Getting seriously weirded out here... (thermal interface related?)

[Cathar]

Please, this is not a cynical attempt for promotion. I just want to share/ask people if they've got any ideas as to what's going on here.

If you're not familiar already, recently some kind individual managed to source some pure silver to make a few of these blocks.

Okay, so I've done some temperature comparisons between the copper Cascade and the silver Cascade, and basically the nod goes to the silver Cascade to the tune of just over 0.5C (best that I can determine) given my test CPU heat load.

What was weird to begin with though was when it came to overclocking. I must've mounted the copper Cascade onto this CPU (AMD Barton XP2800+) maybe 20-25 times in the last 2 months, and compared it to a number of other highly respected commercial blocks. In all cases the copper Cascade came out ahead of the other blocks both in terms of overclocking potential and temperature. The White Water came in a tied second with another commercial block (which is a lot more expensive than the White Water) being about 2C and 15-20MHz behind the Cascade.

Okay, so I've run dozens of tests and have a pretty good idea of where things stand in relation to each other, and the silver Cascade is an "in theory matching reality" predicted ~0.5C better than the copper Cascade.

That is until I tried to overclock. I had never been able to achieve greater than 2680MHz Prime 95 TT stable with the copper Cascade and the Barton. The machine could boot into WinXP at 2695MHz, but was unstable and would constantly restart the "Explorer" interface, so basically it was good enough to take a quick screen-shot and that was it.

With the silver block I was able to hit 2720MHz Prime95 TT stable and had issues getting the machine stable above that point "at all". Almost like a razor-fine line had been drawn in the sand with respect to stability/bootability. If I tried any higher, the system failed to make it into Windows and sometimes wouldn't even POST.

That was, until about 2 hours ago. Just browsing the web and glanced at the CPU temperature and it was about 3.5C lower than where it had been sitting at for the last 4 days of testing with the silver block. Nothing had changed with the system, and the room temperature (and correspondingly the water temperature) was slightly warmer than "normal" as I've got the flu and have gotten the heater turned up a bit.

Okay, it's just a thermistor error is my first reaction, but hopeful, I try to overclock past 2720MHz....and it works!

I'm now typing this at 2737MHz with Prime95 TT running in the background. I tried at 2749MHz and it booted into WinXP just fine and even ran Prime95 TT for a few minutes before failing, so I dropped back to this speed.

I am totally weirded out now, because this is something I've never ever seen or experienced the likes of before. A ~60MHz overclock gain is something I would have never ever though possible.

I have a theory though, and here we get to the point of the post (finally). The block is made out of pure silver, which is fairly soft and deformable in comparison to copper. Pure silver almost reaches the levels of pure gold in terms of softness. You can bite it and leave an indentation and it scratches pretty easily. The silver plate is thick enough and strong enough, however, to not flex under mounting pressure (which is about twice the AMD recommend maximum pressure at present), but that doesn't mean that the base-plate hasn't micro-scopically deformed to more closely mate with the CPU's surface.

Is such possible? Could this be the explanation? Is perhaps the next step for better thermal contact actually the use of soft deformable metals as base-plate material?

[BrianW]

Well if that is the case, the block would need some good lapping if you move to a different motherboard or cpu die size?

I do think what you theorize is pretty close to the mark.

BrianW

[AXDA2600DKV3D]

wow, i think you discovered something great and something that will blow everyone away if this is the case

anyone care to try it with gold? hehe

but it sounds like a good theory... what type of thermal paste or material are you using?

[winewood]

I have seen systems gain and loose efficiency with time. The theory I am thinking... a.) the heat and constant pressure allowed the silver to form to the core and thus improved the tim heat transfer. b.) there may be a natural air pocket or stale zone that is hard to get flowing. The pocket of resistance just gave way, and thus the flow pattern changed.

Upon thinking of b I would recommend the following. Take a single tube and point it away from the outlet barb. Allow it to start a motion for the water to circulate within the block. This allows for no heat pockets and constant motion.

.02 cents.

[Cathar]

Originally posted by BrianW
Well if that is the case, the block would need some good lapping if you move to a different motherboard or cpu die size?

I do think what you theorize is pretty close to the mark.

