|
|
Water Block Design / Construction Building your own block? Need info on designing one? Heres where to do it |
Thread Tools |
07-16-2002, 10:27 AM | #1 |
Responsible for 2%
of all the posts here. Join Date: May 2002
Location: Texas, U.S.A.
Posts: 8,302
|
Ultimate block? Theory.
I found this link while searching about thermal transfers. Thought it might come in handy.
Over at another forum, this guy came up with this design. It's not new, but I thought I'd share the pic he posted, and see if anyone had any ideas on what the ultimate WB would be like (an old theme, I know!). Last edited by bigben2k; 07-16-2002 at 10:32 AM. |
07-16-2002, 10:37 AM | #2 |
Responsible for 2%
of all the posts here. Join Date: May 2002
Location: Texas, U.S.A.
Posts: 8,302
|
I have a couple more links to share:
http://www.coolingzone.com/Content/D...Flomerics.html http://www.coolingzone.com/Content/D.../baseframe.htm and of course another conversion tool. http://www.coolingzone.com/Content/D...ott_fr_un0.htm |
07-16-2002, 11:27 AM | #3 |
Hottest Stank of them All
Join Date: Mar 2002
Location: Washington, DC area
Posts: 533
|
I thought about trying to do a block like that, but here are two considerations.
1) the flow through the channels will be very laminar = bad. 2) If I made a block like that, I'd switch the ins/outs so that they are on opposite ends. As it is, the channels closest to the ins/outs will get the highest flow.
__________________
Does a radioactive cat have 18 half lifes? --Kenny my pimpin' rig: ...previously poorly cooled... 486DX-2 66mhz @ 75mhz, 4Mb ram, shared, 256Mb hard drive Onboard VGA, watercooled, of course 16-color monitor, labtec speakers, 28kbs USRobotics modem Windows XP (think about it) Maze 3, DDen GF3 block, Eheim 1250, econoline van HC, 1/2" hose w/ 5/8" fittings Comair 172mm fan |
07-16-2002, 11:30 AM | #4 | |
Responsible for 2%
of all the posts here. Join Date: May 2002
Location: Texas, U.S.A.
Posts: 8,302
|
Quote:
Maybe we could even add twisters, like in the TC-4 block. What do you think? |
|
07-16-2002, 12:23 PM | #5 | ||
Cooling Savant
Join Date: May 2002
Location: home
Posts: 365
|
Quote:
Ah no. Not sure where you get your info, but for internal flows, the crossover occurs at a Reynold's number of about 2300. Re = rho * V * D / mu. With rho = 1000 kg/m^2, V ~ 1 m/s, D = 0.05 m, and mu = 0.001 N-s/m^2, Re ~ 50,000 which is well into the turbulent region. Laminar internal flows require extremely low velocity. Quote:
As to BigBen2k's original question, it comes down to velocity and surface area. Them's that provide high velocity (good convection coefficient) and lots of area (q = h*A*delta-T) win. With enough pressure, direct impingement puts a high velocity jet right above the core. Performance drops with a weak pump. Some low-loss designs can make great use of surface area, but don't get great velocity. The real trick is matching a block to a pump without buying more pump than you need. |
||
07-16-2002, 12:40 PM | #6 | |
Responsible for 2%
of all the posts here. Join Date: May 2002
Location: Texas, U.S.A.
Posts: 8,302
|
Quote:
So turbulent flow IS achieved in ECUPirate's design? (putting aside the first few cross-channel problem). I believe that the cross-flow design is probably the best, since it is optimized for high surface. I'm sure that with the right manifold, the coolant flow can be distributed more or less equally. So myv65, what's your suggestion for a good block? |
|
07-16-2002, 01:07 PM | #7 |
Cooling Savant
Join Date: May 2002
Location: home
Posts: 365
|
Well, first I should clarify that flow wouldn't be laminar in the block you linked with the possible exception of stagnant areas. The reason stuff would tend to stagnant (actually in all but the last couple of tubes) is that fluids don't like changing directions. More accurately, it'll seek the path of least resistance. When the cross flow pathways are of lower resistance than the "manifold" feeding them, the flow will tend to make the long loop. In the picture, the "manifold" has much lower area than the combined cross tubes. Like I said, it doesn't make a whole lot of intuitive sense, but is true nonetheless.
