Like Bill says, it's a bit of a mystery even to me. I can plug values in for h based upon observations, and reckon I might have an idea, but h is not something that I'm mathematically determining. Basically it's a process of "do some theory", build a block, measure it, revise the theory, make changes, and repeat.

I'm fairly confident that for lower flow rates (<4LPM) that the areas immediately under the jets where stagnation occurs are the hottest sections. At higher flow rates I believe that behavior slowly changes, but that perhaps the ratio of jet diameter to bp thickness is still perhaps a little too low. I also believe that there's also a small issue with the triangle cross-over between 3 adjacent cups at lower flow-rates, which is another thing that's changed on the XXX.

I think the main issue here is that h is varying very widely depending on both the flow-rate, and the actual location within any one cell, which of course is fairly obvious given jet theory, but I guess what I think I didn't anticipate is the magnitude of the effect that it is having on CPU overclock stability, which has caused me to re-think my strategy on ultra-thin (<1mm) bp's.

Some piccies of the first-cut XXX prototype next to a regular Cascade. Note, neither block has been cleaned up after machining so they look a little messy.

http://www.employees.org/~slf/images/xxx/xxx-3.jpg

http://www.employees.org/~slf/images/xxx/xxx-1.jpg

http://www.employees.org/~slf/images/xxx/xxx-2.jpg

It impossible to get a value of " h based upon observations" with no true DeltaT or W and an assumed C/W(TIM).

You can determine where h is roughly at on an average. Is it exact? No. I didn't even begin to claim that it was. 'tis all theory and approximation until nailed down by someone with substantially more research funds and time than I have.

whew
gonna need a filter on that puppy
the Cascade I tested was filthy when I borrowed it, and I'll be damned if it was not somewhat dirty after I finished - and I have a 20 micron filter
-> high performance has high requirements

I've actually found that the Johnson/Jabsco in-line strainers to be pretty effective. I fill my loop with water that's been fed through a 10 micron filter first anyway, and so long as the loop was clean initially the blocks remain fine.

Still though, yes, "micro-structure" blocks do place a very high requirement for a clean system.

Here suggested some "possible h"s for the WW.
However with Kryotherm only dealing with channels am unable to do the same with the Cascade.
Some rough numbers would be interesting.

Rough average estimates only for h for Copper Cascade.

h =~ 80000 at 10LPM
h =~ 55000 at 5LPM
h =~ 30000 at 2LPM

Ta
Seem in keeping with Flomerics based guesses.
Wonder what TAS would make of them.

what are you guys doing ?
a nominal device average is worthless
h is location specific and different also per the conditions
this is what the gridding and nodes are all about

Point taken, but this an average within a cup.
There being 50+ cups per device, it may be a starting point.
Only this guy's opinion.

perhaps, but this touches on my heartburn with CFD for wbs
sure, one can sim a cup, test, correct, and validate
but this does not address the cups' interconnection necessary for any device calc;
nor, more importantly, does it enable 'looking' at the effects of cup changes (of any sort I suspect)

yea, perhaps better than nothing - but for me not worth my time if not predictive
Flotherm is very capable for many things, but I have reservations about its abitity to model mbs w/o extensive parametric validation
if I have to pay $28K pa per seat for Flotherm, plus a pHD to run it; and then STILL have to test 'till hell freezes over to validate, . . . its cheaper to build and test

Cathar,

No material expert here, either. In going through my book I find that the electrode potential differential from copper to silver is ~0.5V, copper to chromium (stainless surface chromium dioxide) is ~1V, and copper to aluminum is ~2V. All this is at 25°C.

From what you described of that pump, I'm reasonably confident it was a matter of contamination. Nonetheless, if contamination is likely in a typical system then I'd agree stainless ought to be avoided, especially in extremely narrow passageways.

Would this affect my system much? Metals in contact with water = brass, copper, silver. Ive not noticed any corrosion yet but I dont want 0.5v coming from the waterblocks. Also note the silver is not touching any other metal in contact with water.
Maybe this would be god for a new thread, on corrosion testing? If someone could deisgn a test for corrosion, that could be done accuratly, I do have silver, brass, copper, steel, and alu I can test with.

Cathar your xxx block pics look like they will be great performers. Could I have a XXX-SS block once you start making them (Ill supply the silver, got plenty here :D)
Also how much more jets do you think will be possible to add into that small an area?
Any plans to make GPU versions?

Will you continue making blocks once youve desiged something better for non-reseach purposes (if you ever did?) or will the old designs either get discontinued or licenced to other companys for manufacture? (like D.Tec/WW).

|kbn|. In a standard water-loop copper/brass/silver are basically galvanically unreactive with each other. The silver->brass difference is pushing it somewhat, but it is basically safe for long periods (many months -> years). At least that's what I've been able to ascertain on-line, and having run a copper/silver/brass loop totally unprotected in terms of corrosion inhibitor for about 6 months now the silver has only just started to develop a mild discoloring, which is about the sort of reaction level I would have expected.

