Storm/G7 Prototype + Piccies

[Les]

Originally Posted by bobo5195

Its was numerical analysis and they didnt record the results. Also re reading the papers there reynolds numbers are alot higher than the ones for pc cooling so its less useful for us. The fluid was air which at high reynolds numbers is going to behave like water. Our low numbers of Re (under a 1000, using what should become know as the bobo equation for reynolds number in a round pipe with flow of Q lpm; Re = 21.2*Q/D) this not the case so all im after now is lamina jets.

Not sure about that, from here one of my fumbles with Reynolds for the White Water :



Note: "Experimental Convection Coeff" values are calculated using the old C/W(TIM)=0.1

[Cathar]

Re in the Storm/G7's jets is around 400*Q, where Q is the block input flow-rate in LPM. In the Storm/G4's it is around 850 * Q.

TBH, can't say that I've noticed that the flow-regime within the jet itself is all that important with respect to the results seen. Being submerged jets the Cascade block sat the jets up high enough from the cup base so that the shearing action of the water rushing out of the cup really turbulated the flow coming out of the jet by the time it got towards the bottom of the cup, hence the "mash in a cup" effect as I eventually came to describe the Cascade's workings.

The Storm disrupts the jet flow prior to striking the base-plate and so induces turbulence artificially.

[bobo5195]

while i remember the results for increased performance were probably due ot the updraft effect of two jets. Where the two paterns colide a upwards spike of flow is created which is good for cooling. Overall the jets without this effect have lower cooling ability as they obviously cover a much smaller area and tend to interfere wiht each other if im reading this right.

Im trying to find jet pulsing stuff. Hoewer some chinese guys seemed to have some sucess with moving the jets around so the jet center is in a different location. This approach is broadly similar to the pulsing in that its trying to break up the boundary layer at the surface to aid heat transfer. It also has the side effect of increasing the cooling area under the jet flow. How to do this is slightly problematic. My crazy idea and i highlight this is crazy, would be to use the pump pressure to shift the nozzle plate around. Take two bypass loops out to the sides of the blocks and alternate them (using electronic valves) using flow momentum to shift the nozzle block around a little (obviously suspending it first). Not perfect but certainly acceptable for the high end bling. I would prefer piezo electric or some kind of water fan powered arrangement but moving arround the nozzles but it needs a good think. Im sure that some how turbulence could be used as well.

[Cathar]

Plate wobble could probably be done in a similar fashion to how nutating flow meters work...

[bobo5195]

im talking about straight pipe entry reynolds here. V beign average pipe velocity and L being diameter,so nothing too fancy.

Are you talking about turbulators in the nozzles as i thought they were just drilled holes. Im fairly sure that the nozzle length is long enough that the flow in them is fully developed lamina flow even if the entry flow is turblent. End turbulators are going to effect the flow (for good from what ive seen) but on a 0.45mm hole getting them right is going to be tricky.

cengel and boles (my heat transfer lecturer fac text book) gives lamina fully developed flow length at l= 0.05*Re*D. turbulent developed flow is approximated at l = 10*D. either way those nozzles are going to create nice pipe flow.

[bobo5195]

Originally Posted by Cathar

Plate wobble could probably be done in a similar fashion to how nutating flow meters work...

Could be good. Im going to have to re read this paper, non of the figures are included in the version i have which is annoying.

[Cathar]

Originally Posted by bobo5195

Are you talking about turbulators in the nozzles as i thought they were just drilled holes.

No, the jets are just drilled holes. The turbulators poke up from the cup base (which also has the nice benefit of added metal-water surface area). Alignment of the whole is therefore naturally of critical importance.

[Les]

Cathar's Reynolds appear correct - quick check with SF Pressure drop5.2(the old free one)
Do seem to obey(ish) Re=21.2*lpm/d where d is in metres

[bobo5195]

ah, thats some tricky machining. You sure its worth the machining accuracy for that feature? (unless of course its free thanks to the cnc machine)

Not seen any research into turbulators anywhere but if it gets gains it gets gains. Could be useful as the core of the jet as a surface length to transfer accross and it would have the effect of breaking up the boundary layer.

[bobo5195]

Originally Posted by Les

Cathar's Reynolds appear correct - quick check with SF Pressure drop5.2(the old free one)
Do seem to obey(ish) Re=21.2*lpm/d where d is in metres

It owuld as i derived it from first principals.

kinematic viscosity = 1*10-6 and work from there, subsituting V=Q/a wher a is 0.25*pi*v^2. LPm = 5/3 * 10^-5 m^3/s as well as while m is annoying m^3/s is a bit hard to comprehend in water cooling scales. Its close enough and saves me some maths here and there. Shouldnt of given it a name when its merely a unit conversion but i would like to have my own eq .

