Made my own waterblock!

[MadDogMe]

I don't understand why people think that a thicker base will spread the heat quicker. This is backwards. A thinner base will spread out heat faster than a thicker base.

Sorry I was labouring under the impression that lateral heat spread is improved with a thicker(than 1mm) base, I thought it'd allow more heat 'units' to move quicker(sideways!) to a larger surface area(wider channel) allowing more convection area, allowing a lower C/W. funnily enough I seem to remember reading this off one of Les's charts :shrug: , but anyway my post was'nt to make a statement for or against (sorry if I did) but more to say experiment!. With a 3mm BP you could cross cut it with a juinor hacksaw and have fins or pins inside the block 1.5mm deep...

In any case, moving heat through copper is exactly what I *don't* want. I want to move the heat to the water.

Picked a bad medium for a block then did'nt you? ...

The scores were to be as deep as you can make them(0.5mm?,1mm?. It's something I keep meaning to look into) to act as surface area channels(like I said I like your block but think it could 'always' do with more surface area), not just scratches to cause turbulance...

I'd like to see a centre inlet design if you could bend your head around mounting it?...

[mikildemion]

Originally posted by Graystar

However, I may be misunderstanding again. Are you suggesting that the water be pumped through the holes in the middle layer? I ask because I was thinking that the top and bottom layers would be fed in the same way that it works now. If you're thinking of pumping the water through the holes, then, I don't like the idea, mostly because you've now created a place to trap air (the original design was also meant to pump air through as well.) Also, in that configuration there doesn't appear to be any real "route" for the water. I mean...yes, the water will go in and come out, but you don't get a sense that every water molecule is on a definite track getting in and out. You might have some hot spots.

Of course, trying it is the only real way to tell. Go for it!

You are not misunderstanding again. That is what I was asking if was good idea or not. And with your explination it looks like it would be a very bad idea. But I may just have to try it to see how bad is bad.

Appreciate the feed back.

MD

[Tweety]

I'm back again...
I had a little thinking session yesterday at work...(I had nothing more interesting to do...)

The Pro's of this block:
The small size
Universally usable, it will fit ANY motherboard, without exceptions
Easy to make, low cost, little material/special tools

The Con's of this block:
The block not touching the pads make me nervous...
Laminar flow(fixed with grooves)
The fact that I didn't come up with the idea...

And yes... I didn't add lack of suface area to the above list... why?
I don't think it lacks any surface area, again why?
Well... Greystar used a method wich is to me, more accurate then any thermometer, he said that the block felt cool when he touched it...
With such a small block, the chance of the topside being cool if the block didn't actually do the job it should is slim to none, the job here being to transfer the heat away from the cpu fast/faster than enough to keep the temps down. If the setup has been running a while (20-30 minutes) to get the temps up to normal running conditions, the temps are stable and the block still feels cool, then I don't think a thicker base plate will affect the performance very much...
The only thing it will do is add more metall to be heated, it won't add to cooling performance...

Laminar flow on the other hand can and probably is a factor holding back performance... easily combatted with grooves or any other measure to add a small amount of turbulence...

As for your idea to make the channel wider Graystar...
Well it migth add to performance, IF the with of the channel is the bottleneck, since I don't have the exact figures I can't tell...
The channel migth actually have the same amount of flow rigth now as the inlets, or smaller/larger... the ideal here would be to have the same flow in both the in/outlets and the channel... If the inlets have a lower flowrate than the channel, making the channel wider, and increasing the flow you will only add to the laminar flow, and make the water "stagnate" in the block heating up...

Laminar flow means that the water closest to a flat suface slows down, heats up and this creates a "tunnel" in the middle of the waterflow with cooler water escaping the block faster, leaving the hot water behind as an insulator... just to explain the word...

Now... I'm gonna give a suggestion going against one of my previous arguments... How about making the bottom plate squared? big enough to rest on the pads, but keep the other parts of the block the same... It will add metall to be heated, but the question is how much will this reduce performance? Is it worth it in terms of security?

BTW the clip, is it a universal heatsink clip? It looks like one in the pics but I'm not sure? Did you have to modify it? bending it to reduce/add pressure?

