2 Pumps, 2 Loops, 1 Reservoir: Have you ever?

[Zarathustra[H]]

Hey all,

I have been considering adding some radiator capacity to my build for some time, but in doing so, I am not convinced my single D5 will be sufficient to keep up with the desired flow rate of > 1 GPM.

I started considering various solutions, like dual pump tops, and putting pumps in various locations of the loop, but none of them really appealed to me.

Then I had an idea. What if I split it into two loops?

No, I'm not talking the way people normally do it, with a separate loop per component. That is inefficient and I don't want to do it.

I was thinking as follows. Two pumps in separate loops, both pulling from the bottom of a single large reservoir, and returning at the top of that reservoir.

Loop 1:
Reservoir -> GPU -> CPU -> Reservoir

Loop 2:
Reservoir -> All the radiators -> Radiators

The idea is the hot return from the block loop mixes with the cold return from the radiators in the reservoir.

The potential benefit is, because there are only two blocks in one of the loops, that flow rate can go pretty damned high minimizing the DeltaT between the coolant and the die.Meanwhile, if there is a lot of resistance in the radiator loop, and the flow slows down, it's not as much of a big deal

I would set pump speed for the block loop based on the max of the coolant temp leaving each block, and the pump speed and fan speed for the radiator loop based on the temperature of the coolant leaving the reservoir.

Functional idea, or stupidity?

I'd appreciate your input.

 

[owcraftsman]

Makes no sense to me, it seems like you would want to send the coolest water available to the blocks for best performance vs mixing warm into coolest available which is all a bit moot when you factor in normalization after the loop has been running for an hour or so under load. Bottom line the loop is only as good as the components that comprise it. If you are looking to improve cooling add more radiator surface or better fans either would benefit you more than this scenario.

 

[Tsumi]

I don't see the benefit in that. I run a dual loop single reservoir system, mostly because that's the only way my reservoir of choice will work (Alphacool Repack dual D5 bay reservoir). Mine is set up as follows:

Loop 1: Reservoir -> CPU -> 280 intake radiator -> reservoir
Loop 2: Reservoir -> GPU -> 480 exhaust radiator -> reservoir

I have 4 additional 120mm fans for intake, so the 480 isn't going to be starved for cool air. If I could find a dual D5 in series bay reservoir that looked as nice as the Repack, I would go for that. I have an RGB LED strip that lights up the entire reservoir, and I'm not a huge fan of the tube reservoirs.

 

[Smoked Brisket]

I am no scientist, but it seems that the normalization of the loop temp should still hold true because its not really two loops, its more like a figure 8 hybrid loop if i am picturing it right. Why not try it? Maybe set it up on a test bench first if possible.

 

[thesmokingman]

That's not two loops and as far as best use practices, it's a terrible idea.

 

[Zarathustra[H]]

That's not two loops.

Well yeah, they are merged, but I struggled to describe it concisely without calling it that.

They are like two loops, sharing the same reservoir.

 

[thesmokingman]

Well yeah, they are merged, but I struggled to describe it concisely without calling it that.

They are like two loops, sharing the same reservoir.

Yea, and its very bad. One loop fails, so will the other. Not sure you get that? All that added complexity for what?

 

[RazorWind]

I think a better solution is probably something like a single huge circuit with two pumps. I could see the shared reservoir arrangement leading to a situation where you have hot coolant in the jacket loop and cool coolant in the radiator loop and not enough mixing in the reservoir to actually do any cooling.

I'd be interested to see some test results, though.

Edit: I could see test results from this being difficult to replicate, since the level of mixing in the reservoir would depend on so many variables, such that one person might find it works great for them, and someone else might not, thanks to things like how they arranged the tubes going into the reservoir.

 

[Zarathustra[H]]

Yea, and its very bad. One loop fails, so will the other. Not sure you get that? All that added complexity for what?

It was really just to simplify routing.

With the added radiators and the case I am considering this wound up making for a very clean routing whereas adding multiple pumps and a single loop around a of the components wound up being a little awkward.

There are other benefits as well.

When you use a dual pump top you wind up with higher pressure in the lines right after the pump. I was concerned this might result in added risk of leaks compared to keeping pressures lower with a single pump per "loop".