And you should be able to calculate pressure by hand its only colebrook or 64/re for the friction factor. I would be tempted in water cooling to use coffin mason as most flows exist in the transition region (for pipes at least, tried to make a calculator and went doah when most values were nearer transition). Theres always playing with the reynolds number (using maximum instead of mean average velocity).

[bobo5195]

have a little look around about pulsed jets. The main cases available are air jets but for either case it was found that pulsing created large eddies whihc improved turblence of the jet. However this turbulence resulted in a lowering of overall heat transfer performance. At the center of the jet (distance from the center of the jet) heat transfer was lost signficantly however further away from the jet more heat was transfered due to higher turbulence.

At jet heights over 15d you lose any beinfits. The highest benfical effects would occur at distances of 3d as here the flow is axisymetric.

One interesting point is that pulsed jets show much lower pressure drop. A jet amplitude of 5% (flow 100% to 95% of max flow rate) created a lower pressure drop of 7%

[mwolfman]

Just a reflection, when I was studding CE, we did some of calculations on the air intake… The sound wave can be used as an turbo (and for 2-strockers the exit is extremely important), but it can also do the opposite… it should work here as well…

(I think its Lockheed SR-71A Blackbird that gets (or should I say got) 68% of its propulsions from the sound wave…)

[GlassMan]

Originally Posted by mwolfman

The reason why I ask is due to the fact that my Sapphire card sounds like a electric grass trimmer...
I can’t say that DDR3 is cool, I have active cooling (air), and they definitely wouldn’t survive without that... (45-50° after 20 min in DoW)...

They (Watercool.de) use modules, one cooler but many diffrent configurations for diffrent cards...

By ignored, I meant as a water cooling problem. My thinking was that if Cathar was going to develope a GPU cooler, he should concentrate on the GPU, and it's mounting problems and not add in all the additional mounting problems of memory.
The DD nVidia type solution was what I was thinking should be avoided.
I can't comment on your experience without the details, but there are many things I would try before I tried the crab solution you refered to.

[bobo5195]

Originally Posted by GlassMan

By ignored, I meant as a water cooling problem. My thinking was that if Cathar was going to develope a GPU cooler, he should concentrate on the GPU, and it's mounting problems and not add in all the additional mounting problems of memory.
The DD nVidia type solution was what I was thinking should be avoided.
I can't comment on your experience without the details, but there are many things I would try before I tried the crab solution you refered to.

problem with gpus is that they are increbily non standard. Half the enigneering difficulty is designing somethign that can work on everything.

In order to bring cost down you could do build to order (ie set the cnc do a custom job based on hole sizes and die size) but that s quite complicated to make an automatic system.

GPu are also easier to cool as they are more accepting of a high temprature differential. However space constraints mean that its hard to get a good wc unit onto the card. If your for a two slot solution why not just use and air cooler which is just as acceptable.

Also some kind of jet impingment design is hard to do as there is a difficult mounting angle to place the incoming water and your going to have to use small tubing to keep things tidy and within size.

[Cathar]

The non-standard nature of GPU's means making blocks for them is difficult, especially so since one often has to buy a $500 card to figure out whether or not the waterblock is going to fit.

It would be nice if ATI and nVidia started acting a bit more like AMD and Intel, and released technical white-papers that documented the clearance zones and mounting points and mechanisms for their video cards, but they don't. Quite often its a case of make a block and hope it fits. GPU-only blocks are easy enough to do but almost all mounting mechanisms are fiddly and require a significant amount of setup overhead to start stamping out the customised plates and pieces required for mounting.

This has been the real reason why I've held back on GPU blocks for so long. Got no problem making a very high performance low-profile block, I just know that unless I throw a few thousand dollars at the mounting issue for each card type that people are always going to complain that the blocks suck for purposes of mounting, and mounting inconsistency is the place where everything can go pear-shaped.

Making GPU waterblocks, seem to me, to be less about the actual performance and more about getting the mounting right, and the problem is with the non-standard nature of various cards and component clearances, even on so-called reference cards between different models even though the mount holes may remain in the same place, and it quickly becomes a rather expensive exercise for the contracting hobbyist.

Have a number of GPU prototypes here, but the stumbling block has always been with the mounting consistency. I'll spend some more time on that issue soon though, but so far its been enough to put a GPU block release on the back-burner indefinitely.

[GlassMan]

The non-standard nature of GPU's means making blocks for them is difficult, especially so since one often has to buy a $500 card to figure out whether or not the waterblock is going to fit.

I hear that, and of course that makes it difficult for people like me to know how varied the mountings are. Thanks for letting us know the score.

