RE the IHS with AMD vs. Intel; Intel used to have poor IHS to core mating, back in the Northwood days. I experienced a 2.6 which after removing the IHS I lost 9C, with 3 other Northy's that I popped the lid I lost about 3C. Intel today has a MUCH better way of bonding the IHS to the core and it is virtually impossible to remove the IHS from an Intel CPU today. I fear from what I am hearing that AMD has garbage IHS to core bond, until they fix this issue it would be best to pop the lid, this I am sure of.

current and applicable, if not defensible would not be used (goddammit Scott, you know me)
there is a 'bad' data set, the causes of which are speculative as some crucial facts/conditions are not known
but there are many good sets of data, both before and after the 'bad' set was generated

the TTV data is also generally confirmed by their CPU based testing

this is an amazing rush to prejudgment, nay outright condemnation, by an authority in the field
but I could be the fool instead:
Cathar has had a MCW55 for a number of months,
could it be that he DID test it and knows for a fact that the Apogee (derived from the MCW55) cannot possibly equal the Storm ??

the plot thickens
Cathar, did you test the MCW55 ?
anything you wish to share ?

Bah I got locked out of forums and couldn't reply earlier. I (lol?) forgot my password.

Anyway my rationale for sticking with AMD CPUs for testing is due to the diode being accessible for my data acquisition. Motherboard readings are both coarse and unreliable and that's a bad combo.

But while playing with my hacked up AMD64 I quickly noticed something that probably plagues all modern CPU users: the damn things pop out of the sockets when you try to unmount coolers! This bends pins, this can mess with my wiring, and this probably affects the TIM interface as well over time. It also gives me pause before spending a lot of time getting test numbers (imagine if I have to start over with a new CPU every couple weeks!).

If I were to "pop the top" on my test box then this would be the reason why.

And to be fair, questions regarding the TTV started to surface LONG before the Apogee block, want to make that clear. As per Bill's testing (I think) the MCW55 just about matches the Storm as per Swiftech's data. A post by Les went largely ignored in the beginning of Sept. as seen HERE .

I hesitate to suggest such heresy, but are no Intel CPUs candidates for OCing ?
focusing on the overclockability capability of the sink (yes I understand that there is compression/loss of 'sensitivity'), but even so ?

if you go with AMD do consider popping the top just for consistency
- just like resurfacing a heat die, need to add this devil to the mix too

Hehe Bill, I run my P4 at 4530Mhz every day, 24/7 at load too folding. Here is what it will do with my plain'ol water setup lol.

Validated: http://valid.x86-secret.com/show_oc?id=19923

lol
so can we save ourselves some grief and, for durability, use an Intel CPU to characterize CPU/sink performance when an IHS is used ?
(AMD data would be equal or worse depending on the TIM joint - lord that must tickoff someone loving AMD)

Yes for IHS based CPU's Intel is where it is at in terms of IHS to core integrity IMO. Also, the NEWER IHS are far better than the old IMO.

New style
http://www.anonforums.com/builds/640/cpu2.jpg

Old
http://www.anonforums.com/builds/32e/unlappeds.jpg

And AMDs current IHS is similar to old style intel as shown above - more or less identical tbh. Whereas Intel's NEW IHS is very similar to earlier AMD CPUs... K6-450 era...

http://www.watch.impress.co.jp/pc/do...309/k62450.jpg

altho obvious refinements due to die shrinkage.

ok, groove it to the exact center to take no more than a 40ga TC (smallest from Omega sets the size)
think about the TC tip and getting the IHS top surface temp

hope Lee or pH wants to do this, it would address the larger portion even of the enthauast users and 100% of the more casual owners who upgrade

err, not the AMD one

Wow that looks good there Marci, didnt realize they switched. Now, can that lid be easily popped? Or have they gone to solder/epoxy as intel has? FWIW, I tried to get a soldered/epoxy'd ihs off of an Intel cpu, to no success. I stuck in in a 400 degree oven for an hour, froze it overnight, soaked it in googone, nothing worked. Believe it or not, the CPU still runs to this day.

This whole IHS removal should just be excluded from the minds of testers. Until now, waterblocks that were tested were tested against "stock" cpu's. No modifications required.

Now... all of a sudden a company creates a WB specifically to address the new/larger heating surface... and testers think that it should be discounted because "real enthusiasts" pop off the IHS? Ridiculous.

I've been overclocking since 1996. If I'm not an enthusiast... I don't know who is. Yet I REFUSE to take a knife to my $600 X2 4400+ processor. It ISN'T going to happen, and I am quite sure I am not alone here.

Why would you risk your investment, when you may be able to accomplish the same cooling with a different waterblock? Not saying either way which block accomplishes this, but that is the claim.

