Liquid Cooling flow path

[Scott Mueller]

Is there an optimal path through components for a liquid cooling setup with 2 GPUs, a CPU, and a 480mm radiator? I was planning on having coolant flow like this:
reservoir -> pump -> GPUs -> CPU -> radiator -> reservoir

 

[Batboy88]

Is there an optimal path through components for a liquid cooling setup with 2 GPUs, a CPU, and a 480mm radiator? I was planning on having coolant flow like this:
reservoir -> pump -> GPUs -> CPU -> radiator -> reservoir

Nah go Rez>Pump>Rad> then everything else etc

Telling ya there is a difference it does cool off a bit before it gets to everything when you do it that way. same tried before with block first always runs like 3-5 degrees hotter...is that significant? idk maybe not...

 

[rgMekanic]

Is there an optimal path through components for a liquid cooling setup with 2 GPUs, a CPU, and a 480mm radiator? I was planning on having coolant flow like this:
reservoir -> pump -> GPUs -> CPU -> radiator -> reservoir

Flow is just fine there, the only connection that really matters is you MUST go Res --> Pump, other than that it doesn't matter much, after the radiators and coolant temperature balance, it will all be the same within a couple degrees.

The only other consideration you ned make is flow path for getting the air to bleed out of the system efficiently

 

[VanGoghComplex]

Loop order doesn't matter, except for that your reservoir be placed higher than your pump, and feeds it directly.

 

[Emission]

It matters in terms of heat distribution. Doing GPUs -> CPU is going to heat up your CPU appreciably, whereas doing CPU -> GPUs will make less of an impact on GPU temps and give you much better CPU temps.

 

[VanGoghComplex]

It matters in terms of heat distribution. Doing GPUs -> CPU is going to heat up your CPU appreciably, whereas doing CPU -> GPUs will make less of an impact on GPU temps and give you much better CPU temps.

If your flow rates are abysmal, maybe.

 

[Emission]

If your flow rates are abysmal, maybe.

Your flow rates would have to be absurdly high for it to not make any impact.

 

[VanGoghComplex]

Your flow rates would have to be absurdly high for it to not make any impact.

Can you show me where you're getting this info? The watercooling guides and stuff I've read and watched all say that block order isn't relevant. The fluid in all points of the loop equalize in temperature, meaning they're equally effective at removing heat from all blocks. The only caveat being that if your flow rate is too low, you might see a difference. But for mainstream pumps like the DDC and D5, it's a non-concern.

 

[guitarslingerchris]

Your flow rates would have to be absurdly high for it to not make any impact.

I would love to see you back that up with actual data instead of your opinion.

 

[rgMekanic]

Can you show me where you're getting this info? The watercooling guides and stuff I've read and watched all say that block order isn't relevant. The fluid in all points of the loop equalize in temperature, meaning they're equally effective at removing heat from all blocks. The only caveat being that if your flow rate is too low, you might see a difference. But for mainstream pumps like the DDC and D5, it's a non-concern.

^ this

Eventually you are going to normalize the temperature of the coolant throughout the entirety of the loop

 

[Emission]

Can you show me where you're getting this info? The watercooling guides and stuff I've read and watched all say that block order isn't relevant. The fluid in all points of the loop equalize in temperature, meaning they're equally effective at removing heat from all blocks. The only caveat being that if your flow rate is too low, you might see a difference. But for mainstream pumps like the DDC and D5, it's a non-concern.

We're those guides written by anyone with a knowledge of basic physics? There is no such thing as "normalization of the coolant temperature throughout the entirety of a loop." It's incredibly simple and I can't believe it's lost on so many of you.

Let me break it down into an easy-to-digest format.

x Temperature water -> heat source -> higher-than-x temperature water.

It's literally that simple. The CPU is going to see hotter or colder water depending on where it sits in the loop order if it is not the only heat-generating device present. If the question is "will this make an appreciable difference in terms of what temperature the device being cooled operates at?" then for the average scenario, probably not. However, when every degree of temperature counts in getting an overclock stable or ensuring a component lives a little longer, perhaps its more appreciable, especially when you're dealing with components that pump out lots of heat.