As to what would make a good block, that's the $64K question. It sounds like a cop-out answer, but IMHO there is no single ideal. What I really mean is that there are conflicting properties based upon your own priorities. You want dead silent cooling at stock speed/voltage? Fine, you can do that with < 10 gph and the right block. You want maximum overclocking? Fine, you can do that with > 50 gph and the right block. Obviously, there's more to it than this in that you must also select the right range of pump and radiator/fans (and I'm totalling ignoring things like pelts, etc.). My own personal bent is towards low volume, low noise cooling without regard to temperatures so long as things remain completely stable. I toy with overclocking a little, but don't get into like a lot of people do. I'll take sub-30 dBA and stock speed any day. Funny thing is that efficient block design shares many of the same qualities as efficient radiator design. In each case you are trying to balance variable thermal resistances such that the total resistance is a minimum. Reducing one resistance generally raises another. It's a lot like electrical resistors in parallel where the individual resistors must add up to a fixed sum. The lowest net resistance will occur when all resistances have the same value. Oh yeah, and in my post above I said "rho = 1000 kg/m^2" should have been "rho = 1000 kg/m^3". Sue me. Last edited by myv65; 07-16-2002 at 01:10 PM. |
07-16-2002, 01:17 PM | #8 |
Cooling Savant
Join Date: May 2002
Location: home
Posts: 365
|
All right, here's a sample of what I would consider a low-flow block. There are two pathways from inlet (right side) to outlet that are mirror images of one another. This keeps flow through each pathway balanced pretty well. The pathway dimensions restrict flow rate to a fairly low number while creating a lot of contact area between fluid and block. The metal between the pathways needs to be sized for the heat load. If there's too little metal left, only the bottom of the channel benefits. If there's too much metal left, convection suffers and becomes the bottleneck (requiring more flow to compensate).
Herein lies another rub. Balancing of resistances requires specification of both flow rate and heat dissipation. You can only optimize for a given set of conditions. Yeah, performance would be OK across some range, but only truly peaks at one particular set. |
07-16-2002, 01:17 PM | #9 |
Hottest Stank of them All
Join Date: Mar 2002
Location: Washington, DC area
Posts: 533
|
your assertion makes sense. somewhat. on the inlet side, the water would likely rush past the first couple of inlets, all things equal. But, in Ben's design, the outlet (suction side) would likely suck from the path of least resistance, i.e., the closest holes, yes? Perhaps this would even things out. Maybe my design would be worse, based on the above logic?
How about this? The center (over the core) should have the highest flow (in theory), but the difference between flow in any of the channels shouldn't be that great.
__________________
Does a radioactive cat have 18 half lifes? --Kenny my pimpin' rig: ...previously poorly cooled... 486DX-2 66mhz @ 75mhz, 4Mb ram, shared, 256Mb hard drive Onboard VGA, watercooled, of course 16-color monitor, labtec speakers, 28kbs USRobotics modem Windows XP (think about it) Maze 3, DDen GF3 block, Eheim 1250, econoline van HC, 1/2" hose w/ 5/8" fittings Comair 172mm fan |
07-16-2002, 02:07 PM | #10 |
Cooling Savant
Join Date: May 2002
Location: home
Posts: 365
|
That's sort of where the confusion comes into play. On the return side, there will be a velocity established by the cross tubes farthest from the inlet/outlet. This velocity profile makes it difficult for flow to exit the cross tubes nearer the inlet/outlet.
Here's an experiment for the really curious. Take a length of pipe and drill a few holes along its length such that the hole area totals ~ 2-3X the cross area of the pipe. Now hook a pump up to one end of the pipe and cap the other. Dunk the pipe under water and turn on the pump. More flow will enter through the holes farthest from the pump. If you repeat the same experiment with holes totalling ~ 1X the area of the pipe or less, more flow will enter nearest the pump. It's only when you consider orifice losses and the hydraulic grade profile (function of depth and fluid velocity) that this stuff starts to make sense. Oh, and your observation about your most recent picture is correct. Highest flow will occur in the center. Deviation from average depends on size/quantity of little runs versus flow and inlet/outlet size. |
07-16-2002, 02:14 PM | #11 |
Responsible for 2%
of all the posts here. Join Date: May 2002
Location: Texas, U.S.A.
