Custom WC setup - Removing 350W passive

[heatwave]

Provided a WC setup is connected to a very strong pump with a very high head - i.e. Iwakis, would it be possible to remove 350W of heat passively by "stringing" 2 or more good heatcores together?

Space and size is not of concern - provided size is not excessively large.

This assumes the pump can overcome the flowrate restriction of the CPU + GPU waterblock(s), pipe bends, heatcore(s), etc.

I am trying to design a scalable watercooling system that is able to cool the current hottest CPU and GPU and even the next generation models.

The WC system would still be able to run the system stable even when room temperature reaches up to 35C in the summer.

As an example: a 3.0GHz AMD XP CPU at 2.3V produces 170W (from a web XP calculator) and the hottest card today is the 6800 Ultra Extreme producing 110W at maximum draw.

I'd just like to get some qualitative comments before I hit my Heat Transfer + Perry's Chemical Engineering books and start some number crunching.

[AngryAlpaca]

At which temperature rise? An 80mm radiator can remove 350W while passive in a WC system if you're willing to go to extreme temperatures...

[Cathar]

Yes, it is possible. Build a thermal chimney.

Make a small tower about 4' tall and large enough to mount your radiators flat. Mount the radiators flat about midway up the chimney. Make cutouts at the bottom of your chimney so air can get in at the bottom. Air closest to the floor is coolest, which is what you want rising through your radiators if you're going passive. Heat from the radiators will warm up the air in the upper half of the box, causing it to rise and suck in cool air from the floor, setting up a fairly efficient convectional cycle. The chimney shape is useful because it allows the warm air to escape cleanly at some distance from the radiators, and allows fresh cool air to always be presented.

Think "mini-nuclear-cooler-tower" and you've basically got the idea.

Will still need fairly large radiators if you want to keep temperatures sane for 350W. I would suggest at least two dual-12cm-fan cores.

However, I think you're grossly overestimating the heat load of the CPU + GPU. You'd be very lucky if there was ever more than 160W of heat going into your loop.

[heatwave]

Angry - obviously the temperature rise will have to be kept under acceptable conditions.

Cathar - if you are the same cathar that made the twin *big @$$* radiators in Overclockers.au , yes I grossly overestimate the heat load for both CPU and GPU so this will be my watercooler for quite some time. My 9800 Pro only puts out 65W but the 6800 Pro does put out 110W under max. load. If I'l be running Vcores above 2V, I'll be producing more heat than the hottest CPUs today which some exceed 100W already - thats 210W already combined.

I am assuming the temperature of the CPU to be cooled from 75C to 45C - where the temperature of the air increases from 35C to 45C.

Using the heat transfer law Q = U*A*delta T, for a flowrate of 3L a minute, that is like removing 2000W of heat based on the specific heat of water and its flowrate.

Also only 4 square meters of area is required but this equation although useful for simplicity, doesn't show the properties of the metal being used as the heat exchanger, let alone what flowrate of air is required to cool it down. I'll look into more details for more equations.

[AngryAlpaca]

But what are acceptable conditions? I'd consider 50C too high, while others may be willing to push processors to the limit (~80C) Anything above 2V is considered to be dangerous to your processor. Incidentally, a top end Opteron produces 89W and the FX produces even more.

75C-45C? Umm... As this is a steady state system, it's closer to going from 20C to 45C, unless you intend to be sending the water in bursts and in small amounts. 2000W assumes much better heat transfer than we get if you want a 22C delta T. With 2000W and a thermal resistance of 0.12C/W (block alone, as measured by Joe Citarella) you'd get temperatures in the hundreds...

That's an excellent idea, Cathar. Wow... That's just... Wow.

[LPorc]

If by stringing them together you mean in series, think parallel instead. The temperature difference between the water and air will be greater over a larger surface area in parallel than in series.

Interesting thought with the chimney... I have had a similar notion but less passive. Some thought experiments and validation by tinkering have led me to believe that most fan shrouds are way too close to the radiators. For purposes of footprint, I figured going vertical would be best. I have been contemplating a vertical duct of about 3 foot height with the radiator mounted about 1/3 of the way up the length with a pair of side-by-side fans exhausting upwards about 1 foot above the radiator.

Another notion I am kicking around has parallel runs of 1/4 ID copper tubing run back and forth several times in a long narrow rectangular box, open at the bottom, vented at the top with some quiet fans to aid convection. I suppose that's a chimney of sorts, too. The idea was inspired by a Xice passive cooler that used plastic tubing wound around a frame in a convection box to do a barely adequate job for a single CPU. I am thinking it could be built into my desk as a combination backsplash/modesty panel.

