Follow Slashdot stories on Twitter


Forgot your password?

Liquid Cooling More than One Component? 65

static0verdrive asks: "I am new to liquid-cooling, and I have designed a system for use in a micro-ATX OpenBSD server, with the following layout: Fillport > Reservoir/Pump > Y Split (one to CPU and the other to chip-set) > Y Reconnect > Radiator/Fan > Back to the fillport. I don't like the idea of having the hot coolant coming from the CPU going directly to the chip-set, hence the Y split. Could this split cause any problems? Would there be a difference in pressure (considering the CPU is most likely a lot hotter) that could cause an issue? How would you handle liquid-cooling more than one component? What if I wanted to cool 3 components, such as in the case where I add a video card to this setup later on?"
This discussion has been archived. No new comments can be posted.

Liquid Cooling More than One Component?

Comments Filter:
  • UPenn prof (Score:5, Informative)

    by 2.7182 ( 819680 ) on Friday June 23, 2006 @03:35PM (#15591880)
    Jean Gallier at UPenn has a lot of stuff about this on his webpage: here []
    • by Fry-kun ( 619632 )
      sorry in advance for nagging but please refrain from naming your URLs undescriptive names such as "here" or "click".
      There are 2 problems with naming a link "here": those who use a link summary program won't know what the link is about unless they read the post - which completely defeats point of using the link summary program. Obviously, they get frustarated as hell.
      Second problem is that search engines (yes i mean google :P ) use the link title to categorize it. If you name the link properly, the search en
    • thanks for linking us to the front page of the university... terribly helpful.
  • suspect idea (Score:3, Insightful)

    by Jeff DeMaagd ( 2015 ) on Friday June 23, 2006 @03:36PM (#15591886) Homepage Journal
    I think the idea is potentially suspect because you have no guarantee that both sections of the split line will have equivalent flow resistance. if one is more resistant than the other, more of your flow will go to the path of least resistance. It seems overkill to liquid cool the chipset's core chip. I'd suggest doing a series connection, cool water going to the chipset then CPU because its heat output should be much smaller than the CPU.
    • Re:suspect idea (Score:3, Interesting)

      by B'Trey ( 111263 )
      So insert valves and flow meters and adjust until you get the ratio you want. (Extra credit if you have the computer monitor and automatically adjust its own cooling.)

      • Oooooooo!! Now THAT'S a whoop-ass idea! Any thoughts on how to accomplish this?!
        • Well, with no budget considerations... :) (used to work in a research lab doing some very fancy aerodynamics research)

          Digitally acutated valves are easy to find, as are digital flow meters (Google for a decent industrial supply place - McMaster-Carr [] is a good one). Find an I/O card (PCI I/O boards are cheap), then write some code to control the valves based on flow meter feedback and tie into your chip's temperature readouts and regulate temperature that way. Be sure to allow for calibration procedures

      • No.

        Extra credit if the valves are directly thermally controlled. No computers.

        If a central heating can do it why is that so difficult for a computer cooling system.

        Hint: the valves for floor central heating use valves which work based on the return fluid temperature. It should not be that hard to make something similar for Nx coolant distribution.
        • I'd imagine that, although the scale is wrong, an automotive style thermostat would work perfectly. The valve opens and closes using a spring made from two different metals laminated together which expand and contract at different rates. As the fluid temperature increases, the valve opens wider and flow is increased and heat is dissapated faster because a larger volume of fluid moves of the cooling apparatus.
      • by $RANDOMLUSER ( 804576 ) on Friday June 23, 2006 @04:46PM (#15592421)
        Damnit Jim!! I'm a programmer, not a plumber!
      • d0 - d7 of a parallel port would be a rather easy way. That's 8 possible motor speeds, or 3 possible speeds over 2 pumps + 2 on / off relays if you wanted to do it that way.

        You'd need, of course 8 relays done something like this:

        d0 ---> //1.5 v switching spst controlling 6 v//
        d1 ---> //1.5 v switching spst controlling 5 v//
        d2 ---> //1.5 v switching spst controlling 3 v//

        Assuming your pumps vary in speed from 3v to 6v. An identical setup for d3 d4 and d5.

        d6 and d7 would simply be a master on/off for
        • This is easily possible, but I would suggest that anyone planning to try this to "do it right" and use proper parallel port interfacing techniques. The OP's suggestion has merit, but properly interfacing to the parallel port isn't simply a matter of hooking up some relays to the pins. Yes, it is possible to do this, it is also possible to "blow your port" (and if you are really unlucky, your entire "super i/o" chip). Interface it right (ie, using at minimum a switching NPN transistor with current limiting b
          • I figured on using a parallel port relay board commonly found on e-bay, I don't have a link to the last one I bought or I would post it. They use small reed contacts that switch something that handles a bit more. Its actually a DPDT but wired in as if it was SPDT, one side is "beefier" than the other.