[Graystar]

Originally posted by MadDogMe
Picked a bad medium for a block then did'nt you? ...

Ha! Wiseguy!

Originally posted by MadDogMe
The scores were to be as deep as you can make them(0.5mm?,1mm?. It's something I keep meaning to look into) to act as surface area channels(like I said I like your block but think it could 'always' do with more surface area), not just scratches to cause turbulance...

I'd like to see a centre inlet design if you could bend your head around mounting it?...

The scores were to be just scratches initially. I want to trip up the water, but it don't want water falling into channels. That will just create hot spots. I want the water in and out fast. So I'm still a little undecided on whether I should make the scoring any deeper than merely surface imperfections.

I've thought about it and thought about it and could not come up with an easy mounting system for a central outlet based block.

[Graystar]

Originally posted by Tweety
Now... I'm gonna give a suggestion going against one of my previous arguments... How about making the bottom plate squared? big enough to rest on the pads, but keep the other parts of the block the same... It will add metall to be heated, but the question is how much will this reduce performance? Is it worth it in terms of security?

BTW the clip, is it a universal heatsink clip? It looks like one in the pics but I'm not sure? Did you have to modify it? bending it to reduce/add pressure?

I don't think a large base would reduce performance. However, I do think it is unnecessary. I wish you guys could come here and grab one of the outlets and tug on it. It simply isn't going anywhere. This is the nature of the single point mount. All the force that is holding down the block needs to be overcomed before the block can actually tilt.

Personally, I think chipping dies is an outdated problem that was brought about by thermistors that were sticking up through the center of the socket. I don't think people were pushing down on their CPU before locking them in place. Then, the pressure of the HSF forced the center region of the processor to slip deeper into the socket, and CRACK!

I mean really, those four little pads are like sponges. They compress. They're not gonna stop anything from tilting significantly. So....I don't worry about it.

The clip came from an SK6 heatsink. You can also get the clip from the least expensive Dynatron heatsink. It's something like 6 bucks from Newegg. I only had to add two more pieces of my copper stock to the center of the block in order to get good pressure. I think I can easily add a third piece to get a little better tension. I wish I could get just the clip from someplace.

http://www.newegg.com/app/viewProduc...114-005&depa=1

[JFettig]

If you dont mind, I think Ill make you a base too, how wide is the channel area in there? I think Ill make it with a 1/8inch end mill and leave a fin in the center, how does that sound?

how thick is the thickest peice you can use with that clip? this would end up being 1/4inch+ your top.

I would make the channels 1/8inch deep and leave a 1/8inch base, then probably leave it up to you to lap it


Jon

[Graystar]

Originally posted by JFettig
If you dont mind, I think Ill make you a base too, how wide is the channel area in there? I think Ill make it with a 1/8inch end mill and leave a fin in the center, how does that sound?

how thick is the thickest peice you can use with that clip? this would end up being 1/4inch+ your top.

I would make the channels 1/8inch deep and leave a 1/8inch base, then probably leave it up to you to lap it

Sure thing! That would be cool.

If you don't mind, lets talk in decimal, since my copper is .043".

I've got three layers of copper, and I add another 2 layers for clip pressure. I can go to three additional layers I think. So that's .258" max thickness of the entire block. That gives you .172" as the max base thickness. Since 1/4" is .25, it won't work unless I get a different clip. However, since you are cutting your channels into the block, I think my middle layer would be unnecesary. So that would give you another .043, giving you .215 max. Hmmm.... .25 would be almost like adding another layer. I don't know if I can clip it with a block that high.

So you'll have to make the base, 1 1/4" by 5/8" x .215". The tubing is 3/8" OD. Each tube is positioned 1/8" from the three closes edges. This should tell you where you need to start and end your channels.

I'm assuming that you mean channels running from one outlet to the other. Personally, I think channels are not all they're cracked up to be. What follows are my own thoughts by just sitting and thinking...I don't have any actual experience with a channelled design.

The idea of the channel is to increase the surface area. However, there are two problems introduced when you increase surface area with a channel. First, the area that you added is above, not at the same level as, the base. So heat has travel further to get to where it can be transferred to water. This reduces efficiency. So the gain from channels is a balance between increase surface area and lost efficiency, not simply an absolute increase due to surface area. Second, when you have channels, you create drag, which slows down the water. Slower water doesn't cool as well, so again, you've reduced performance.