You can reduce the likelihood of this by spreading the pumps out far from each other in the same single loop, but that becomes even more of a difficult routing.

 

[Tsumi]

You will lose some efficiency because you won't get the hottest water going to the radiators. Lower temperature delta = less heat transfer. In practice, I would be surprised to see more than a 3 degree difference between dual parallel loop and single loop. If it makes your tube routing significantly easier, I say give it a try.

 

[Zarathustra[H]]

You will lose some efficiency because you won't get the hottest water going to the radiators.

That is a good point. I hadn't thought of that. I'm aware of the Delta T relationship to the rate of heat transfer, just hadn't through that part through yet.

Maybe I'll have to think of something else...

 

[thesmokingman]

It was really just to simplify routing.

With the added radiators and the case I am considering this wound up making for a very clean routing whereas adding multiple pumps and a single loop around a of the components wound up being a little awkward.

There are other benefits as well.

When you use a dual pump top you wind up with higher pressure in the lines right after the pump. I was concerned this might result in added risk of leaks compared to keeping pressures lower with a single pump per "loop".

You can reduce the likelihood of this by spreading the pumps out far from each other in the same single loop, but that becomes even more of a difficult routing.

That is only an issue in your mind I'm sorry to say. The problem with shared res again are two fold, one loops breaks so too will the other. The temps of both loops will equalize in time, thus defeating the point. You have no gains from this setup, only losses. If yer going to run dual loop, then run dual loop or vice versa multiple pumps. There are multi pump tops for a reason. Personally I'd just run a dual pump top and regulate pump/fan speed via water temp and be done.

 

[Zarathustra[H]]

That is only an issue in your mind I'm sorry to say. The problem with shared res again are two fold, one loops breaks so too will the other. The temps of both loops will equalize in time, thus defeating the point. You have no gains from this setup, only losses. If yer going to run dual loop, then run dual loop or vice versa multiple pumps. There are multi pump tops for a reason. Personally I'd just run a dual pump top and regulate pump/fan speed via water temp and be done.

Are pump failures really common in your experience? Because I have never seen one, but I have only been doing this for 4 years or so.

As far as other failures go, the number of fittings/joints are going to be the same either way, maybe slightly less with the two "loop" solution if the routing is better.

 

[Zarathustra[H]]

That is a good point. I hadn't thought of that. I'm aware of the Delta T relationship to the rate of heat transfer, just hadn't through that part through yet.

Maybe I'll have to think of something else...

Come to think of it, if the flow is high enough, which it will be, the temp in the reservoir should.be pretty much the same as it is anywhere in the loop, so the efficiency losses will likely be negligible.

 

[thesmokingman]

Are pump failures really common in your experience? Because I have never seen one, but I have only been doing this for 4 years or so.

As far as other failures go, the number of fittings/joints are going to be the same either way, maybe slightly less with the two "loop" solution if the routing is better.

It's not for pump failures specifically but the doubling of head pressure. You have for all intents and purposes two loops, you can't run them both off one pump can you? If you were going for pump redundancy then you'd need to add another pump on top. I'm assuming you need two pumps just to get to baseline.

 

[Tsumi]

One of my D5 pumps has been going strong since 2011 (built 2007), and the other since 2014. D5 pumps aren't very likely to fail due to overheating, unlike 35x pumps.

 

[hititnquitit]

Try it. Your going to have plenty of surface area for both cpu and gpu so it isnt as if your temps will ever be a problem. Your temps will be the same as if you were running a single loop. Well your essentially doing just that...
As for pump failures, ive had several ddcs of all flavors die(4 or 5) but never had one of my d5s die.
Adding a second pump wouldnt be a terrible idea for redundancies sake but not something thats going to make or break anything. One d5 should still do fine once the loop is filled.
Give it a try and see how you like it. The worst that can happen is you go back to a traditional loop.

 

[VanGoghComplex]

I don't know how effective this will be, but I can't help myself from liking it. You're changing the meta of the liquid cooling loop. Your reservoir is becoming a liquid to liquid heat exchanger.

Try it for science. I don't expect performance gains but I'd be tickled pink to see it work.

 

[Zarathustra[H]]

I don't know how effective this will be, but I can't help myself from liking it. You're changing the meta of the liquid cooling loop. Your reservoir is becoming a liquid to liquid heat exchanger.