I would love to see you back that up with actual data instead of your opinion.

Let me preface this by saying that I do not currently possess the hard data to prove my point, so do whatever you like with this information, but I have designed cooling systems capable of cooling a 200W+ CPU, along with 1500W+ worth of GPUs, all in a single loop. The GPUs can shrug off seeing water saddled with 200W worth of heat after coming out of the radiator (in this case there are 4 of them, all cooled with full cover blocks, so plenty of metal to distribute the relatively small amount of heat through) but if you were to reverse the order and expose the CPU to 1500W worth of heat, you are passing that relatively hot water through a small metal block that is also trying to absorb the heat off of the CPU, and the CPU will absolutely run hotter in this scenario.

Relating this back to flow rate, heat transfer from water to another material is not instantaneous, and therefore the slower or faster the water flow, the more or less heat the water absorbs on its way through a block (surface area inside the block plays a big part in this as well).

Also, I just need to nit-pick something for a minute here, as this is something anyone should be able to understand:

The fluid in all points of the loop equalize in temperature, meaning they're equally effective at removing heat from all blocks.

So what you're suggesting is that the fluid temperature going into your radiator is identical to the temperature exiting it? That would mean the radiator is doing nothing. The only correct part of this statement is how effective the liquid is at removing heat at a given temperature, which for water is constant unless it is experiencing a phase change.

To end my rant, if you're cooling a 90W CPU and a 90-180W, hell even a 250W GPU, it's not going to make a gigantic difference in temperatures, so you can comfortably configure your loop in a way that suits your sanity, but the differences in temperature exist and they are measurable. You're welcome to experiment with it yourself if you don't believe me.

 

[Batboy88]

We're those guides written by anyone with a knowledge of basic physics? There is no such thing as "normalization of the coolant temperature throughout the entirety of a loop." It's incredibly simple and I can't believe it's lost on so many of you.

Let me break it down into an easy-to-digest format.

x Temperature water -> heat source -> higher-than-x temperature water.

It's literally that simple. The CPU is going to see hotter or colder water depending on where it sits in the loop order if it is not the only heat-generating device present. If the question is "will this make an appreciable difference in terms of what temperature the device being cooled operates at?" then for the average scenario, probably not. However, when every degree of temperature counts in getting an overclock stable or ensuring a component lives a little longer, perhaps its more appreciable, especially when you're dealing with components that pump out lots of heat.



Let me preface this by saying that I do not currently possess the hard data to prove my point, so do whatever you like with this information, but I have designed cooling systems capable of cooling a 200W+ CPU, along with 1500W+ worth of GPUs, all in a single loop. The GPUs can shrug off seeing water saddled with 200W worth of heat after coming out of the radiator (in this case there are 4 of them, all cooled with full cover blocks, so plenty of metal to distribute the relatively small amount of heat through) but if you were to reverse the order and expose the CPU to 1500W worth of heat, you are passing that relatively hot water through a small metal block that is also trying to absorb the heat off of the CPU, and the CPU will absolutely run hotter in this scenario.

Relating this back to flow rate, heat transfer from water to another material is not instantaneous, and therefore the slower or faster the water flow, the more or less heat the water absorbs on its way through a block (surface area inside the block plays a big part in this as well).

Also, I just need to nit-pick something for a minute here, as this is something anyone should be able to understand:



So what you're suggesting is that the fluid temperature going into your radiator is identical to the temperature exiting it? That would mean the radiator is doing nothing. The only correct part of this statement is how effective the liquid is at removing heat at a given temperature, which for water is constant unless it is experiencing a phase change.

To end my rant, if you're cooling a 90W CPU and a 90-180W, hell even a 250W GPU, it's not going to make a gigantic difference in temperatures, so you can comfortably configure your loop in a way that suits your sanity, but the differences in temperature exist and they are measurable. You're welcome to experiment with it yourself if you don't believe me.

Well it's not like the flow and it all stays in the radiator that long and thinking aluminum is the best heat conductor and off to everything so you do kinda have an equalization throughout the loop. It does make more sense to pass through the rad then to the blocks...