BrianW

If it is what is happening, it would be on an absolutely tiny scale. I would imagine that a quick 1 minute lap with some 1500-grit paper would be enough to remove any "imprinting" given a very flat base to start with.

[Cathar]

Originally posted by winewood
b.) there may be a natural air pocket or stale zone that is hard to get flowing. The pocket of resistance just gave way, and thus the flow pattern changed.

I have ~12PSI of pumping pressure applied to this system. Each jet's velocity is approaching 7m/s and was flow tested in open air. There are no air-pockets in the block.

[mad mikee]

Sometimes after running for awhile at a higher speed, the cpu gets 'acclimated' to the higher speed and sometimes the 'burn in', helps reduce some of the resistance, allowing cooler temps and even ability to reduce voltage to get the same speed. Some variant of this may have happened though I seem to have hit a brick wall around 2400 mhz (still tinkering).

Also details about compound, time since installed? Did you need to increase voltage to get past 2720?

Last but not least, the fact that you were able to run at higher speed/temp may of course generate more heat which would tie in with your idea about what is happening. I would guess it is some combination of all the above. I think I'm going to let my machine fold at 2500 mhz for awhile so I can get a little more. Thanks for the inspiration!

[airspirit]

It makes sense that a soft metal would deform around the die. I would hesitate to remove it and risk mismounting it next time, but if you did and managed to measure the profile of it I bet you would find an indentation by the die combined with a general warping of the base plate of your block. Silver is very soft and it doesn't take much pressure and heat to make it pretty malleable.

[Gulp35]

if you are using a silver based thermal Interface material could the thermal transfer from silver to silver be better than from silver to copper? I have no idea if this is the case I am just throwing out the idea. It could be horribly off kilter.

Again no scientific backup, just a thought, go ahead flame me.

[AXDA2600DKV3D]

actually i was thinking something to do with arctic silver if thats what he was using

wouldnt arctic silver do the same as a silver base atleast? silver gettin crushed into the holes, grooves, and gaps?

so i dont see what the difference would be

[redleader]

Coppers a very soft metal anyway. I don't see why silver would be any different.

Probably just a fluke.

[Cathar]

Just to answer some of the questions:

Using AS3. Noticed this "temp-drop" effect about 16 hours after mounting the block.

With the 2680MHz speed, I was doing that at 2.20v with the copper block. Regardless of voltage increase past this point, the CPU would not get stable above that speed. With the silver base, the CPU is stable at 2720MHz at 2.20v. 2730MHz at 2.22v, and 2737MHz at 2.25v. Pushing beyond that voltage achieves nothing for stability at at the higher 2748MHz speed, and in some sense, stability is actually reversed with the higher voltages (>2.25v), so clearly the extra heat from the extra voltage is overwhelming the ability to keep the CPU "hot spots" cool.

Mind you, the voltages I'm saying above are just the BIOS settings. The actual voltages are more like 0.10v lower, so not quite as freakishly dangerous as they sound.

Actually since yesterday, my 3yo daughter walked in and bumped the hosing quite badly. The temperatures after that climbed back to what they were before, but the overclocking potential still remains, which is strange. Disrupted thermal contact for only a portion of the die?

Clearly this seems to be a rather "sensitive" event, and perhaps something that shouldn't be relied upon. I'm personally chalking it up to one of those "twilight zone" moments until I run some more tests and gather more evidence over a longer period of time. I was just pretty excited and confused at the same time over what I was seeing.

[sevisehda]

If you play with the theory that the soft silver has deformed to match the die then any jerk would most likely undo it. The changes in the silver are VERY small, it would be impossible to remount it even on the same CPU because you would never get it in the exact same spot again. Your daugter may have hit it hard enough to unseat the block just enough to undo the deformation.

I'm not sure how soft silver is and even how much heat makes it softer. But would it be possible to not use a TIM, then like some TIMs burn it in for a while to allow it to deform.

[flyingass]

billa's monochromatic interference-based surface viewer might be useful in determining levels of deformation.

my initial impression is fairly skeptical - i rather doubt this sort of result would be regularly reproducible. if not, however... there was another thread here discussing annealing, and i believe annealed copper is both softer and slightly more conductive than hard worked - which could provide even more of an improvement in a well-designed block.


you might want to research the effect of annealing on hard worked silver, though good information is difficult to find. the other thread was focusing mainly on copper, and a 10% improvement in conductivity was speculated....