Try it for science. I don't expect performance gains but I'd be tickled pink to see it work.

Yeah, I wasn't expecting any efficiency gains. In fact, trying to think about it theoretically, the potential changes in efficiency are all negative unless you can keep a high enough flow. At high flow rates it should be approximately equivalent to having everything in the same loop.

I got the idea when laying out the routing as it would be easier, and then thought it might be a fun thing to try.

This will probably be a future state though.

I think for my next loop upgrade I am going to just stick with the single pump. When I add more radiators down the line, this will be on my list of things to try.

 

[Nimisys]

Personally I like the idea, and if I had a second pump I'd give it a go, just because.

I am debating going fully parallel on my next setup. I have had the blocks (2x gpu, 1x CPU) parallel, but the radiators in serial without issue for many a year now. But reality is, given the flow rate, it's not like you get a ton of DeltaT across the radiators.

So why not just put both radiators in parallel with each other and the blocks (those left in serial) between the pump and return to reservoir? It's not as if I won't be getting coolant through all the components, and it'll be higher flow rate than all of them in serial anyways?

 

[VanGoghComplex]

Zarathustra[H] !

I realize now (six months after the fact) why this seemed such an intriguing question to me! And I know it will work!

This is very similar in concept to a cooling loop that is set up at the plant where I manage the controls system. The pump splits into two parallel loops - one side goes to the thermal load and the other side goes to the heat exchanger that brings the temps down. The two loops then recombine right at the suction side of the pump.

The loops we have at work are several thousand gallons and cool huge heat loads and use gigantic outdoor chillers attached to a glycol loop... but the theory is exactly the same as if you're talking about chips and radiators instead.

I can't say whether this will perform as well as a traditional loop, but I know it will perform: I see it in action every day!

 

[Nimisys]

Zarathustra[H] !

I realize now (six months after the fact) why this seemed such an intriguing question to me! And I know it will work!

This is very similar in concept to a cooling loop that is set up at the plant where I manage the controls system. The pump splits into two parallel loops - one side goes to the thermal load and the other side goes to the heat exchanger that brings the temps down. The two loops then recombine right at the suction side of the pump.

The loops we have at work are several thousand gallons and cool huge heat loads and use gigantic outdoor chillers attached to a glycol loop... but the theory is exactly the same as if you're talking about chips and radiators instead.

I can't say whether this will perform as well as a traditional loop, but I know it will perform: I see it in action every day!

I haven't posted pictures yet, but this is exactly how I have set up my latest version on the "Your doing it wrong" setup I use..



Still running twin 7970s at GHz speeds as well as the 1700x, so it will see quite a bit of thermal load.

 

[VanGoghComplex]

I haven't posted pictures yet, but this is exactly how I have set up my latest version on the "Your doing it wrong" setup I use..

View attachment 265428

Still running twin 7970s at GHz speeds as well as the 1700x, so it will see quite a bit of thermal load.

Sweet crap you are doing it wrong.

That looks really intriguing, but all the metal chips everywhere and the power supply tucked under there make me nervous as heck. Do I see a custom res sandwiched between the sides of those rat-rod Swifty rads?

 

[Nimisys]

Sweet crap you are doing it wrong.

That looks really intriguing, but all the metal chips everywhere and the power supply tucked under there make me nervous as heck. Do I see a custom res sandwiched between the sides of those rat-rod Swifty rads?

Pic was taken during the mock up hardline stage. It's currently apart for paint and other details. Res feeds pump, pump feeds splitter. Rads in parallel, blocks in series, back to the other splitter and returned to the res. QDC and soft tube at each block right now.

And yeah that res is custom. Got tired of them cracking. So it's 1/4" plexiglass and brass with copper fittings soldered to the brass end caps. holds close to a liter. Fill port and drain valve also incorporated.

I'll post more once I get it finished.

 

[sabrewolf732]

Uhm I need to see this Nimisys

 

[BitRage]

Hey all,

I have been considering adding some radiator capacity to my build for some time, but in doing so, I am not convinced my single D5 will be sufficient to keep up with the desired flow rate of > 1 GPM.

I started considering various solutions, like dual pump tops, and putting pumps in various locations of the loop, but none of them really appealed to me.