Posts: 8,302
|
This requires brain power...
Ok, so what if I cross-drilled channels a bit larger over the core, since I want more flow there? I know, it depends on the thickness of the base plate. So ECUPirate's NEW revised design would still have the flow going through mostly the center cross channels, I can see that. I've been toying with this manifold idea for years, and I know that it's not easy to get even flow. But do we want even flow, given that the heat comes from the area touching the core? Again, I think that it depends on the thickness of the baseplate. ECU: I think that what myv65 is saying is that in the first pic, the flow would be concentrated at the rear channels, because the water coming in would have a tendancy to flow right past the first few cross-holes, but if the manifold was bigger, it would act differently. |
07-16-2002, 02:22 PM | #12 | |
Responsible for 2%
of all the posts here. Join Date: May 2002
Location: Texas, U.S.A.
Posts: 8,302
|
Quote:
Last edited by bigben2k; 07-16-2002 at 02:32 PM. |
|
07-16-2002, 02:31 PM | #13 |
Responsible for 2%
of all the posts here. Join Date: May 2002
Location: Texas, U.S.A.
Posts: 8,302
|
(just a quick basic fitting recap)
First off, any waterblock must fit over the CPU, and must cover the core. Then, to make sure it doesn't crush the core, it must extend to the 4 pads on the CPU, so that it stays flat, because if it tips, it will crush/damage the core. Then, to hold it down, we have a number of options: 1-extend the baseplate with notches, and clamp it to the CPU socket tabs. 2-extend the baseplate with tabs (with holes) so that the plate can be screwed into the mobo. Alternatively, extend the cover plate. 3-use a hold down that is bolted to the mobo, then applies pressure to the block from the top 4-Thermal epoxy the whole kit-n-kaboodle. (not recommended!!!) OK. back to the cooling part... Last edited by bigben2k; 07-16-2002 at 02:33 PM. |
07-16-2002, 05:35 PM | #14 |
Cooling Savant
Join Date: May 2002
Location: Ohio-State! <--Money Hussle! SUCKS
Posts: 292
|
Basically what I'm seeing is... what I've been pondering for a long time; a "heater core" type design approach for a water block. That's sorta what the pic lends me to picturing. I would say that having a larger central channel over the core would be most efficient for mass water to heat absorption. I would make some changes just for experimentation:
1)I would move the inlet/outlet to the opposing sides or 2)Like the previous pic, have the inlet and outlet in the same path way... the problem then is how much flow is going to go through the very outter holes... 3)Providing that you have a 200GPH+@1'head pump, stagent flow is not really going to be an issue in some of the holes for the fact that the pump is doing 2 jobs; it's not just pushing water, it's ALSO pulling it and with the 1'head buffer, there would really be no problem or major issue with "slow" to zero flow. 4) Maybe the oppsite would help outter flow resistance, in that you make the outter holes a slight bit larger, to compensate for them being at the very extreme edges of the block. That way you might have faster/higher flowage going to and from the middle of the block, BUT the larger holes lending to more MASS flowage and exit... Who know's... Maybe I will run into someone in the flow-mechanics lab around here and ponder the idea across to the pro's... Looks like a great performer, I'd like to see someone build a prototype...ahah Imagine that-- folks just sitting around having time and access to CNC millers to experiment and play around with copper designs... BASTARDS!
__________________
XP 1600+ AGOIA-Y (GreenLabel, UNLOCKED!@/ NF7-S 2.0/ PC 2700 DDR 10.5*172FSB(1806/ 1991 high), Vcore 1.792...@84.95Watts --86' Chevette Heater Core --1/2" Vinyl tubes --Via1300-Stealth Mod(QUIET!) ---Home made copper-tube block ---PVC small internal res |
07-17-2002, 03:47 AM | #15 |
Thermophile
Join Date: Mar 2001
Location: France
Posts: 1,221
|
CFD to the rescue
People, try a search on "CFD" in this forum (it was in the thread title).
