by BillA

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 ?

There may not be measurements, but there is overwhelming anecdotal evidence. And that is something that someone in BillA's position really ought to know or check before blowing off steam at us.

[ricecrispi]

We got our review by robotech who did another fine job. I feel we don't give him enough credit that he deserves. He really went out of his way on this one and his reviews are getting better.

Maybe the next thing we need to do is finding out why the intel TTV is producing such different results. Maybe someone should contact Swiftech and bring up the topic. It was their data and their block that brought all this hoopla. THey should have a hand in explaining why they got different reults other than saying different testbeds. Maybe they should retest or try some changes in the setup.

As for the intel TTV, I can't trust data coming from it unless we found out why the results it got vary so much from what robotech got. I'm not going to say it's a bad or invalid tesbed, but I trust robotech's review more. We talk about grooving ihs, what the thermocouple is measuring and IHS variablity but that doesn't prove anything until we actual find out why. We can make as many theories as we want but they don't meaning anything if we do go out, test them, and prove them
.

[jaydee]

Good god.

so far it seems only Lee and I seen to understand the purpose of a heat die, a simulator - not THE TRUTH
I too would expect similar results from pH, and I would expect his conclusions to be similar to Lee's

So all the data you done on your die sim posted on overclockers and Thermal Managment website is now crap as with all of your data Les was using to make predictions and models.


Basically all that work tells us no more than we knew before the tests.

[Les]

Originally Posted by jaydee

..... ..now crap as with all of your data Les was using to make predictions and models. ....

No, JD (link)
To repost:
"It would be nice, but this is the problem
There is not " a good approximation for C/w vs Q", if you are referring to measured "C/W".
Good data for "Swittech Storm"(G4) but as discussed here interpretation is impossible without more details.
Good data for LRWW but requires leap of faith re C/W(TIM)(different TIM and wb surface finish) and possibly "sensor offset allowance" (Die conductivity(360-400w/mk and sensor position(seperately reported as 1.9939mm))
Possibly the only "good and interpretable" data is the SwiftechMCW6000 on a 144sq mm heat-source (Incoherent's presentation here and subsequent posts)."

Edit: The crude models are for uniform-flux from Die (be it CPU or Simulator)
Should apply to Heat-Dies which are designed to give uniform-flux.
Although data may have to be manipulated to account for "sensor introduced errors" (link5 in next Post)

[Les]

Presently focusing on the analysis of the various test beds.
Progressing tortuously along the lines indicated by : link1 , link2 ,link3,link4 and link5 .
Here "link1 & link2" are TIM, "link3 & link4" are "position & size of heat source" and "position & size of sensors", and "link5" is "sensor introduced errors"
Still a long way to go, an IHS has not yet even been introduced, but....

Edit. Changed "link5" from Post to Page

[Incoherent]

I am going to go out on a limb.

Why are we even discussing this?
I have been considering whether or not to embark on a series of tests on a die simulator where I attempt to characterise waterblock behaviour with an IHS.
After having read this thread I am hesitant to do this, I believe the exercise to be futile. Putting a heat spreader on top of a CPU die with a waterblock on top is exactly the same as putting a Heat spreader under a waterblock with a CPU die underneath.
The point of that statement, stupid as it is, is to illustrate that you can turn any high performance waterblock into a complete dog by putting crap in the heatpath. There are lots of analogies. A thoroughbred racehorse will beat shire horse around a racetrack. But not if you have it pulling a heavy cart through mud. This community only deals with unimpeded racehorses as far as I am concerned. The business of cargo transport is a whole other industry, that which I think Swiftech is moving towards with the Apogee. ie general, reliable, sustainable and above all economical. No-one breeds racehorses for cargo transport.

In my world, using an IHS capped CPU for performance should not even be considered. The whole basis for my testing is and always has been to establish performance levels that would enable me to build the best block I possibly could. Having spent time building waterblocks, or money buying waterblocks, or effort designing waterblocks, WTF would I render that irrelevant by hitching up a cart to the racehorse?

This is Procooling, it's always been about the overclock at the end of the day, that's what the watercooling community started from. Leaving an IHS on is not what we are about. Waterblocks designed to adequately cool an IHS capped CPU are another market entirely, potentially a much bigger one, and the right way to go for Swiftech IMO from a purely economical standpoint.