I want the block that can cool my $6-800 processor as it is sold, so if I need to RMA it... I don't have to get out the epoxy and try to rip someone off.

No, I suggest you go read the thread, this is NOT about popping IHS's. The AMD ihs to core integrity is SHIT, no two ways about it. And, you obviously have not been reading or you would KNOW that Intel IHS's are almost impossible to take off :rolleyes:

I have read the thread, but seemingly mis-interpreted.

I fail to see the correlation between the AMD->core joint, and the performance of a waterblock.

If you take the same CPU, with its borked IHS... and test two waterblocks... the amount of heat given off by the cpu does not change. The IHS interface remains constant... whether it is borked or not.

ie: the amount of heat a waterblock can remove from said cpu... does not change based on how well the IHS is bolted on.

WB performance vs IHS problems?
I don't see how the two relate. I understand that it makes getting an accurate temp reading more difficult, but I don't see how it relates to the performance of the block.

Exact same heat source, two different blocks. I do not understand how the heat source being screwed up in some way affects the WB's ability to remove the heat.

"Until now, waterblocks that were tested were tested against "stock" cpu's."
not
only those not having or making a heat die, many have made heat dies (not only me)

Because when an IHS is on for shit, each mount will further degrade IHS to core integrity, c'mon i don't claim to be 1/10th as knowledgeable as Bill, Cathar, pH, Lee, Les and a host of others, but surely this fact should be common sense.

I can accept that the AMD bonding is not consistant enough for repeated WB testing (@5 mounts per WB).
Beyond that you have established that Northwood bonding was shit, but not that the current Intel system is superior in any way to the AMD system for normal or extreme use. AMD's TIM may be superior to epoxy. (Didn't Intels overheating heating problems begin with it's use?) :) Some may consider the inability to remove Intels IHS a fatal flaw.
If AMD's joint broke down in the first few remounts you would have a point, but I haven't experienced any problems with either of mine.

edit: "It changes from CPU to CPU, and one IHS capped CPU will appear to favor one waterblock, while the exact sameCPU type from the same manufacturing batch will favor a different waterblock." Now this can indicate a problem, need American, union workers to correct!!!

Some black crap - 10 secs with a craftknife blade and it's off... no difficulty particularly - unless you push in too deep and swipe off a few resistors etc... LX tape on the blade prevents that tho...

k - so you KNOW the power of your cooling system, yeah? So, imagine if in 6 months time you decided to buy a second x2 4400+, and swapped em straight over, but for some reason the new chip is running 10 deg hotter than the old... AMD won't replace as there's no fault - the chip works, just does so slightly hot - but there's nothing that defines what temp the cpu should be running at according to manufacturer, as long as it's below xx deg C... so nothing to go against in the argument that you want your CPU replacing cos it runs 10 deg hotter than another... you're stuck with that CPU. That's what folks are having to do here. They move from one CPU to another, that should give them an expected marginal temp increase.... yet they're seeing a RIDICULOUS temp increase for no apparent reason. S'not in the mountings. I've had em brought in to the store for me to try and sort, and only conclusion is to remove IHS as the chip isn't faulty per-se... or deal with the fact that your PA120.3, Storm G4, & Laing D5 with Panaflo M1A's can't keep the only thing in your loop within 15 deg of ambient at stock under load when in fact it should be more than easily capable. So, a £400 chip vs a £300 cooling system. And you're losing all the benefits of that cooling system cos of some crappy bit of tin they've badly gooped onto the die. No RMA Option on the kit - it ain't the kit's fault either.

For some removing IHS is a necessity. You're lucky that you haven't had a bad chip that has resulted in the NEED to remove the IHS.

Those who've experienced such will likely remove the IHS from every CPU they buy thereafter just to remove the ballache of a chance that you mount it all up and you've got ANOTHER duff one...

The problem I have with IHS's, at least from the point of view of enthusiast overclocks is this.

There have been a number of cases reported in forums where an IHS capped CPU will "favor" one waterblock that ranks worse, while the supposedly better waterblock will both perform worse temperature and achieve a lower overclock. After many suggestions and remounts, this was the conclusion that was reached.

Then said users changed the CPU to another CPU of the exact same type, and the stories are reversed. "Worse ranked" waterblock sees much higher temperatures and a much worse overclock.

What is evident here is that subtleties in the IHS manufacture and waterblock geometry are at play. The CPU die to IHS TIM joint is NOT invariable and cannot ever be assumed to be so, and worse, cannot even be assumed to be consistent on a per cpu type or even on a per waterblock scenario. It changes from CPU to CPU, and one IHS capped CPU will appear to favor one waterblock, while the exact sameCPU type from the same manufacturing batch will favor a different waterblock.