Then I had an idea. What if I split it into two loops?

No, I'm not talking the way people normally do it, with a separate loop per component. That is inefficient and I don't want to do it.

I was thinking as follows. Two pumps in separate loops, both pulling from the bottom of a single large reservoir, and returning at the top of that reservoir.

Loop 1:
Reservoir -> GPU -> CPU -> Reservoir

Loop 2:
Reservoir -> All the radiators -> Radiators

The idea is the hot return from the block loop mixes with the cold return from the radiators in the reservoir.

The potential benefit is, because there are only two blocks in one of the loops, that flow rate can go pretty damned high minimizing the DeltaT between the coolant and the die.Meanwhile, if there is a lot of resistance in the radiator loop, and the flow slows down, it's not as much of a big deal

I would set pump speed for the block loop based on the max of the coolant temp leaving each block, and the pump speed and fan speed for the radiator loop based on the temperature of the coolant leaving the reservoir.

Functional idea, or stupidity?

I'd appreciate your input.

I really dislike how the watercooling community has become "only do it this way" and preaching things they have heard from others in completely different scenarios with different variables.... This is what I am currently doing in a glycol chilled 32 core system with a 4090 and 1900mm of radiator, 2 ddc in series and 2nd loop is a d5.... I have a single 300mm reservoir, "Primary loop" for cpu and gpu and "2ndary loop" for vrm, chipset, via ek pro manifold.... In terms of cooling it allows you to control flow to individual components without losing flow to others, it is more efficient then 2 separate loops as its taking advantage of all radiators, so you don't have rads just sitting when doing cpu intensive loading but not gpu or vise versa... Yes there are more possibilities of failure, more expensive, more complicated, WE ARE OPEN LOOP GUYS THAT SHEET IS A GIVEN, WE ALL MASOCHIST ANYWAYS!
slSay it's "useless" is the same as saying sports cars are useless cuz regular cars will get u there too, high end desktops are useless because a mid tier system will play the games too, watercooling is useless because you can still use computer relatively the same with air cooling.... like common guys... just because it's not ur cup of tea don't call other ppl dumb for liking that cup of tea.

 

[owcraftsman]

technically three pumps sharing the same reservoir is a single loop.

 

[RazorWind]

technically three pumps sharing the same reservoir is a single loop.

Not necessarily. It’s one coolant volume, but if the three pumps aren’t in series, it would technically be three loops.

 

[owcraftsman]

Because all heat dump ends up in the same place its a single loop in series or not.

 

[Tsumi]

I really dislike how the watercooling community has become "only do it this way" and preaching things they have heard from others in completely different scenarios with different variables.... This is what I am currently doing in a glycol chilled 32 core system with a 4090 and 1900mm of radiator, 2 ddc in series and 2nd loop is a d5.... I have a single 300mm reservoir, "Primary loop" for cpu and gpu and "2ndary loop" for vrm, chipset, via ek pro manifold.... In terms of cooling it allows you to control flow to individual components without losing flow to others, it is more efficient then 2 separate loops as its taking advantage of all radiators, so you don't have rads just sitting when doing cpu intensive loading but not gpu or vise versa... Yes there are more possibilities of failure, more expensive, more complicated, WE ARE OPEN LOOP GUYS THAT SHEET IS A GIVEN, WE ALL MASOCHIST ANYWAYS!
slSay it's "useless" is the same as saying sports cars are useless cuz regular cars will get u there too, high end desktops are useless because a mid tier system will play the games too, watercooling is useless because you can still use computer relatively the same with air cooling.... like common guys... just because it's not ur cup of tea don't call other ppl dumb for liking that cup of tea.

This is exactly the kind of post that is useless advice. Things are a certain way for best optimization because that's what's been found out through years of experience.

There is a scientific objective reason as to why the split loop as proposed by the OP is less effective than a single loop: the cooled water from the radiator gets mixed with the hot water from the heat sources, leading to a higher average coolant temperature because the radiators aren't seeing as hot of a temperature and the components are seeing hotter temperatures than they would if things were in a single loop. In practice, whether or not this is significant depends on coolant flow rates. The higher the flow rate, the less significant the difference.