I am going to stick to testing, using a small heat die in some form, because I believe that that best represents a small UNCAPPED, CPU die. Despite the concerns of whether it fully represents a real CPU I believe it to be the only reliable way to establish waterblock performance. Because the conditions in a heatdie a are known and if the setup is known it can be modelled. If it can be modelled then absolute performance can be established from real data.

I refuse to believe this is not possible.

And I am not talking about data only valid for one testbench, I am talking about absolute performance which would be reproducable in any other testbench where the the conditions are known. In a TTV they are not (at least by me), In a real CPU they are not, and you haven't a hope in hell of knowing the conditions in an IHS capped CPU.


P.S. Ben. I also refuse to believe that performance is dependent on power level in a die simulator. If it is the calibration is out. In fact I think that this effect is a way of verifying the calibration of the temperature sensors. Take temperature readings at different power levels, plot a curve dT water-Sensor vs Watts, extrapolate a linear down to 0W. If dT for all sensors is not zero at this point, you need to recalibrate.

[Les]

Originally Posted by Incoherent

I am going to go out on a limb. ..............................

Think I perch on a twiglet of the same limb

[gone_fishin]

Lets get real for a minute. Those on the real cutting edge of overclocking do not bother with watercooling. They spend near $1000 for the very latest cpu (multiple ones of a batch), pry off the IHS's, put a copper cup filled with liquid nitrogen and overclock them cpus with no worry of destroying the entire investment. So do not assume watercooling at above ambient temps is the cutting edge, it is somewhere along the line where the mediocre cheapass section of overclockers reside.

[Marci]

Sorry but that's bollocks. You're talking about the cutting edge of the benchmarking crowd where OC's don't have to be sustainable other than to run the single bench in question. Radically different crowd.

[Incoherent]

Perhaps I should have said:

It's all about the maximum sustainable overclock.

With hardware that is affordable. If you get $1000 dollar CPU performance with one costing $100, you have a lot to play with.

[gone_fishin]

Originally Posted by Marci

Sorry but that's bollocks. You're talking about the cutting edge of the benchmarking crowd where OC's don't have to be sustainable other than to run the single bench in question. Radically different crowd.

No, they overclock. The above ambient crowd overclocks. There is also the phase change below ambient crowd who also overclocks (at sustained periods). The above ambient crowd is the bottom of the barrel so to speak, with the least performance gains and the least investment.

So I ask, what is the point of benchtesting to this crowd at this point (other than it being a hobby). There are physical limitations. For those enthuseists who wish to make the next leap in performance the obvious road is a reliable phase change system.

In that context this block by Swiftech appears to be overdesigned with far too much investment in getting data to use to market it. Now we see in fact that their own investment has turned against them and the very data they used to market it is the flak they are recieving. Nobody in their right minds will ever attempt this path in waterblock marketing to the bottom of the barrel again.

[stev]

[BillA]

Originally Posted by gone_fishin

No, they overclock. The above ambient crowd overclocks. There is also the phase change below ambient crowd who also overclocks (at sustained periods). The above ambient crowd is the bottom of the barrel so to speak, with the least performance gains and the least investment.

So I ask, what is the point of benchtesting to this crowd at this point (other than it being a hobby). There are physical limitations. For those enthuseists who wish to make the next leap in performance the obvious road is a reliable phase change system.

In that context this block by Swiftech appears to be overdesigned with far too much investment in getting data to use to market it. Now we see in fact that their own investment has turned against them and the very data they used to market it is the flak they are recieving. Nobody in their right minds will ever attempt this path in waterblock marketing to the bottom of the barrel again.

Thanks gf, again

On 11/30 a thread was started “product testing today – who is being served ?” specifically addressing the potential for using a CPU as a so-so heat source intended towards low budget / semi-technical review sites. There has been limited response to this topic and the probable reason could be the strong disapproval apparent in this thread for both the thermal impediment and the additional source of uncertainty introduced into the DUT. I believe it is now correct to say that the more vocal on procooling (at least) wish optimum testing done with a bare die sim., I can accept such w/o question, that is the popular opinion.