 

[VanGoghComplex]

We're those guides written by anyone with a knowledge of basic physics? There is no such thing as "normalization of the coolant temperature throughout the entirety of a loop." It's incredibly simple and I can't believe it's lost on so many of you.

Let me break it down into an easy-to-digest format.

x Temperature water -> heat source -> higher-than-x temperature water.

It's literally that simple. The CPU is going to see hotter or colder water depending on where it sits in the loop order if it is not the only heat-generating device present. If the question is "will this make an appreciable difference in terms of what temperature the device being cooled operates at?" then for the average scenario, probably not. However, when every degree of temperature counts in getting an overclock stable or ensuring a component lives a little longer, perhaps its more appreciable, especially when you're dealing with components that pump out lots of heat.



Let me preface this by saying that I do not currently possess the hard data to prove my point, so do whatever you like with this information, but I have designed cooling systems capable of cooling a 200W+ CPU, along with 1500W+ worth of GPUs, all in a single loop. The GPUs can shrug off seeing water saddled with 200W worth of heat after coming out of the radiator (in this case there are 4 of them, all cooled with full cover blocks, so plenty of metal to distribute the relatively small amount of heat through) but if you were to reverse the order and expose the CPU to 1500W worth of heat, you are passing that relatively hot water through a small metal block that is also trying to absorb the heat off of the CPU, and the CPU will absolutely run hotter in this scenario.

Relating this back to flow rate, heat transfer from water to another material is not instantaneous, and therefore the slower or faster the water flow, the more or less heat the water absorbs on its way through a block (surface area inside the block plays a big part in this as well).

Also, I just need to nit-pick something for a minute here, as this is something anyone should be able to understand:



So what you're suggesting is that the fluid temperature going into your radiator is identical to the temperature exiting it? That would mean the radiator is doing nothing. The only correct part of this statement is how effective the liquid is at removing heat at a given temperature, which for water is constant unless it is experiencing a phase change.

To end my rant, if you're cooling a 90W CPU and a 90-180W, hell even a 250W GPU, it's not going to make a gigantic difference in temperatures, so you can comfortably configure your loop in a way that suits your sanity, but the differences in temperature exist and they are measurable. You're welcome to experiment with it yourself if you don't believe me.

I will. The next iteration of my build will have coolant temperature sensors in the coolest and hottest points of the loop. I'm curious about the delta. Everything I've read and watched tells me that it will be a degree celsius or less, but I'll report my findings in either case.

 

[Emission]

I will. The next iteration of my build will have coolant temperature sensors in the coolest and hottest points of the loop. I'm curious about the delta. Everything I've read and watched tells me that it will be a degree celsius or less, but I'll report my findings in either case.

I'm honestly curious as well. I wish I still had access to more parts to test with, and had the time.

My apologies for the rant, it was late and I was quite tired.

 

[Batboy88]

I'm honestly curious as well. I wish I still had access to more parts to test with, and had the time.

My apologies for the rant, it was late and I was quite tired.

You can check it...it don't come out the block much different or into the pump/rez. it all does kinda even out.

 

[THRESHIN]

As mentioned, the thing that really matters is going from res to pump suction. This provides a bit of head pressure, very necessary for centrifugal pumps which these all are.

If you want to nit pick (like me) the outlet of the pump should go to the rad to remove pump heat.

In reality, these are small systems so heat added by the pump will be almost nothing.

As far as a temperature delta, again small. Water is indeed overkill for a computer, but hey that's just fun. The delta will be there though.

To go to an extreme example, I work at a nuclear power plant. Everything is water cooled! But even there, the temperature delta from inlet to outlet of the reactor core is only about 40 C. All the principles are exactly the same, just much bigger and there's very little water to air heat exchangers like the rads we use. It's mostly water to water. Still, it's all the same basic flowpath as your water loop when you simplify it.

 

[Tsumi]

We're those guides written by anyone with a knowledge of basic physics? There is no such thing as "normalization of the coolant temperature throughout the entirety of a loop." It's incredibly simple and I can't believe it's lost on so many of you.