[myv65]

Cathar,

This is why people with inate ability go into mechanical engineering and guys with square heads and those little propeller-twirly hats become electrical engineers. I sincerely doubt you have deformed the silver appreciably, certainly not to the tune of 3.5°C. Alas this is likely the realm of itty bitty quantum things orders of magnitude beneath what can be seen and directly observed. You'll probably have as much luck correlating this with the ocean tides as anything.

The obvious course of action is to try duplicating the results. Mount and dismount that silver block dozens of times and look for patterns to emerge.

I place my bet on it being a fluke, and I'm not referring to the multimeter of choice.

[Since87]

Originally posted by myv65
This is why people with inate ability go into mechanical engineering and guys with square heads and those little propeller-twirly hats become electrical engineers.

Yeah, yeah, yeah. Mechanical engineering just doesn't take enough mental horsepower to require active brain cooling the way EE does.

[Cathar]

Don't know what's going on. Let it sit overnight and it's dropped again.

Maybe the thermistor on this board is on its way out? Maybe it was just the thermal paste settling in? Maybe a little bit of column a) and a little bit of column b)?

[murray13]

Doesn't AS3 have a period of time before it reaches it's best?

With this statement from their site:

During the CPU's initial use, the compound thins out to enhance the filling of the microscopic valleys and ensure the best physical contact between the heatsink and the CPU core. Then the compound thickens slightly over the next 50 to 200 hours of use to its final consistency designed for long-term stability.

Don't know how much affect this had on your measurements but it might be a cause. With your blocks performing so well a small change in the TIM would be more apparent than with a lower performing block.

[msv]

Originally posted by Cathar
Just to answer some of the questions:

Using AS3. Noticed this "temp-drop" effect about 16 hours after mounting the block.

Noob tip, just repeating the obvious (I *know* You already have considered this several times): As far as I´m told there´s no point in reading temps the first 48 hours with AS as compound. It needs a couple of days to mature. Temps are supposed to drop during this time. Though 3.5 c is *a lot* of temp drop.
regards
Mikael S.

[Cathar]

Yes, I'm well aware and familiar with the AS curing thing. I've never, ever seen a 3.5C temp drop in 16 hours though.

As some people suggested, it perhaps is just a fluke/freak occurrence with the AS3. I'll pull off and re-install a couple of times and see if it happens again.

[Blackeagle]

While the AS3 settling down helps, the burn in effect in the CPU of the higher mhz could real easy add more. The silver block perhaps just helped the CPU past a "hump" in it's ablility to O/C to the extremest levels. These seem by far the most likly to me in combination.

A very close examination of the base if you remount it will answer the deformation theory.

Or perhaps as noted above you should check on the state of the tides?

I'd suggest a posting to VR's list of uber O/C results is in order. Should be enough to push you up past a number of folks.

[Blackeagle]

I'd suggest a remount of the copper Cascade you've had in use could also be helpfull.

If the CPU has burned in to O/Cing at a higher level than in the past the copper block will now scale higher as well. If it however doesn't............

Will be real interesting to see how this goes.

[Ewan]

Cathar. How much pressure dare you attach your waterblock to the CPU?

I expect that the higher the pressure the more initial cooling. However high pressure will possibly make the curing process take longer since the goo can't flow from one place to another as easily.

[Cathar]

I apply about 20kgs of pressure to my CPUs.

[Boli]

When you do remove the block I would have a serious look at the block interface with the CPU die... if it was perfectly flat then ruinnning your fingers over it SHOULD detect any "inperfections".

I recon the block has been moulded slightly. Though this is just a hunch remembering that poeple used to warm the metal up in their hands before making jewelry from it and the blobk will get to a much higher temperature... and more importantly a higher preasure.

Speaking along those lines... with a softer metal the force of the water on those impingment cups could have worn them to a slighly more efficient shape. (A Teardrop sort of one? *shrugs*)

Anyway you look at it though a VERY close look at the entire block is worth it. If it seems identical then its the AC3 working extra well with a silver block than a copper.

~ Boli