            I think most of these come with sample asm code to show you how to manipulate the registers, could be the very utility you were talking about.

            I looked into it once so I could have 3 led's, red, green and yellow
    • Just as an aside, this is kind of a neat example of the resistors-in-parallel theorem; if you equate pressure to voltage and electrical resistance to flow resistance, and current to flow rate in volume/time, I'd be very surprised if you didn't get a result where the flow through each cooling block was close to what you'd calculate by dividing the pressure drop across the block by the flow resistance through the same block.

      The reason I bring this up is that it gives you a nice way to easily calculate the fl
    • I'd suggest doing a series connection, cool water going to the chipset then CPU because its heat output should be much smaller than the CPU.
      Personally I'd do it the other way around, have the cool water going to the cpu first, then to the chipset. The cpu is going to generate alot more heat than the chipset, so you want to provide it with the coolest water first IMO. Everything else you mentioned I would agree with though (definitely avoid using Y splitters).
  • As a Slashdot user

    Just my $0.02.

  • by I_am_Rambi ( 536614 ) on Friday June 23, 2006 @03:40PM (#15591913) Homepage
    I've looked into liquid cooling for a machine, but haven't done anything yet. From what I've found out, an ideal solution is to go reservoir->pump->cpu->video card (if doing video card)->north bridge-> radiator->reservoir. You get max cooling this way. Introducing a Y splitter introduces another connection (in which liquid can leak) and split forces of liquid. The liquid will take the path of least resistance, so you might not get max colling that way. I've found that [] has a wealth of information. You might want to pose this question on their forums, as they do more liquid cooling and have people who have done many systems.
    • It's been posted before, but aparently people forget: []

      The idea is to submerge the entire computer in vegetable (or better, mineral) oil. This cools ALL components at a lot less cost than a bunch of watercool components, and at less effort as well.

      • I've seen that done as an experiment, and it's always intrigued me, but I've wondered if anyone has actually used such a system in anything approaching production. Even if "production" was 'average everyday home use.'

        I'm curious as well whether the heat transfer from a chip submerged in fluid like that is better or worse than one that has a cooling block with some sort of coolant forced through it. The dissipation of heat throughout a static volume of fluid might end up being worse than you can achieve by f
  • Loop Planning (Score:3, Informative)

    by labalicious ( 844887 ) on Friday June 23, 2006 @03:48PM (#15591989) Homepage
    You can find some really good advice and watercooling guides, like this one: [].

    The bottom line on your waterloop, in my own experience, you'll find that the order in which the water is flowing results in negligible water temperature increase/decrease.

    I have two machines WC'ed, a P4 (pre-prescott) and a Dual Xeon. The order of the loop for the P4, pump/res> radiator > CPU Waterblock > GPU Waterblock > Flow Indicator > Pump.

    The P4 only gets to about 90F during heavy gaming sessions (ATI X800XL). Then again, I have a triple 80MM fan radiator. Your results may and will vary.
    • Re:Loop Planning (Score:2, Informative)

      by Anonymous Coward
      most people are correct here that serial flow thru components is better. one thing to add is to save that Y-split for your radiators. if you have 2 equal radiators, it is better to attach them parallel so water goes slower and gives up more of its heat, without causing additional back-pressure.

      water sops up heat well enough to not worry about the heat input from your CPU, other chips do not need to be kept as cool. however, this is something to think about when buying/making your waterblocks. if you use a j
      • f you have 2 equal radiators, it is better to attach them parallel so water goes slower and gives up more of its heat

        I agree that the radiators in parallel is better for heat transfer, but your explaination is nonsense. Heat transfer improves with the flow rate of the system. Slower moving fluid transfers heat less efficiently than fast moving fluid. Attatching the radiators in parallel reduces the total flow resistance in the system and increases flow through each radiator.

  • having a good single chip water cooling system and moving to a dual chip (each being a dual core opteron) I had somewhat the same quetion.

    I previously cooled my cpu and graphics card (both 25% overclocked) but because of an initial lack of funding went back to air cooled and
    modest (10%) overclock.

    Anyone tackled with with a single pump/radiator solution?