Let me break it down into an easy-to-digest format.

x Temperature water -> heat source -> higher-than-x temperature water.

It's literally that simple. The CPU is going to see hotter or colder water depending on where it sits in the loop order if it is not the only heat-generating device present. If the question is "will this make an appreciable difference in terms of what temperature the device being cooled operates at?" then for the average scenario, probably not. However, when every degree of temperature counts in getting an overclock stable or ensuring a component lives a little longer, perhaps its more appreciable, especially when you're dealing with components that pump out lots of heat.



Let me preface this by saying that I do not currently possess the hard data to prove my point, so do whatever you like with this information, but I have designed cooling systems capable of cooling a 200W+ CPU, along with 1500W+ worth of GPUs, all in a single loop. The GPUs can shrug off seeing water saddled with 200W worth of heat after coming out of the radiator (in this case there are 4 of them, all cooled with full cover blocks, so plenty of metal to distribute the relatively small amount of heat through) but if you were to reverse the order and expose the CPU to 1500W worth of heat, you are passing that relatively hot water through a small metal block that is also trying to absorb the heat off of the CPU, and the CPU will absolutely run hotter in this scenario.

Relating this back to flow rate, heat transfer from water to another material is not instantaneous, and therefore the slower or faster the water flow, the more or less heat the water absorbs on its way through a block (surface area inside the block plays a big part in this as well).

Also, I just need to nit-pick something for a minute here, as this is something anyone should be able to understand:



So what you're suggesting is that the fluid temperature going into your radiator is identical to the temperature exiting it? That would mean the radiator is doing nothing. The only correct part of this statement is how effective the liquid is at removing heat at a given temperature, which for water is constant unless it is experiencing a phase change.

To end my rant, if you're cooling a 90W CPU and a 90-180W, hell even a 250W GPU, it's not going to make a gigantic difference in temperatures, so you can comfortably configure your loop in a way that suits your sanity, but the differences in temperature exist and they are measurable. You're welcome to experiment with it yourself if you don't believe me.

Anyone with intermediate physics knowledge would know the higher the flow rate, the smaller the delta between inlet and outlet until the delta becomes essentially zero. This is not an open system where coolant is being fed in at a constant temperature. This is a closed system with several points of heat exchange.

Let's take a 200 watt CPU. It will be transferring 200 joules of heat every second to the waterblock. The specific heat of water is 4.186 joules per gram per C. That means it takes 4.186 joules to heat up one gram of water 1 celsius. In order to have a 5 degree increase in temperature, you would have to have a flow rate of approximately 9.6 grams per second of water (200 ÷ 4.186 ÷ 5). That is a flow rate of 573 ml/min, or 0.15 gallons a minute. That is a very low flow rate. It is recommended to operate at 1 gallon per minute or higher. At that flow rate (63 grams per second), the delta is 0.76 degrees. Even with your ludicrous example of 1500 watts (anyone running that kind of setup will be running individual loops or very high flow loops anyways), the temperature delta is at most 5.7 degrees, which again would drop at higher flow rates.

Additionally, just because the inlet and outlet of the radiator are at the same approximate temperature does not mean no heat transfer is being done. As long as the outlet of the radiator is at a higher temperature than the air moving through the radiator, there will be heat transfer to air. The radiator does not magically cool the coolant to air temperature unless heat loads are very low or flow rates are very low. At low flow rates, you are not efficiently using the radiator because all the heat transfer occurs at the beginning with almost nothing at the end.

 

[noxqzs]

Makes me remember the times my teacher used to flip out saying HEAT and Temperature are two different things.

 

[Brian_B]

Your flow rates would have to be absurdly high for it to not make any impact.

Apart from AIOs, which usually have much smaller pumps, most external pumps do have absurdly high flow rates for the amount of heat transfer that actually needs to get accomplished. And most enthusiasts don't mind that at all, because it means greater dT, which means more efficient heat transfer and lower temps on the die.

 

[Emission]

Anyone with intermediate physics knowledge would know the higher the flow rate, the smaller the delta between inlet and outlet until the delta becomes essentially zero. This is not an open system where coolant is being fed in at a constant temperature. This is a closed system with several points of heat exchange.