So the gain from an increase in surface area must overcome the losses from decreases in efficiency caused by greater conduction and slower flow. Also, you have to consider design very carefully with fins. You must insure that every water molecule has a definite path in and out of the channels, and the pressure to move it. Otherwise, you'll end up with water sitting in the channel. That is why I suggested getting rid of my middle layer if you are going to cut channels. All the water would move through my wider top channel, because there's less resistance, and water would just sit in the lower channels.

I think with 1/8" deep channels you're going to have a similar problem *unless* the entry point of the water is at the base of the channel. That will put pressure where it needs to be to move the water. If you add such a base to my design, you'll only move the water at the top of the channels.

So in addition to cutting channels, you will also need to drill 3/8" blind holes at each end of the channel to accept my tubing. Otherwise, I'm pretty sure we're just going to get hotspots from bad flow.

[JFettig]

graystar, you got a LOT to learn just gonna tell ya that much

and 3/8 blind holes? dont get what ya mean?

[Graystar]

Originally posted by JFettig
graystar, you got a LOT to learn just gonna tell ya that much

Could be...we'll see when I get a rad and pump on my block.

Originally posted by JFettig
and 3/8 blind holes? dont get what ya mean?

Blind holes and holes that stop in the middle of the work piece. I would suggest a blind hole, matching the depth of your channels, so that 3/8" tubing can extend downward. This means that water is introduced to the channels from the channel ends. If you simply place the tube on top of the channels then the water is introduced from the top, forced down into the channels, and then tries to flow across the block. We can try that if you want, but I don't think it will work as well.

[MadDogMe]

The idea of the channel is to increase the surface area. However, there are two problems introduced when you increase surface area with a channel. First, the area that you added is above, not at the same level as, the base. So heat has travel further to get to where it can be transferred to water. This reduces efficiency. So the gain from channels is a balance between increase surface area and lost efficiency, not simply an absolute increase due to surface area. Second, when you have channels, you create drag, which slows down the water. Slower water doesn't cool as well, so again, you've reduced performance.

I think you're overly hung up on the max efficiency thing GS, it's more important to remove more units of heat than remove it at a certain temp. The extra distance heat has to travel is tiny and is greatly offset by having double the surface area(assuming a square shaped channel, more if deeper) for convection to happen...

The drag is neglegable, what difference is there between it and that caused by cross scored 'turbulance' marks?...

PS, have you measured the cross~sectional area of your blocks water channel?, how does it compare to the barb sizes or tube inner diametre?. If it's bigger you'll get a pressure drop, if smaller you'll cause resistance but boost velocity, have you thought about tweaking it to match your pumps power?...

[Graystar]

Originally posted by MadDogMe
I think you're overly hung up on the max efficiency thing GS, it's more important to remove more units of heat than remove it at a certain temp. The extra distance heat has to travel is tiny and is greatly offset by having double the surface area(assuming a square shaped channel, more if deeper) for convection to happen...

You say "greatly offset" but I don't see the numbers demonstrating that. It could very well be. I tend not to believe anything until I see the numbers.

Originally posted by MadDogMe The drag is neglegable, what difference is there between it and that caused by cross scored 'turbulance' marks?...[/b]

The difference is that the drag is on a laminar flow pattern. It cannot be compared to a turbulent flow pattern.

Originally posted by MadDogMe PS, have you measured the cross~sectional area of your blocks water channel?, how does it compare to the barb sizes or tube inner diametre?. If it's bigger you'll get a pressure drop, if smaller you'll cause resistance but boost velocity, have you thought about tweaking it to match your pumps power?... [/b]

The tubing is .375" OD with a .035" wall. The channel is .375" x .043". The area of the tube works out to .07306 sq. in. and the area of the channel is .016125 sq. in. So the area of the channel is only 22% of the area of the tubing, or a ratio of 4.53 to 1.

What exactly does it mean to tweak it to match the pump power? How do you do that? Is there a formula?

[bigben2k]

About designing it for the pump, MadDogMe is referring to this: get the PQ curve for the pump. You will likely find a curved line. Draw a line at the axis intersects, and see which point on the curve is furthest away from your line: that's your pump's point of maximum efficiency.