There is no need for my technical concurrence on using a bare die sim, I and others have used such for years. The limitations of practical use however, I believe I may be uniquely aware of as I alone have visually observed with optical flats the surface wear over time. But it is quite clear that ‘facts’ are not going to be accepted on my say-so alone as I am not a “trusted’ source, and that virtually any anecdotal un-named source can trump my ‘facts’. Yes, such is the court of public opinion.

The ignorant conceit displayed by some, in a presumably technical forum, is amazing to one having started in engn some 40 yrs ago. Our technical society is built on Stds and Specs (ASTM, API, AWWA, NSF, ISO, DIN, etc., etc); these and company/product specs are how companies communicate. Contrast this with the comments of those who ‘cannot trust’ data because it was generated by the company itself. OK, now it is the presumed veniality of the company, they must lie for advantage. Well kids, keep that shit on procooling – you cannot work in engn with this mentality.

The Scientific Method: Define the Conclusion, and Assemble the Facts to Fit
Re the TTV, why not start a thread soliciting hands-on experience with TTVs ?
(I am a ‘not trusted source’ so my experience is invalidated; commercial interests, blah blah)
Sport has been had misquoting out of context the P4 TTV info. Anyone read the goddanm thing ? What is the heating element ? Whose hairbrained idea was it to consider the TTV as a CPU ? And transfer (presumed) problems ‘generally attributable’ to IHS TIM joints also to TTVs ? GOOD science fellows, why not consider that almost no one here knows anything about TTVs ? But of course it’s a source problem again. I suggest that TTVs disappear from our vocabulary, other than as a topic to beat up Swiftech of course.

It was distressing to have Stew raise the issue of my benefiting by pimping Swiftech products, and being a disgruntled ex-employee as well. If I was not a pimp for years before, why would I change ? Ah says the dirt-digger, because Gabe owes you money ! Liar Gabe is paying $380/wk, when he is so inclined; but if Gabe goes bellyup I am paid at once. So why would I be pimping Gabe’s stuff ? Again it returns to the issue of technical impartiality, now Stew too has called me out (to be polite, lol). [Why was this issue raised ? It is not technical, it is a personal slur.]

BillA, the technical whore – just ‘cause it feels good
I’ll take a break and see if I cannot develop a more positive attitude, later

[Annirak]

Originally Posted by BillA

But it is quite clear that ‘facts’ are not going to be accepted on my say-so alone as I am not a “trusted’ source, and that virtually any anecdotal un-named source can trump my ‘facts’. Yes, such is the court of public opinion.

No, Bill, it has nothing to do with you claiming that some things are facts, and that any un-named source can trump you. It's that your assertions are don't pass sanity checks, and you insist on ignoring variables. That ain't engineering, it's marketing. Start being an engineer again, and we'll believe you.

Originally Posted by BillA

Contrast this with the comments of those who ‘cannot trust’ data because it was generated by the company itself. OK, now it is the presumed veniality of the company, they must lie for advantage. Well kids, keep that shit on procooling – you cannot work in engn with this mentality.

It's not so much that we--well... I--don't trust the data because it comes from the company, it's that I don't trust the data because the testbed isn't characterised, is designed for validation, not thourough testing, and is missing a critical variable, which is the only one that anyone who's trying to overclock CARES ABOUT!

Originally Posted by BillA

The Scientific Method: Define the Conclusion, and Assemble the Facts to Fit

Again, we have marketing, not the scientific method.

Originally Posted by Scientific method

Scientists use observations and reasoning to propose tentative explanations for natural phenomena, termed hypotheses. Under the working assumption of methodological materialism, observable events in nature are explained only by natural causes without assuming the existence or non-existence of the supernatural. Predictions from these hypotheses are tested by various experiments, which should be reproducible. An important aspect of a hypothesis is that it must be falsifiable, in other words, it must be conceivable to prove the hypothesis to be false. If a proposition is not falsifiable, then it is not a hypothesis, and instead an opinion or statement outside of the scope of scientific inquiry.

The scientific method is a process that follows "observe a phenomenon, formulate a hypothesis, test the hypothesis, repeat"