The XXX blocks I only intend to make in silver for the bases. Am giving very serious thought to putting copper tops on them though just for strength as I don't really want to risk another repeat of a bad polycarb batch resulting in cracking tops. This would also allow the block to be thinner, as in 16mm high instead of the current 20mm.

Am giving serious thought to calling such blocks "Cascade XS", and dropping the XXX code-name moniker.

Damn, now the "I'd hit that" joke when you post pics of the finished product isn't going to be nearly as funny.

When I took out my silver gpu block it had no noticable change in colour. I had been using it for a month or two.

Cathar as silver is a lot more expensive than copper, and that the acctual cooling area in your block is ~30mm2 at most? You could do what I intend to reduce the cost some what.
My dad has made the silver into 20mm squares 8mm thick and they are accuratly made. We intend of fitting the silver into accuratly made copper blocks so that we get the advantage of silver, but without the cost. If this could be accuratly done on the scale you intend without problems of the jets/ cups not lining up, I think it would be a good idea and make silver blocks more affordable. They just wont look as shiny - silver is very easy to get a very nice shine.
Silver will perform better but would depend a lot on the deisgn to take advantage of it, I look forward to any test results from this new block in both materials.
some people might still want 100% silver bases though.
Copper tops would be nice, (none of this annodised alu crap :eek: ), but plastic may be best for your design. I recommend non brittle plastics like polyethelene instead of accrylic, never used poly carb though.

XS is a nicer name than XXX-SS :)

It will be interesting to see how this new block compares to the original with half the jets blocked off, when used on an athlon xp. Any estimate on how the pressure drop will compare to the original?

Which code are you using and how small do you want your capillaries?
 

Which finite element code are you running? ANSYS? or NASTRAN? OR ? and have you considered using self assembled or lithographically etched tubes?

I am using a concoction of my own writing. I suppose it could be classified as a finite-element analysis tool, but it's horribly primitive by modern day professional standards and has no fancy visualisation tools. Best bit though is that it is free, apart from me spending some time to code it up.

Self-assembled?

Have discussed lithographically etched tubes with someone. Biggest issue appears to be that everything needs to be squarish.

All you need is a synchrotron.....
 

No - I mean self assembled nanotubes using a mask and PMMA like the big boys.

My issue seems to be that we're talking 10-50 nanometers for the tube using self assembly and more for a mask like that used for the wafer.

Maybe I can try something at this end.

-talc

Nanometers? We're still dealing with hundreds of microns here.

Big boys? Sorry, doing what I can on a budget that amounts to around $1500 US max.

Happy to know if there's some way to do things better though.

Cather, at the risk of stating the bleedin' obvious. The "smoothing" effect of a given volume of copper must be related to it's heat capacity. Coppers heat capacity is less than that of water but not by a huge amount when stated relative to volume (4.2ish vs 3.5ish J/cm^3.°K).
That said, water IS better, by 15-20% so the more water in the vicinity the better the dampening must be. i.e. the thinner the baseplate the better. The dirty big fly in the ointment is the TIM-BP-H2O thermal resistance, which might nullify any benefit, leading to, thicker BP must be better, above a certain thickness.What that thickness is I don't know, it might be near zero. A direct die cascade would be interesting. ie no cup bases, the base of the "cup" being the surface of the CPU die.
What I am say is that I am concerned that all the tests showing that direct die cooling gives higher temperatures may be missing the point. The, lets call it "Maximum Attainable Stable Frequency/°C", might in fact be higher, thanks to the high specific heat capacity of water, even though the actual cpu temp could be higher. What the WW and than Cascade showed for me was that mass of copper (or silver) is not important for maintaining good temperatures. What is important is the way the water is delivered to the hot area. Deliver it in the same way direct to die and I can not see how it could be worse.
I am not a direct die advocate, actually I've thought it inferior because of the necessarily smaller surface area, but with impingement in the equation and your observations of overclock stability, I am beginning to wonder.

Cheers

Incoherent

I happen to have page 5 of this thread from before the crash still open. I'll repost. I hope no one minds. Unfortunately, those spiffy diagrams of cup bases and discussions of such were on page 4. :cry:

Thank God someone managed to save page 5 at least! Sad we lost page four, but perhaps we can resume the thread from here.

Great job guntledweasel! ! :D

Man!, I can hardly wait to see the results of the next prototype comparisson of the two alturnatives for advancing the Cascade XS design.

Now I have to admit I'm sort of glad I missed on the last order of SS Cascades. Now I'll be able to afford a Casscade XS instead.