  • my setup (Score:2, Interesting) []
    out to cpu, split one into geforce 6800 ultra OC, meet up with the split 1/4 " pipes to go back into the reservoir. works beautifully... in a non-airconditioned room in muggy SE Pennsylvania playing the Prey demo... only got to 42 Celcius. Add AC and it maxes out at 39 Celcius.
    It's a Koolance rig.
  • My take (Score:2, Informative)

    by jonging ( 981292 )
    Don't use "Y-splits" to redirect fluid. I understand your concern that you don't want hot parts connected in series but a series connection is superior to a parallel setup. The resistance overhead for a series setup is negligible. Also, was previously mentioned, it is more advantageous for you to minimize the amount of junctions and thereby increase the reliability of your setup.
  • by grumpygrodyguy ( 603716 ) on Friday June 23, 2006 @04:03PM (#15592117)
    "I am new to liquid-cooling, and I have designed a system for use in a micro-ATX OpenBSD server, with the following layout: Fillport > Reservoir/Pump > Y Split (one to CPU and the other to chip-set) > Y Reconnect > Radiator/Fan > Back to the fillport. I don't like the idea of having the hot coolant coming from the CPU going directly to the chip-set, hence the Y split. Could this split cause any problems? Would there be a difference in pressure (considering the CPU is most likely a lot hotter) that could cause an issue? How would you handle liquid-cooling more than one component? What if I wanted to cool 3 components, such as in the case where I add a video card to this setup later on?"

    Splitting the main(Y-split) to cool several devices is generally not recommend over cooling multiple components in serial.

    Most people go: Pump->CPU->radiator->reservoir

    some people go: Pump->CPU->Video GPU->radiator->reservoir

    and very few people go: Pump->CPU->Video GPU->Chipset->radiator->reservoir

    ( or even Pump->CPU->Video GPU #1>-Video GPU #2->Chipset->Memory->radiator->reservoir)

    As you add more and more stuff to the circuit, you'll also need a bigger pump, a bigger radiator, and you'll have to seal the connection points more carefully to gaurd against leaks resulting from higher pressure. It's the general consensus that splitting the coolant in a parallel fashion like you're describing is less effective than connecting the components in serial. The primary reason is that flowrate X volume is king in water cooling...with a Y-split you're cutting your CPU water cooling volume in half, and probably restricting flow even further with narrower tubing. Also, pressure drop in the system is a function of how much tubing you use. More tube, less pressure. People have tried this before...and their results weren't comparable with serial.

    I'm assuming you want to water cool so you can overclock. If this is so, then you need to prioritize your CPU over everything else. If you don't plan on overclocking and just want the silence, then you're still better off using serial because it's cheaper and safer(less connection points means a lower leak probability).

    Don't worry about warm water returning from the CPU and 'heating' the chipset. Fast flowrate and the high heat capacity of water keep this from being a problem. Generally the water temperature across the entire circuit is nearly homogenous(maybe 1-2 degrees difference).

    To learn more:

    1) Goto 46c9ab33c79d52f8485eff396&f=71 []
    2) Spend at least 2 hours reading the stickies etc. (or register and ask your own question, the folks there are very knowledgeable.)

    Good luck with it!
  • Tube size (Score:4, Interesting)

    by llZENll ( 545605 ) on Friday June 23, 2006 @04:05PM (#15592135)
    Why not just use a smaller diameter tube for the components that don't need as much cooling? KISS
  • ..maybe check out HP's new air cooling rig [] instead.
  • stop worrying (Score:2, Informative)

    hi. I'm a mechanical engineer (who loves to tinker) and has used watercooling. I'm going to suggest a few steps to help you see that your system is successful. (I'm not going to use any numbers or calculation, because I'm very lazy: 1) I did the work once for my own systems and found that the work was overkill, 2) water can carry more heat than you realize 3) I'm lazy.

    1. Relax, be lazy and have a beer. You're dealing with (I think) a server in your home that is probably going to remain lightly
  • As a Slashdot user (who knows nothing about liquid cooling), it seems to me like the way to ensure each component gets adequate cooling is to have a big, shared radiator and a smaller, separate pump for each component. That way they are each ensured their own coolant flow. The downside, of course, is that you need three pumps, which will cost more and make more noise.

    Just my $0.02.

  • by deacon ( 40533 ) on Friday June 23, 2006 @06:15PM (#15593003) Journal
    Reason 1:

    Your lower water flow in each cold-plate due to splitting the flow with a Y lowers the velocity thru each cold-plate and thus lowers the heat transfer between the water and the cold-plate.