[Les]

Returning to h(eff)
Have 2 methods for determining the Heat Tramsfer Coeeficient(h(conv)): (1) Sieder-Tate for channels(Kryotherm) (2) Flomerics for impingement.
Use both for the WW then convert to h(eff)**
Presented, in earlier post, the "Sieder-Tate model" which concurs with your postulate.
However the "Flomerics model" predictions.are still a better fit to the only data available - Billa.
Originaly the predictions were matched to the exerimental by adding C/W(TIM)=0.1c/w. However the sensor offset of ~0.05c/w was overlooked. The upshot is that with the accepted C/W(TIM)=0.06c/w and ~0.05c/w offset the "Flomerics model" is the better fit.
With the at-time-evidence pointing to bp<1mm being best at all flow rates the "Flomerics model" was and is still the preferred

Have attached the "Flomerics model" predictions.This suggests that at ~6watt (~ 10lpm P/Q link)
the h(eff) is ~ 90-100kw/m^2c and not 60-70kw/m^2c.
With no10x10mm die test data and no model(to connect h(conv) and h(eff)) for the Storm or Cascade think your suggested h(eff) values are a big Dunno.

** I use Kryotherm to convert h(conv) to h(eff) ...............h(eff) is the effective heat transfer coefficient acting on the finned surface of the bp(as defined here)
Then use Waterloo to convert h(eff) to Thermal Resistance(rather loosely designated C/W).

EDIT: Corrected Flomerics Prediction graph and edited text accordingly (in magenta)
Was using a "memory version" for (heff)
Found my Excel for predicted h(eff) of "Production WW"

[bobo5195]

There are models other than sieder tate, cant remember what they are called (huaman?? comes to mind) besides im not sure they are that helpful in this case. The heat transfer of impinged jets is hard science. Also they are exceptionally efficent and at the surface where the jet is focused they are nearly ideal, so you have to consider the whole surface rather than a point. A simple model is not going to predict to much. At the end of the day simple heat capacity minus 'losses' (thermal paste etc, might be an idea to include minimum thickness of al or cu to give a nice spread of heat on the surface as this is mostly unavoidable) this gives you a nice k value. Ie if k=1 then there is perfect heat trasfer. There are special case where k can be lower than 1 if you use pump power to cool the fluid by making it go through a sudden expanssion but using k values gives you a nice graph. The surface is not macroscaled finned as well so any finned model is an approximation to a different type.

Everything is relative after all, use of a peltier to increase the gradient (but only to just above dew point) will give substantial heat gains. Even assuming performance is the only goal is slightly pointless, as there are many factors which effect overall valuation.

[bobo5195]

To top up another search of the mech eng archives, i found this but cant get hold of the full text.

"
Optimized heat transfer for high power electronic cooling using arrays of microjets
Fabbri, Matteo (Mechanical and Aerospace Engineering Department, Henry Samuely School of Engineering and Applied Science, University of California Los Angeles); Dhir, Vijay K. Source: Journal of Heat Transfer, v 127, n 7, July, 2005, p 760-769
ISSN: 0022-1481 CODEN: JHTRAO
Publisher: American Society of Mechanical Engineers

Abstract: Electronic cooling has become a subject of interest in recent years due to the rapidly decreasing size of microchips while increasing the amount of heat flux that they must dissipate. Conventional forced air cooling techniques cannot satisfy the cooling requirements and new methods have to be sought. Jet cooling has been used in other industrial fields and has demonstrated the capability of sustaining high heat transfer rates. In this work the heat transfer under arrays of microjets is investigated. Ten different arrays have been tested using deionized water and FC40 as test fluids. The jet diameters employed ranged between 69 and 250 μm and the jet Reynolds number varied from 73 to 3813. A maximum surface heat flux of 310 W/cm2 was achieved using water jets of 173.6 μm diameter and 3 mm spacing, impinging at 12.5 m /s on a circular 19.3 mm diameter copper surface. The impinging water temperature was 23.1°C and the surface temperature was 73.9°C. The heat transfer results, consistent with those reported in the literature, have been correlated using only three independent dimensionless parameters. With the use of the correlation developed, an optimal configuration of the main geometrical parameters can be established once the cooling requirements of the electronic component are specified. Copyright © 2005 by ASME. (11 refs.) "

Also in a side note i found a paper which suggest that increasing surface roughness by the use of turbulators is a bad idea and that the jet impingment surface should be smooth and a whole lot other stuff.

[bigben2k]

Originally Posted by bobo5195

...Ie if k=1 then there is perfect heat trasfer. There are special case where k can be lower than 1 if you use pump power to cool the fluid by making it go through a sudden expanssion ...

...and this applies to liquids as well as gases?!?

Originally Posted by bobo5195

...
Also in a side note i found a paper which suggest that increasing surface roughness by the use of turbulators is a bad idea and that the jet impingment surface should be smooth and a whole lot other stuff.