[heatwave]

Lporc - i'm not so sure how you would connect the radiators in parallel? The reason I opt for series was the water would subsequently get cooler and cooler as it passes through each radiator. Its pretty much the same effect as having one large radiator but with smaller radiators in series, you'll also have entrance and exit losses - but since I am assuming the pump is able to overcome all flow restrictions.

Angry - I think you are using an incorrect equation there. When we design heat exchangers, we don't use thermal resistance, but overall heat transfer coefficient - U, which has the SI units W/m2.C. The typical heat exchanger is the counterflow of 2 liquids within 2 tubes and these having typical U values of 110-350 for a water-oil system. The heat transfer that occurs in the WC system is convection, not conduction.

For the watercooling system, we are interested in water-air and the typical U values for a "finned-tube heat exchanger, water in tubes, air across tubes" is between 25-55.

I got the value of 2kW but that was an error, it is actually larger as it depended on the:
* water flowrate
* temperature difference

I assumed:
* all units are in SI
* flowrate of water = 3 Litres/min = 0.05 kg/s (m)
* specific heat of water = 4180 J/kg.K (Cp)
* delta T = 75-45 = 30C (delta T)

The formula to calculate heat, q = m * Cp * delta T (where q is in Watts)

Therefore with the above values:

q = (0.05) * (4180) * (30)
= 6,270W
= 6.27kW!

That is a very large heat loss. But do remember that is assuming it takes 3L/min of water and 30C temperature difference. Now this does not show what material is being used - but these equations are used to design large commercial heat exchangers in power plants where heat removal is in the excess of hundreds of kilowatts.

Lets speculate what materials they could have used:

Diamond = 2300
Silver (pure) = 410
Copper (pure) = 385
Aluminum (pure) = 202
Iron (pure) = 73
Carbon steel (1% C) = 43

Those values are the "thermal conductivities" which is in W/m.C - the higher the number the better the material is to be used in a heat exchanger. For example, it takes 2300W of heat to bring 1m of the material up one degree C.

Diamond and silver are obviously cost prohibitive, iron and carbon steel are cheap but sizewise they would have to be built 3 to 10x the size of copper and aluminum. We all know that copper heatsinks are better than its aluminum counterparts so we can assume that these equations hold for our mini watercooling system.

To solve for the required cooling surface area the equation q = U * A * Tm is used.

That q is the same as before, the A is the surface area in m2, U is the heat transfer coefficient (see above) and I chose 25 as the worst case scenario, Tm is the log mean temperature difference which is:

Tm = [ (Tw1-Tw2) - (Ta2-Ta1) ] / ln [ (Tw1-Tw2) / (Ta2-Ta1) ]

where:
* Tw1 - initial water temperature (75C)
* Tw2 - final water temperature (45C)
* Ta1 - initial air temperature (35C)
* Ta2 - final water temperature (45C)

solve for Tm = 18.2C

I chose an initial air temperature at 35C because this is pretty much as hot a typical room gets in the summer period. The two 45C shows the temperature I expect the CPU to be operating at under maximum load.

Rearranging the equation to solve for A:

A = q / (U * Tm)

= 6270 / (25 * 18.2)
= 13.8m2

Now this example shows how the equations work and relate to one another.

NOTE:
Don't be mistaken that this WC system assumes the CPU is at 75C! Its not. I assumed that in a stagnant system, the CPU at such high speed will be heating the water to 75C. Thus at 3L/min water flowrate and with a 14m2 surface area, the water temperature will drop to 45C and stay so in a steady state operation.

Albeit I am attempting to solve for only 350W heat dissipation, the surface area would be considerably smaller than above.

Here's an interesting article on a undergraduate project from the chemical engineering department of University of Brunswick in Canada - http://www.unb.ca/che/Undergrad/proposed/radiator.pdf

I have the same book "Heat Transfer by J.P. Hollman" which is an excellent book on heat transfer. I will be investigating this PDF file more closely.....

[LPorc]

Originally Posted by heatwave

Lporc - i'm not so sure how you would connect the radiators in parallel?

I suppose given your equations a monster head pump the difference in m is more affected by the split in flow than by the restriction of the radiators, thus making it a more meaningful drop than delta-T in the second radiator. Parallel would be .5m * delta-t * 2, serial would be m * delta-t + m * delta-delta-t.