    Reason 2:

    You do not have the equipment needed to measure the flow thru the 2 branches of the Y so you risk having 1 component be hotter than needed and not know it. Some will suggest using valves to choke flow to the higher-flowing cold-plate, but this way you are wasting pump head.

    Sadly, water cooling has come from being done right (like by IBM and the water cooled version of the VAX 9000, which was changed to air cooling before being shipped) to the use of feeble pumps and undersized radiators. In many cases, water cooling in PCs has become the equivalent of a "Type R" sticker on a Honda sedan.

    There is nothing magical about water cooling. An air cooled setup can have the same performance, given good heat sink surface area, good fin efficiency of the heatsink, and 600 feet/minute airflow. A water cooling setup CAN let you to increase the effective heatsink area for ejecting heat into the room air without a fin efficiency penalty. But to do this you need enough flow and enough radiator area, and to keep costs down most kits are marginal on both.
  • by Cyno01 ( 573917 ) <> on Friday June 23, 2006 @06:20PM (#15593031) Homepage
    It shouldn't matter, but if you're that worried about your chipset temps you could throw another 80mm rad in between the proc and northbridge. Plenty of people i know though run everything on one loop with only 1 large rad or just an aditional one between the proc and graphics card(s).
  • Parallel vs. Series (Score:3, Informative)

    by Wonko the Sane ( 25252 ) * on Friday June 23, 2006 @07:04PM (#15593245) Journal
    A parallel tubing configuration will always be superior to a series configuration (unless you are using a positive displacement pump for some reason). IF you have a method to balance the flow correctly to each component (such as throttle valves). The reduced flow resistance will allow the pump to operate at a higher flowrate and will make heat transfer in the radiator more efficient. This will reduce the outlet temperature of the radiator, supplying cooler water to each component. This makes all your components run cooler and also reduces the power usage of the pump.

    If you don't have a method to correctly balance the flow then your best bet is a hybrid series-parallel configuration. The best solution will depend on the heat load of every component you want to cool and the physical characteristics of the pump and radiator.
  • serial connections. (Score:3, Informative)

    by aquabat ( 724032 ) on Friday June 23, 2006 @09:20PM (#15593917) Journal
    I've been water cooling for a couple of years now, and it has been my experience that connecting all your blocks in a chain has no detrimental effect on their cooling ability.

    My current loop is as follows:

    Athlon MP2800+ -> Athlon MP2800+ -> AMD 762 Northbridge -> FireGL X1-128 -> Koolance 2 DIMM RAM cooler-> Reserator.

    Once the temperatures reach steady state, the difference in temperature between any two points in the loop is less than 2 degrees C. In doesn't really matter how much water is circulating or how fast (these do matter though, in determining how fast the steady state can change when a cooled component suddenly changes its temperature). What really matters is the surface area of your radiator, and the airflow over it's fins.

    Under full load, on a 30C day, the Reserator is very warm to the touch. I can drop the temperature to below room temperature by putting a fan behind it. Whatever temperature the radiator is at is the temperature the blocks are at (at steady state conditions).

    If you want to cool more components, you don't have to fill up your case with parallel cooling loops. Instead, add them in series and add another radiator is series also. You only ever need one input hose and one output hose piercing the case. The Reserators work really well for this, since you only need a pump in one of them. The other(s) are just extra surface area.

    • Sir, I submit to you that you are lying, or that your measurement technique is shit. Unless you are doing some sort of evaporative cooling, you cannot cool something below room temperature by blowing room temperature air on it.

      Yours truly,
      The Laws of Thermodynamics

      PS the Reserator is total rice, thus proving that you are a fag. You probably run Ubuntu on your "rig".
      • You are correct. I just put a temperature sensor on the radiator, and it is 27C with the fan blowing on it, while the current ambient is 25C.

        I had only put my hand on it to feel the temperature when I made the previous statement. I should have used the sensor, so yeah, my measurement technique was shit.

        As for the Reserator, it works just fine. I like it because it has no fans, and is therefore quiet. I also like that it doesn't look like an automobile part. It has surface area to radiate enough heat to

  • I drilled a couple holes each in the CPU and chipset to improve water flow. Hooked them up with a Y split to an old VW Beetle radiator. Works like a charm. Never goes over room temp.
  • Just submerge the entire system in Fluorinert [] (like the Cray-2 [] I use to admin at NASA Langley), circulate and cool the liquid, and be done with it.

    Of course, the liquid is about $270 a liter.

To write good code is a worthy challenge, and a source of civilized delight. -- stolen and paraphrased from William Safire