Well there you go; you can't believe everything you read.

The jet against a flat plate has been tried, and it's nowhere near the performance range of the Storm block.

When I originally came up with the concept (here), I was looking for a way to minimize what we're now going to refer to as the "conductive boundary layer" or the "conductive part of the boundary layer". But in the same effort, I was trying to figure out a fin pattern that would work with it, because for everything I've read, It was only possible to improve on a flat plate by using fins.

Maybe this warrants more experiments, but I really believe that the cups, which form an intricate fin pattern are an essential element. Otherwise I agree that turbulators within the jets are a waste; one wouldn't want to do anything to impede the flow within the jet nozzle.

[Cathar]

In my reading of various papers, those that achieve k~=1 factors are doing so at flow rates so low that the thermal capacity of the water becomes a major limitation. I've pretty much always considered pursuit of "k=1" as a bit of a misdirected quest for the holy grail as in "Great! k=1 but now the water's boiling!"

A maximum surface heat flux of 310 W/cm2 was achieved using water jets of 173.6 μm diameter and 3 mm spacing, impinging at 12.5 m /s on a circular 19.3 mm diameter copper surface. The impinging water temperature was 23.1°C and the surface temperature was 73.9°C.

So if we calculate that, the average convectional rate is ~61000 W/m²K. That's nice but that's also against a totally flat surface so h(conv) == h(eff). Compare that to even my conservative (by Les's statements above) estimate of h(eff) as being ~105000 W/m²K and 61000 W/m²K isn't that spectacular. Sure, k is going to be a heck of a lot closer to 1 in the micro-jet implementation, but that's rather pointless if it can't deliver a better h(conv) than ~61000 W/m²K.

I'm all about lowering temperatures as far as possible, and doing so without the theoretical constraints that some of these guys place over themselves of achieving close to k=1 efficiencies. Once I see research papers start to discuss k efficiencies I take that as a pretty good sign that they're on the wrong track for purposes of maximising cooling performance. It's a false grail IMO.

[Les]

Uncertain which "k" Bobo is talking about - maybe simply "Absorbed Watts"/"Source Watts"
Suspect Cathar is referring to ε =The effectiveness of a cold plate or heat sink is the ratio of heat transferred to the ideal case in which all fluid achieves the surface temperature:(link .Equivalent to e in the Definitive Nomenclature list*
When Thermal Effectivenesses εs and "C/W"s are considered in relation to flow rate(link) there are situations where low flow may be used to advantage : Peltier chiller(link) or external rad(probably best with rad design with low Rwet at low water flow - high water velocity rad)
Playing with the sums for internal radiator.- difference between dT(die-water) being effected by LMTD(wb), and T(water wb in) being effected TD(rad).
Sums are a can of worms and most probably up the wrong tree - heat balance may simply be wrong.

*BTW Definitive Nomenclature for rads are a beast for - so far identified 48 (link

[Cathar]

Indeed Les, that is what I meant, and that is also what I understood bobo to be referring to, where "k" is the surface temperature delta divided by the water temperature delta, with delta being relative to the entry temperature of the water.

I always considered such an interesting curiosity for purposes of analysis, but by no means the be-all/end-all of assessing cooling effectiveness. I see it being used most often in micro-channel/micro-jet research papers where they're playing with pitiful flow rates and so achieving a "perfect" transferrence ratio is therefore of critical importance.

[bobo5195]

I should never post when im that drunk/tired.

Anyway quickly..

I was talking about effectiveenss (noncap epsilion), but couldnt remember the name. I was trying to get over how effectiveness would be more helpful in showing performance as a) it gives a baseline of possible performance b) shows how a block performs over a wide range, particularly for low flow where temprature graphs become asymtotic.

Oh and while i remember i found a paper by one of my lectures who is trying to use turbulators to improve cooling performance . hes interested in cooling hot air with a cold plate but some of the same principals are going to apply. He used dimples (hemisperical extrusions) instead of machined more brick like surfaces.

[Les]

Originally Posted by bobo5195

I should never post when im that drunk/tired.

Anyway quickly..

I was talking about effectiveenss (noncap epsilion), but couldnt remember the name. I was trying to get over how effectiveness would be more helpful in showing performance as a) it gives a baseline of possible performance b) shows how a block performs over a wide range, particularly for low flow where temprature graphs become asymtotic.

Oh and while i remember i found a paper by one of my lectures who is trying to use turbulators to improve cooling performance . hes interested in cooling hot air with a cold plate but some of the same principals are going to apply. He used dimples (hemisperical extrusions) instead of machined more brick like surfaces.

Complete crap.
Focus on Cathar's postulations and apply any privileged(University Library Passwords) to constructive/destructive analysis