However, that's just the math. Math is perfection, here in reality things get a little messier. I think the math shows us the best we can hope for, however we work with what we have got. Take this article, on the Thermochill radiators. It has been observed that the dissipation of heat is not affected much by the flow rate of the water above a low point, but rather by the flow rate of the air. I think this might be a case where we have to move enough air to hit the full transfer value in your equations. But you are talking about passive...

Based on that, I would expect m to not be much of a factor so long as the split flows were not dropped to too low of a flow rate. That leaves delta-t as the big driver, and parallel would be better. Heck, with a few Y splitters and an extra length of tube you could try it both ways. I won't be satisfied myself until I have run a setup serial and parallel and compared the water tempatures at equilibrium. Right now I just don't happen to have two identical radiators and a monster-head pump. The pump is on order, tho

[heatwave]

The maths above is as simple as it gets in heat transfer. There are lots of stuff to worry about and to simplify:

* natural convection equations for the heatcore

* how to even calculate the effective surface area of a heatcore is hard to find

* the real scenario is actually cooling 2 heat points (GPU and CPU) - not cooling a uniformly heated fluid which is the usual case of a heat exchanger

* need to perform a heat balance with the GPU and the CPU - this is a conduction and convection problem all in one - i.e. pain in the backside

* simplify the problem by ignoring flow restrictions

* Plus several others I might have missed.

Maths is perfection - being a chemical engineer myself I know the difference between theory and practice. But with maths, it gives us quantitative numbers which we can work and compare with.

I've been in several extreme cooling boards and most folks are not even sure if their WC system is up to scratch. They just follow others and buy top of the line everything which in the end isn't cost effective as you could have bought the high end CPU and save yourself the hassle of researching and implementing watercooling.

The equations aren't hard for heat transfer, you just have to find and use the right one. By understanding the science, problems which may be trivial could be solved easily.

[HammerSandwich]

Some of your assumptions smell wrong to me, Heatwave.

First of all, a 30C difference in water temp is not feasible for this application. More than 6kW would melt the PC.

Next, you assume 45C water will match the CPU temp, but this completely ignores the waterblock's function. Remember that the block will have a C/W of (very roughly) .1, so CPU temp will always be significantly higher than water temp. With your 6270W example, the CPU would be more than 600C hotter than the water. This might hurt your overclock a bit...

Your radiator calculations have similar problems, I think. Is U=25 a reasonable number for passive operation? Air's specific heat is about 1/4 water's. Thus cooling 3lpm of water 30C with a 10C increase in air temp would require 36kg of air per minute. That works out to about 1000cfm (sorry for the units, but cfm is something we all know). Will you get 1000cfm without fans, even with a 14m^2 radiator?

I suggest you put that 350W goal into your equations and try again.

(edited for typo)

[AngryAlpaca]

Ok, I see how it works now. How do you plan to drop the temperature 40C over 4 seconds to only 10C above ambient? It's especially hard when passive, and why is the water stagnant at all?

problems which may be trivial could be solved easily.

All problems that are trivial can be solved easily...

[Ozymand]

Originally Posted by LPorc

Another notion I am kicking around has parallel runs of 1/4 ID copper tubing run back and forth several times in a long narrow rectangular box, open at the bottom, vented at the top with some quiet fans to aid convection. I suppose that's a chimney of sorts, too. The idea was inspired by a Xice passive cooler that used plastic tubing wound around a frame in a convection box to do a barely adequate job for a single CPU. I am thinking it could be built into my desk as a combination backsplash/modesty panel.

I have always wondered if it would of been possible to improve upon that design by using copper to aid in thermal transfer (mostly because I'm hoping to find more of a narrow/tall design for a external water-cooling unit rather than a short/wide form factor.. Xice initially interested me but then lost out when the performance was sub-par).

A few times that topics similar to this have come up in ProCooling (such as this topic) , the idea seems to recieve rather poorly with most people citing the large flow resistance of having so much copper tubing outweighing the improvements (compared to the conventional heatercore setup, of course).

Though, there's nothing saying that it shouldn't be done. I'm going to harass a friend that has experience working with copper tubing to see what can be done. With the good low-flow performance of the Swiftech 6000 waterblock has given hope of making such a system work well (maybe not outstanding, but well for a compact and clean setup that won't take up too much space nor generate too much noise).

[LPorc]

Originally Posted by Ozymand

I have always wondered if it would of been possible to improve upon that design by using copper to aid in thermal transfer (mostly because I'm hoping to find more of a narrow/tall design for a external water-cooling unit rather than a short/wide form factor.. Xice initially interested me but then lost out when the performance was sub-par).