Then, design your block to operate with those parameters. If you use a heatercore, you have to account for it too, and I can tell you right now, the result/parameter is actually different.

[Graystar]

I still don't understand. Do you have an example?

[bigben2k]

Ok.

Here are the common Eheim curves.

In red, I drew the line I was talking about, along with the perpendicular. Where that line intersects the curve, is where the pump is most efficient, because that's where it provides the most amount of flow for the pressure.

In green is where you measure it all.

For a 1046, it is most efficient at 3 lpm, with 0.75m head.
For a 1048, it is most efficient at 6 lpm, with 1m head.
For a 1250, it is most efficient at 11.5 lpm, with 1.3m head.

Now if you're choosing a pump, you do this exercise in reverse: figure out the flow/pressure drop that you need/want, and select a pump with the above info, so that it's most efficient at it.

Note that most people achieve 50 gph (~4 lpm) in their rigs, so I can tell you with confidence that anyone using an Eheim 1250 isn't running it efficiently.

[bigben2k]

Here's (one of two possible) curve for my Little giant 2-MDQ-SC.

Most efficient at 350 gph, 8.5 feet of head, aka 6 gpm (~24 lpm) and a bit less than 3 meters.

[Graystar]

Oh, I understand what you're getting at now.

Such a calculation is not critical if you have a flow-bypass. You just need to make sure the pump has sufficient head for your application.

[bigben2k]

Originally posted by Graystar
The difference is that the drag is on a laminar flow pattern. It cannot be compared to a turbulent flow pattern.

That'c correct. Friction isn't a significant factor in laminar flow, but becomes significant with turbulent flow, i.e. reynolds # > 4'000. I don't know if you're actually going to achieve that though, you might want to run the numbers.

Originally posted by Graystar
Such a calculation is not critical if you have a flow-bypass. You just need to make sure the pump has sufficient head for your application.

I don't see how a flow-bypass makes the calculation "non-critical", but you get the idea.

If you have a heatercore, you should substract it's curve from the pump's, and re-plot from there. Finding that info is the hard part, of course but you can use BillA's "Surplus" core, from his roundup.

[Graystar]

Originally posted by bigben2k
Since you're already running it, you can do a simple test: compare your water temp to your CPU temp. If the difference is more than 10 degrees C, your design stands to be improved.

Just got some numbers on this. I bought a Radioshack Indoor/Outdoor thermometer. I put the outdoor probe into the reservoir. The readout is sitting next to the case. I also tested the water temperature with a Vicks electronic thermometer (for body temperature) Both read exactly the same.

The ambient temperature is 22.8C.
The water temperature is 31.3C.
The CPU temp is 42-43C (switching back and forth)

The difference seems to be around 11.5C difference. Is there any accounting for radiator and pump in this equation? I am using a 140gph/3' head pump and that crappy 80mm Iceberg1 radiator with 1/4" barb.

Forgot to mention....this is at 100% load, since my computer folds all the time. I have a model 680 2200+. Power ratings are 67.9W max and 61.7W typical.

My Velocity pump deal got hosed...so to speak. I bought a Danner Supreme Mag-Drive 250 and I will try to hook that up with my new heatercore. That should make a real difference.

[MadDogMe]

What exactly does it mean to tweak it to match the pump power? How do you do that? Is there a formula?

Being ghetto and not an 'argue by numbers' kinda person (I'm no expert, I give opinions I don't mean for them to come across as statements ),
I'd have the pump push the water through the tubing you'll use(right lenght) rad then block and see which inner diametre throws/jets the water the furthest out of the end of the blocks body, not the exit barb. Or something like that with a bit more thought put in . It's not perfect, it does'nt allow for back pressure but that's how I'd calculate my 'jets' for an impingment block, I just crossed it over to your scenario...



The pump and radiator most definatly are included in the equation, it'll be nice to see how your little wonder performs with a decent pump and HC. If the HC is'nt removing the heat from the water the water temp will rise and so will the CPU temp. The Pump will dictate how often the water cycles through the system and at what velocity the water and copper surfaces meet for optimum forced convection, the more the merrier...