There is also another reason why when having two pumps, single loop is generally regarded as superior: redundancy. If one pump fails in a single loop, the other pump can continue coolant flow at a reduced flow rate until the other pump can be replaced. In a split flow with a pump powering each flow, if one pump dies, the system is unusable when thermal limits are reached because one side would have lost all heat transfer capabilities. Of course, this can be mitigated by having four pumps, two on each side, but that greatly increases the expense of the loop. In fact, one of my D5s just died a few months ago and I didn't realize until my CPU started overheating (older pump that had no RPM output). Thankfully the passive heat transfer was enough to let me use the computer in idle state, but gaming was out of the question. If the pumps were in series I could have carried on until the new pumps arrived as my system has more than enough capacity even at reduced flow rates.

There is only one reason you would ever want to reduce flow, and that is pump noise. Otherwise, the more flow you have the better. There really is no advantage to being able to control flow to each loop.

In a single series setup, you are always maximizing cooling to all components. Even in the situation where you have one hot component after the other, the flow rates are generally high enough that the loop temperature stays pretty constant throughout, especially with two pumps.

I imagine with the large industrial systems, a factor other than maximizing cooling potential is driving the use case for having the split flows. It could be for ease of component management, minimizing restriction for high flow pumps, or somehow less costly to build and maintain. For the PC watercooling space though, series setups are generally superior to parallel setups. How much superior depends on the specifics of the build in question.

 

[Enigma]

This is exactly the kind of post that is useless advice. Things are a certain way for best optimization because that's what's been found out through years of experience.

There is a scientific objective reason as to why the split loop as proposed by the OP is less effective than a single loop: the cooled water from the radiator gets mixed with the hot water from the heat sources, leading to a higher average coolant temperature because the radiators aren't seeing as hot of a temperature and the components are seeing hotter temperatures than they would if things were in a single loop. In practice, whether or not this is significant depends on coolant flow rates. The higher the flow rate, the less significant the difference.

There is also another reason why when having two pumps, single loop is generally regarded as superior: redundancy. If one pump fails in a single loop, the other pump can continue coolant flow at a reduced flow rate until the other pump can be replaced. In a split flow with a pump powering each flow, if one pump dies, the system is unusable when thermal limits are reached because one side would have lost all heat transfer capabilities. Of course, this can be mitigated by having four pumps, two on each side, but that greatly increases the expense of the loop. In fact, one of my D5s just died a few months ago and I didn't realize until my CPU started overheating (older pump that had no RPM output). Thankfully the passive heat transfer was enough to let me use the computer in idle state, but gaming was out of the question. If the pumps were in series I could have carried on until the new pumps arrived as my system has more than enough capacity even at reduced flow rates.

There is only one reason you would ever want to reduce flow, and that is pump noise. Otherwise, the more flow you have the better. There really is no advantage to being able to control flow to each loop.

In a single series setup, you are always maximizing cooling to all components. Even in the situation where you have one hot component after the other, the flow rates are generally high enough that the loop temperature stays pretty constant throughout, especially with two pumps.

I imagine with the large industrial systems, a factor other than maximizing cooling potential is driving the use case for having the split flows. It could be for ease of component management, minimizing restriction for high flow pumps, or somehow less costly to build and maintain. For the PC watercooling space though, series setups are generally superior to parallel setups. How much superior depends on the specifics of the build in question.

Amen

 

[Zarathustra[H]]

This is exactly the kind of post that is useless advice. Things are a certain way for best optimization because that's what's been found out through years of experience.

There is a scientific objective reason as to why the split loop as proposed by the OP is less effective than a single loop: the cooled water from the radiator gets mixed with the hot water from the heat sources, leading to a higher average coolant temperature because the radiators aren't seeing as hot of a temperature and the components are seeing hotter temperatures than they would if things were in a single loop. In practice, whether or not this is significant depends on coolant flow rates. The higher the flow rate, the less significant the difference.

There is also another reason why when having two pumps, single loop is generally regarded as superior: redundancy. If one pump fails in a single loop, the other pump can continue coolant flow at a reduced flow rate until the other pump can be replaced. In a split flow with a pump powering each flow, if one pump dies, the system is unusable when thermal limits are reached because one side would have lost all heat transfer capabilities. Of course, this can be mitigated by having four pumps, two on each side, but that greatly increases the expense of the loop. In fact, one of my D5s just died a few months ago and I didn't realize until my CPU started overheating (older pump that had no RPM output). Thankfully the passive heat transfer was enough to let me use the computer in idle state, but gaming was out of the question. If the pumps were in series I could have carried on until the new pumps arrived as my system has more than enough capacity even at reduced flow rates.