You'll get a lot of answers. (I'm at work so i don't have time to do it for U, sorry) Ah and look at the K4.1 block (which is K4.2 now i think). It's a top performer, and has a design you'll find familiar... I think it has "stagnant flow" in some spots but somehow it doesnt hamper its performance. Anyways, designing an efficient fluid system is very counter-intuitive. Thats where CFD or great math skillz help a lot (hello Bernouilli, Navier-Stokes...). There are *so* many important parameters that solving this by hand or intuition is impossible, unless you've got the skills of Newton or Enstein. What you'll want to consider is - flow restriction vs pressure and turbulence - flow path length vs backpressure - base plate thickness vs heat spreading / resistance / capacity - channel cross section shape vs manufacturing costs - channel pathway for optimal cooling (crossflow ?) vs flow path length and restriction - inlet / outlet placement and multiple inlets or outlets ? - flow path complexity (surface area) vs manufacturability and efficiency - channel shape vs liquid properties (glycol, methanol, WW, etc...) Each of these is solvable, but not easily, even with CFD. No wonder ppl are earning a living out of this. The problem is heavily multi-dimensional , and I'm not sure there's only one solution to it. There's no such thing as "the ultimate block". One good thing to do is, build it and experiment V12: point 3 - i dont get it. You could have a nuclear turbine as a pump, you'd still get "stagnant flow" spots (ok much less, but relatively speaking) ..? |
07-17-2002, 09:16 AM | #16 |
Responsible for 2%
of all the posts here. Join Date: May 2002
Location: Texas, U.S.A.
Posts: 8,302
|
|
07-17-2002, 10:50 AM | #17 |
Responsible for 2%
of all the posts here. Join Date: May 2002
Location: Texas, U.S.A.
Posts: 8,302
|
Re: CFD to the rescue
Ok so we have:
1- flow restriction vs pressure and turbulence 2- flow path length vs backpressure 3- base plate thickness vs heat spreading / resistance / capacity 4- channel cross section shape vs manufacturing costs 5- channel pathway for optimal cooling (crossflow ?) vs flow path length and restriction 6- inlet / outlet placement and multiple inlets or outlets ? 7- flow path complexity (surface area) vs manufacturability and efficiency 8- channel shape vs liquid properties (glycol, methanol, WW, etc...) -Let's ignore #4 for now. -Let's forgo the manifold #6 for now, we can tackle it later. -I think #7 refers to a combination of #1, 2 and 4, as well as the possibility of manufacturing. Let's toss that out too, for now! -For #8, let's stick to water (for effective purposes) for now, but I'd like to include "windshield wiper fluid". We're left with: 1- flow restriction vs pressure and turbulence 2- flow path length vs backpressure 3- base plate thickness vs heat spreading / resistance / capacity 5- channel pathway for optimal cooling (crossflow ?) vs flow path length and restriction 8- channel shape vs liquid properties of water (or wiper fluid) Let's assume for cost purposes, that we're talking about copper. Unless someone believe that a silver plating would make any kind of difference. Does that include everything? Put aside for now: 4- channel cross section shape vs manufacturing costs 6- inlet / outlet placement and multiple inlets or outlets ? 7- flow path complexity (surface area) vs manufacturability and efficiency |
07-17-2002, 03:47 PM | #18 |
Cooling Savant
Join Date: Jun 2001
Location: Connecticut
Posts: 140
|
I will say from experience that I have designed and built 4 copper waterblocks all very similar to ECUPirate's diagram above. There are only 4 channels in each block but the rest of the design is similar, all channels are of equal size.
I have compared my blocks to DD Maze 2, Maze 1 and the jagged edge block(forgot who makes it). Using the identical system and testing parameters, my block performed equally or better than all of the blocks above. I don't have a lot of fancy graphs and pictures to show because I did the testing mostly for my own curiousity. That being said, a different radiator or pump combined with each block may have yielded different results. For this reason, I have always believed that having the inlet over the center of the block is generally a waste of time and effort. With the exception of the spirals, it almost always leads to a design with a lot of flow restriction. We all know that poor flow through the block is not a good thing. IMHO the best block is a straight through, one pass design with maximum flow through the center of the block. |
07-18-2002, 11:13 PM | #19 |
Hottest Stank of them All
Join Date: Mar 2002
Location: Washington, DC area
Posts: 533
|
Hi, I'm back. (went to carowinds amusement park).