[bigben2k]

Originally posted by Graystar
The ambient temperature is 22.8C.
The water temperature is 31.3C.
The CPU temp is 42-43C (switching back and forth)

The difference seems to be around 11.5C difference. Is there any accounting for radiator and pump in this equation? I am using a 140gph/3' head pump and that crappy 80mm Iceberg1 radiator with 1/4" barb.

Forgot to mention....this is at 100% load, since my computer folds all the time. I have a model 680 2200+. Power ratings are 67.9W max and 61.7W typical.

Someone check my calculations: I'm getting a C/W of 0.29 (or is it 0.17 ?!?). By comparison, WW is rated at 0.19@ 1gpm.

Your water temp is not "amazing": you ought to be able to bring it down closer to 25 degC, but I'm not sure if it's the rad or the airflow through it that's limiting you.

Either way, That's a nice CPU temp.

Sorry to hear about the pump deal falling through.

[Graystar]

Originally posted by bigben2k
Someone check my calculations: I'm getting a C/W of 0.29 (or is it 0.17 ?!?). By comparison, WW is rated at 0.19@ 1gpm.

I thought it was (Tcpu - Twaterout) / Watt. That's .17...no?

[Gulp35]

c/w is = (Tcpu-Tambient)/W

Hope this helps.

BB2k I got a C/W of .29 also, which for a watercooling system is not too good (I got .16 with my alpha 8045 on a o/c'd 1.4Ghz Tbird) though a pump upgrade should fix that .

I was scanning over this post and was seeing generally the thought that channels don't produce turbulence. Which infact they do (if they take up the whole chasm so that water cannot flow over them instead of through) If you want I can Email you an Excel spreadsheet that can calc the possible heat transfer for paralell(sp?) channel blocks.

[Graystar]

Originally posted by Gulp35
c/w is = (Tcpu-Tambient)/W

Hope this helps.

BB2k I got a C/W of .29 also, which for a watercooling system is not too good (I got .16 with my alpha 8045 on a o/c'd 1.4Ghz Tbird) though a pump upgrade should fix that .

This is the guiding document on C/W by Bill A:

http://www.overclockers.com/articles654/index02.asp

C/W of the system = (Die - ambient) / CPU power(watts)
C/W of the block = (Die - water inlet temp) / CPU power(watts)

I switched my routing so that the water went from the reservior to the CPU. That way, I can measure the inlet temp. Here are the latest temps from this configuration:

Ambient - 22.0C
Water inlet - 28.4C
Die temp - 40C
Load - 67.9W (my machine folds 24x7. And I'll leave out the 7W from my submersed pump.)

C/W of the system = (40-22)/67.9 = .265
C/W of the block = (40-28.4)/67.9 = .171

My system sucks. However, that was established already as we know I'm using the Ahanix Iceberg1 WC kit. I simply replaced the kit WB with my WB.

My block, however, seems to be doing great.

But, I could have misinterpreted something. Anyone see any mistakes?

BTW, it would appear that my system is working a little better with the flow going from rad to res/pump to block. Indeed, I have gained another 1C over the air-cooled processor. However, last time I tested this the temp went up by 1C.

I think the real reason is that my WB twisted around on the die while I was switching the hose.

In the machining world, very precise blocks called gauge blocks are assembled to give a precise height for machine work. The blocks are so precise that air trapped between the blocks will throw off the measurement. As such, there is a technique of assembling gauge blocks where you twist the blocks together and squeeze the air out. You know you've got it when air pressure holds the blocks together.

I think something like this happened when my block was twisted. Either air was expelled or the TIM was thinned. In any case, that's what I attribute the slightly better performance of the system to.

[Since87]

Originally posted by Graystar

But, I could have misinterpreted something. Anyone see any mistakes?

It appears that you believe that you know the power dissipation in the CPU to 0.1 Watt.

Why do you think that you know the CPU temperature with any accuracy.

[Graystar]

Originally posted by Since87
Why do you think that you know the CPU temperature with any accuracy.

Did you mean the CPU wattage?

The temperature is coming from the internal die.

The power consumption comes from AMD tech docs for my processor when at maximum load. I use the AMD CPUid utility to verify my processor model as a 680 2200+.