I've said it many times before, but I'll repeat it again. If we're looking to quantify the cooling effect of a generic IHS scenario, then we HAVE TO remove the variability of the CPU Die -> IHS joint, which means machining a heat die that looks just like an IHS capped CPU, but is a solid one-metal-no-join piece design.

For those struggling to picture what I mean, imagine something like a 13x13mm (or whatever sized) copper heat die with a "mushroom" IHS lid on it, but all one solid piece. In fact the word "mushroom" pretty much exactly describes the sort of one-piece thing I'm talking about. Heat energy is applied at the mushroom stalk, the temperature of the die is taken from the center of the stalk just below where the IHS "head" starts to fan out. Heck, groove the IHS head and stick a TC in there too, ala Intel TTV style, and then we can validate that the relationship between IHS surface temp and stem temp is consistent.

In this manner the variability of the CPU die -> IHS joint is totally removed and we now have a fixed invariable model that is not subject to TIM joint degradation or block-IHS-TIM-die geometry variations.

Of course we can argue all day that "this is not exactly what an IHS capped CPU looks like", but what are we going to do? Measure and present results based upon a device with what many have observed to have huge variations? The best we can do is to model/measure against an invariable model and the day that the CPU makers get their act together and provide an invariable CPU die -> IHS interface, then it'll all make sense, and certainly a lot more sense than measuring performance based upon a variable laden model where if we start to tune just for that specific model we really don't know what variable we're actually tuning for and which way the effects of that variable are swinging in terms of waterblock->IHS->die geomterical interaction.

Keep it simple. Keep it consistent. Don't play the variable game, and don't develop waterblocks around unquantified variables.

That's what I'm proposing.

Got curious about this statement, from athe Apogee page
The Storm (@2g/m) is superior to the Apogee (@3.3 g/m) with any single core cpu made the last few years, and the next few years to come. I guess I had better toss my 114mm San Diego and 14Xmm Newcastle as they are "earlier generation", along with my Storm.
Gotta love market speak.

I don't understand how the Apogee can cool the center of the IHS better than the Storm, single or dual, as the cooling power is concentrated there. If the diameter of the jets matrix on a Storm is less than 21.7mm (apprx. diagonal of X2 2Mg cache) I can see how the Apogee can outperform the Storm on a dual core as far as overclocking.
Can't wait for more data to see if the results hold up.

Oh, and I tried to quantify the MCW55 on my video card GPU, but lack of an adequate temperature resolution on the card prevented me from gathering results good enough for me to make any conclusive statements about it. What was evident to me though was that it was performing roughly on par with an old Hydra prototype which was assessed on my CPU testbed to be around 0.5-1.0C worse than old original all-copper White Water at ~110W heat load. Still, such results are not conclusive in any shape or fashion and I have deliberately held off making any comments about the MCW55, other than to suggest in public that it is certainly a good performer. No GPU block has ever enabled a higher GPU overclock than the old Hydra prototype and the MCW55 was certainly the closest thing I've seen yet to it. Still, the old Hydra prototype is now WW beater, let along a Storm equaller. Take that for what it's worth, which in my opinion is worth very little and such an observation should be taken with a very small grain of salt. I have next to no faith in the finality of that performance assessment of the MCW55 other than to say "it seems pretty good for GPU's".

Am presently trying to get a suitable CPU mounting bracket for the MCW55 machined up so I can quantify it directly on my CPU testbeds. One of my machinist's CNC mill's broke down though and so I have next to zero machine time available prior to Christmas.

Happy that Cathar talk about that, it was the same idea as mine to make my proper die with IHS but without TIM to avoid assembly, never realized because no time... Not difficult to mill at all, some care should be take to avoid conduction secondary paths. I would to use a MOSFET because of the small size and capable to dissipate a lot of power easily and easy to control, power resistors or cartbridge are generally too big for a small and light die like that (no need to heat 500 gr of copper, too long and useless...). TIM die/IHS is only a gap in Rth, its thickness is too small to change flux repartition, so we could pull out this element and get a solid one piece die with a flat surface above. IHS and die center would be instrumented with very small TC using EDM holes (diam. <0.3mm). This kind of die would be much better than bare die for sure, nearer to reality in terms of flux spreading (always different with any WB !)

In french on attached pic but you'll understand... I think Cathar's idea is like mine if I understood well.

Yep, that's exactly it Roscal. Easy to machine.

I've actually been saying this off and on for quite a while now. Seems no one took notice. I myself have been working towards getting one made up, but always seemed to have more pressing matters with my real job, but lately with all this IHS waffle I've been becoming increasingly more motivated to do so.