I understand what you're saying, myv65. It actually makes sense. I think if you made a block like the one I drew, it would do well. I'd make it w/ 2 or 3 layers of cross holes, for maximum flow. Maybe to get more flow to the outside channels, you could add a little "island" to the inlet side of the block, like in the revised picture.
__________________
Does a radioactive cat have 18 half lifes? --Kenny my pimpin' rig: ...previously poorly cooled... 486DX-2 66mhz @ 75mhz, 4Mb ram, shared, 256Mb hard drive Onboard VGA, watercooled, of course 16-color monitor, labtec speakers, 28kbs USRobotics modem Windows XP (think about it) Maze 3, DDen GF3 block, Eheim 1250, econoline van HC, 1/2" hose w/ 5/8" fittings Comair 172mm fan |
07-19-2002, 05:39 PM | #20 |
Cooling Savant
Join Date: Feb 2002
Location: Dione, sector 4s1256
Posts: 852
|
you might want to add matter distribution.
i.e. of the average cross sectional print of the entire block, how much is copper and how much is liquid.... a good ratio I found to be about 60% liquid and 40% copper. obviously the granularity of the divisions will have an astronomical effect on efficiency as well. |
07-19-2002, 06:11 PM | #21 | |
Responsible for 2%
of all the posts here. Join Date: May 2002
Location: Texas, U.S.A.
Posts: 8,302
|
Quote:
1- flow restriction vs pressure and turbulence 2- flow path length vs backpressure 3- base plate thickness vs heat spreading / resistance / capacity 4- channel pathway for optimal cooling (crossflow ?) vs flow path length and restriction 5- channel shape vs liquid properties of water (or wiper fluid) To which I'm going to add point #6 (your point). 6-Cross sectional material print (It touches on #1, but that's ok) For baseplate (#3), i found something in the OC forum (I'll get a link later) that indicates that 300 gph effective flow reduces the difference between Alu and Cu to 1 degree of effective cooling. It was preliminary, but is worthy of mention. It's going to be critical, because it will also link to the flow rate. |
|
07-19-2002, 06:17 PM | #22 |
Cooling Savant
Join Date: Sep 2001
Location: Slovenia
Posts: 468
|
My take for ultimate block here: http://forums.procooling.com/vbb/sho...&threadid=3815
__________________
[My ftp, with lots of pics, hope to be home page someday |
07-19-2002, 06:25 PM | #23 |
Cooling Neophyte
Join Date: Jul 2002
Location: missouri city, tx
Posts: 7
|
hi, 1st ppost in this forum, reffered to by bigben2k
yeah, those cone desings seem to be a good idea, lagrish surface are, but more importantly, little resistance between the core and that surface area. wouldnt be as good for pelts though. an improvement over my design, to fux those probs yall have mentioned.
__________________
xp 1600+, still to be oc'ed, 768 mb ram, msi kt266pro2, baracuuda 4 40gig, linksis nik+cable modem, jaton gf2mx400 64 meg http://www.overclockers.com/tips835, cd+cdrw fixing to do water cooling Last edited by athlonnerd; 07-19-2002 at 06:31 PM. |
07-19-2002, 06:25 PM | #24 |
Cooling Neophyte
Join Date: Jul 2002
Location: missouri city, tx
Posts: 7
|
oops wrong pic, sorry, this one:
__________________
xp 1600+, still to be oc'ed, 768 mb ram, msi kt266pro2, baracuuda 4 40gig, linksis nik+cable modem, jaton gf2mx400 64 meg http://www.overclockers.com/tips835, cd+cdrw fixing to do water cooling |
07-19-2002, 06:46 PM | #25 |
Cooling Savant
Join Date: Jun 2001
Location: Connecticut
Posts: 140
|
If myv65 is correct, then by reducing the orefice size at the outer channels, the flow should even out. This would create less velocity in the flow furthest from the outlets and allow the flow from the first couple cross channels to escape.
|
Currently Active Users Viewing This Thread: 1 (0 members and 1 guests) | |
|
|