After seeing some setups of varying degrees of sophistication, from downright ghetto to high art, I figure it's worth a shot. Rather than use a single run like Xice did, I'd think parallel runs would reduce restriction a lot. Regardless, the Swifty 6000 is a block that makes sense in a lot of ways, not just low flow performance. It is complex in its simplicity rather than simple in its complexity.

Check out the bottom of thread.

Relevent info from lolito__fr: 50m of 4/6mm coiled LDPE pipe, 28 parallel runs coupled with 16/18mm copper manifolds. Total surface area: 0.94m2. Preliminary tests indicate very low flow resistance, and 7°C delta T at 45W (slightly worse than the hoped 5°C) -- and it is an open frame setup with 1mm wall plastic tube!

Also see Medusa. Looks like the pictures are gone, though.

Despite it being in a 3 TEC system relevent info, 38c water 50m 8mm microbore soft copper tubing (whatever that is) in completely open still air setup, dropped to 26c water when unspecified domestic fan blown over it, which seemed to match or exceed the performance of his Saab heater core in a cooling tower.

I also saw in a post here, but cannot find, what looked like an evap bong but was a coiled tube forced air radiator that the owner claimed performed better than a heater core.

I think there is plenty of room for experimentation with long parallel runs of copper tubing/piping for passive with a chimney or forced air. I guess the rub is that with automotive heater cores being so cheap the majority of enthusiasts aren't interested, and even if you came up with something that worked well it would still be likely to be bigger than what would fit in a case, and even if it was the size of a small briefcase and had a handle for the guys that want to move their systems around, nobody would want to make it a product because of the materials cost and shipping weight

Bill A made a comment, probably off-hand, about having some ideas about radiator design for w/c, but nothing more developed out of it (at least in this forum, the only w/c forum I lurk in). I doubt he'd be working in the same paradigm, language barrier aside

[Incoherent]

Originally Posted by heatwave

[b]

I assumed:
* all units are in SI
* flowrate of water = 3 Litres/min = 0.05 kg/s (m)
* specific heat of water = 4180 J/kg.K (Cp)
* delta T = 75-45 = 30C (delta T)

The formula to calculate heat, q = m * Cp * delta T (where q is in Watts)

Therefore with the above values:

q = (0.05) * (4180) * (30)
= 6,270W
= 6.27kW!

That is a very large heat loss.

Indeed it is. I believe the delta T in this equation should be the difference between inlet and outlet water temperatures, not the delta between starting temp and final temp. 30C for any radiator of the size we talk about, at this flow rate, in air, is not possible. Replace the 30 with 0.3 and you'd be nearer the mark I'd say.

Originally Posted by heatwave

... they would have to be built 3 to 10x the size of copper and aluminum.

I'd say not. The proportion of the radiators thermal resistance due to conduction is minimal. As you say, convection dominates.

To solve for the required cooling surface area the equation q = U * A * Tm is used.

That q is the same as before, the A is the surface area in m2, U is the heat transfer coefficient (see above) and I chose 25 as the worst case scenario, Tm is the log mean temperature difference which is:

Tm = [ (Tw1-Tw2) - (Ta2-Ta1) ] / ln [ (Tw1-Tw2) / (Ta2-Ta1) ]

where:
* Tw1 - initial water temperature (75C)
* Tw2 - final water temperature (45C)
* Ta1 - initial air temperature (35C)
* Ta2 - final water temperature (45C)

solve for Tm = 18.2C

I chose an initial air temperature at 35C because this is pretty much as hot a typical room gets in the summer period. The two 45C shows the temperature I expect the CPU to be operating at under maximum load.

An operating CPU is never at water temperature. It is at a temperature determined by the thermal resistance of the block in C/W, the power output of the CPU, the water temperature and a host of other variables. Think 15C above water T. Typically-ish for a good block.

Rearranging the equation to solve for A:

A = q / (U * Tm)

= 6270 / (25 * 18.2)
= 13.8m2

Now this example shows how the equations work and relate to one another.

NOTE:
Don't be mistaken that this WC system assumes the CPU is at 75C! Its not. I assumed that in a stagnant system, the CPU at such high speed will be heating the water to 75C. Thus at 3L/min water flowrate and with a 14m2 surface area, the water temperature will drop to 45C and stay so in a steady state operation.

If are saying that ALL the water in the system was pre-heated to 75C in a stagnant system then OK. Maybe I'm not reading this correctly but I think your assumptions are not quite right. Aren't you just plugging random numbers into formulae?