Problem is to know the die area below IHS, I put 12x12mm at one moment, but now dies are larger and often change with generations. Make a die every year is a pain... I don't know really if differences between 12x12 and 16x12 for example will be great, not sure, because IHS acts as a good smoother... Need tests and data, or numerical studies to get a first approach, I'll make tests with Fluent one day, not difficult to model at all...

out of the 800 or so computers I've sold through my retail pc store in the last year.... All have had AMD processors. At least half have ben Athlon64's with IHS on them.

I've only had two of them report extremely abnormal temps. Of those 2, both cases were caused by faulty motherboards.

I fail to believe that re-mounting on the same cpu will degrade the IHS weld to the core. It is either on right, or not on right. It doesn't magically start coming off by itself. ... or at least I've never heard of it happening.

CPU die sizes, including the dual-core dies, seem to be mainly sitting between 140-220mm^2, with the AMD dies typically being the most oblong, and the Intel dies being the squarest, although that has changed somewhat with some of the dual-core Intel dies. The most oblong long:short edge ratio seems to be no higher than 1.8:1 to date.

So perhaps a ~180mm^2 die size, with a long:short edged ratio of 1.4:1 would be a suitable middle ground? A 16.0 x 11.5mm die size shape? Just throwing a suggestion out there. Everyone's going to have different opinions on what would be most representative, or what would be the most acceptable "middle ground" unless testers are prepared to provide results on multiple IHS testbeds.

no
it is improper to characterize all CPU/IHS TIM joints as being intrinsically variable to a huge extent
that some may degrade so over time and use has nothing to do with testing (don't use old shit)
that some may be inconsistantally mfgd is another separate issue (don't buy cheap shit)

Cathar, take your entire post above and read it in the light of a consistant TTV CPU/IHS TIM joint being a fact
there is nothing to your post, it is ALL totally predicated on variable TIM joint performance
remove that variability and all that's required is learning what the new numbers signify

I have cross-tested multiple TTVs, and tested them over time; they are fine tools
-> what is perhaps not known/being applied is an appropriate correction factor
nothing wrong with a TTV

the real difficulty is that TTVs are not available

anything positive ?
there DO seem to be (today) CPUs not having IHS issues
"Keep it simple."
we agree, test the actual configuration where possible; and since there ARE CPUs w/o an IHS problem . . . .
but this still leaves the quantification of heat unresolved if using a CPU source

". . . which means machining a heat die that looks just like an IHS capped CPU, but is a solid one-metal-no-join piece design."
and this will replicate the heat flux through the TIM joint, and the subsequent lateral spreading through the IHS ?
Roscal says so, I have no good basis to object
where is the DIE TEMP to be measured, and how ? (I missed this in your post)
Roscal did make a suggestion, but
- how do you establish the correspondence of this measured slug temp with a CPU temp ? where the temp is taken and its value are irrelevant, the correlation is key
there are some assumptions here, as did I with my old heat die; I defined it so
how many today would accept that ?

junk the hat, work on valid CPU specific correction factors for a simple post
NB if you want to assess dual core cooling you will need a dual core source (or do junk testing)

-> how is the die face to be maintained ?

do you know what the TTV has that no die will ever have ?
repeatability, lesser absolute performance but a greatly reduced range

Orkan
many mountings or severe thermal cycling will do it
outside of the scope of 'normal use'

Marci's experience is different due to his customers

Bill, sorry, but I'm going to have to say it.

I really don't understand your steadfast defence of a testbed where the TIM layer is not quantified. Really. Please. You don't know the die temp on the TTV, so you cannot EVER conclusively say that it is invariable now can you? Why? Because you don't really know.

Truly, I find this head-in-the-sand approach from you most out of character and disappointing.

There. I said it. I just had to get that off my chest. I have never ever witnessed ANY person who claims to hold to a scientific testing methodology and then blatantly dismiss what is a potentially variable, especially in light of the admitted "bad data" that has been observed, and then say "sorry - no - I don't care - it's invariable".

Seriously, please. Pleading for ignorance now? Why? I cannot understand this mentality from you. It is so against everything you've ever stood for in the last 5 years.

Can't have it both ways Bill. Dual core CPU's are just a single piece of silicon and look just like a single CPU silicon die externally. Heck, some of the newer dual core silicon dies are even smaller than the older single core CPU silicon dies and present more localised and compact thermal footprint.

While championing the TTV for dual cores, you're also putting forwards a case that a single middle-of-IHS case-temp is suitable for all scenarios.

Can't have it both way.

I find this defense of the TTV model utterly and totally incomprehensible in light of the admitted and unexplained bad data. In scientific terms, the TTV is a theoretical model in which observed phenomena cannot be explained. Either find the explanation or the model is as invalid as any aspersions you care to lay at any other testbed's feet.

What will happen Bill if independent testers start to find results that do not agree with the TTV? Will you then engage in tearing down every independent tester that does not usa a TTV?