There is only one reason you would ever want to reduce flow, and that is pump noise. Otherwise, the more flow you have the better. There really is no advantage to being able to control flow to each loop.

In a single series setup, you are always maximizing cooling to all components. Even in the situation where you have one hot component after the other, the flow rates are generally high enough that the loop temperature stays pretty constant throughout, especially with two pumps.

I imagine with the large industrial systems, a factor other than maximizing cooling potential is driving the use case for having the split flows. It could be for ease of component management, minimizing restriction for high flow pumps, or somehow less costly to build and maintain. For the PC watercooling space though, series setups are generally superior to parallel setups. How much superior depends on the specifics of the build in question.

I have been running this system now for 2+ years, and I have to say I disagree.

Running two separate loops with a shared reservoir has been very effective for me.

I would agree that running completely separate loops is inefficient. You want to combine them into the same mass of water sharing all of your radiators, because that maximizes coolant capacity even when one block sees high load, and another is idle. When you run them as separate loops but with the coolant merged such that it goes through all radiators - however - you still get the total radiator capacity shared across all of your coolant.

I have three D5's in total. Two in one loop in series using an EK dual pump top that is my "cold side". Then there is one D5 that pushes coolant through the "hot side" just two blocks and then back into the reservoir.

The reason this is very effective simply because multiple pumps in series don't scale well.

If you have an existing loop, swapping out a single pump for a dual in series pump head gives you a little bit more flow, but not all that much. It increases the pressure and that results in a little bit of an increased flow but it is nowhere near double. It scales way way worse than SLI

By splitting out components into smaller loops you wind up with way less back pressure per loop, and this results in higher flow.

Much like with a racecar, where you get way more benefit from weight reduction than you do from increased power, you get way more benefit from reducing back-pressure than you do from adding another pump.

So the engineering/science actually makes sense here, but ONLY if you have a large complicated loop with lots of back pressure.

 

[Tsumi]

I have been running this system now for 2+ years, and I have to say I disagree.

Running two separate loops with a shared reservoir has been very effective for me.

I would agree that running completely separate loops is inefficient. You want to combine them into the same mass of water sharing all of your radiators, because that maximizes coolant capacity even when one block sees high load, and another is idle. When you run them as separate loops but with the coolant merged such that it goes through all radiators - however - you still get the total radiator capacity shared across all of your coolant.

I have three D5's in total. Two in one loop in series using an EK dual pump top that is my "cold side". Then there is one D5 that pushes coolant through the "hot side" just two blocks and then back into the reservoir.

The reason this is very effective simply because multiple pumps in series don't scale well.

If you have an existing loop, swapping out a single pump for a dual in series pump head gives you a little bit more flow, but not all that much. It increases the pressure and that results in a little bit of an increased flow but it is nowhere near double. It scales way way worse than SLI

By splitting out components into smaller loops you wind up with way less back pressure per loop, and this results in higher flow.

Much like with a racecar, where you get way more benefit from weight reduction than you do from increased power, you get way more benefit from reducing back-pressure than you do from adding another pump.

So the engineering/science actually makes sense here, but ONLY if you have a large complicated loop with lots of back pressure.

Fair point, for running a very large loop splitting the flows makes sense to keep pressure from building up too high, and that's probably why it can be used on large industrial setups.

Martin's Liquid Lab testing (great source of information in the old days) shows series setups netting about 30% gains in flow in a low restriction setup, but a high restriction setup gains about 50% flow.

However, I'm not sure if the gains are as high as you think by running separate loops with shared reservoirs. Restriction is additive at the same flow rate, just like pump pressure is additive. If you're getting 1 GPM through each loop, you'll get 1 GPM putting everything in series. Same localized flow rates, the only place that sees a higher flow rate is at the reservoir.

Laing D5s are rated to about 5.5 psi of pressure. I'm fairly sure our systems can handle a theoretical maximum of 16.5 psi with 3 D5 pumps.