Let's take a 200 watt CPU. It will be transferring 200 joules of heat every second to the waterblock. The specific heat of water is 4.186 joules per gram per C. That means it takes 4.186 joules to heat up one gram of water 1 celsius. In order to have a 5 degree increase in temperature, you would have to have a flow rate of approximately 9.6 grams per second of water (200 ÷ 4.186 ÷ 5). That is a flow rate of 573 ml/min, or 0.15 gallons a minute. That is a very low flow rate. It is recommended to operate at 1 gallon per minute or higher. At that flow rate (63 grams per second), the delta is 0.76 degrees. Even with your ludicrous example of 1500 watts (anyone running that kind of setup will be running individual loops or very high flow loops anyways), the temperature delta is at most 5.7 degrees, which again would drop at higher flow rates.

Additionally, just because the inlet and outlet of the radiator are at the same approximate temperature does not mean no heat transfer is being done. As long as the outlet of the radiator is at a higher temperature than the air moving through the radiator, there will be heat transfer to air. The radiator does not magically cool the coolant to air temperature unless heat loads are very low or flow rates are very low. At low flow rates, you are not efficiently using the radiator because all the heat transfer occurs at the beginning with almost nothing at the end.

"Ludicrous" lol. Good points, even though you gloss over minor details such as rate of heat transfer through the metal of the blocks.

 

[Tsumi]

"Ludicrous" lol. Good points, even though you gloss over minor details such as rate of heat transfer through the metal of the blocks.

Rate of heat transfer through metal doesn't matter once it hits the equilibrium state. Again, anyone with more than a basic understanding of physics would know this. The only thing rate of heat transfer affects in a waterblock is the delta between coolant and CPU die temperature, not inlet/outlet delta.

So you can stop trying to act superior to us experts because you are clearly not one.

Edit: if you really want the details my simplified model ignored, here there are: transferred to motherboard, heat transferred to air, and heat transferred from waterblock to air. All of which reduces heat transferred to the coolant. Oh, and thermal expansion of water, which would reduce the mass of water going through the waterblock.

Heat transfer is classical Newtonian physics. It follows what goes in must come out, otherwise the system is not in equilibration.

 

[Emission]

Rate of heat transfer through metal doesn't matter once it hits the equilibrium state. Again, anyone with more than a basic understanding of physics would know this. The only thing rate of heat transfer affects in a waterblock is the delta between coolant and CPU die temperature, not inlet/outlet delta.

So you can stop trying to act superior to us experts because you are clearly not one.

I'm so sorry that my offering of actual experience is a show of superiority to you.

 

[Tsumi]

I'm so sorry that my offering of actual experience is a show of superiority to you.

We're all waiting for your charts to prove us wrong. Until then, your words are worth less than the hard disk space used to store them.

 

[Emission]

We're all waiting for your charts to prove us wrong. Until then, your words are worth less than the hard disk space used to store them.

I no longer have access to the gear I would need to perform the tests to generate said data. That being said, if you wanna pay for the system I'd be more than happy to perform the testing and show you the data that proves my original point, which is that the CPU will run appreciably hotter in front of a high enough heat load and a low enough coolant flow rate as opposed to running the CPU behind that heat load.

Until then, quit your bitching, it's not adding anything to the conversation.

 

[guitarslingerchris]

I no longer have access to the gear I would need to perform the tests to generate said data. That being said, if you wanna pay for the system I'd be more than happy to perform the testing and show you the data that proves my original point, which is that the CPU will run appreciably hotter in front of a high enough heat load and a low enough coolant flow rate as opposed to running the CPU behind that heat load.

Until then, quit your bitching, it's not adding anything to the conversation.

I don't think that's true either over a long enough time period unless your argument in response is going to be the flow rate of a dripping sink.

 

[Emission]

I don't think that's true either over a long enough time period unless your argument in response is going to be the flow rate of a dripping sink.

More snarky remarks, keep them coming.

 

[VanGoghComplex]

More snarky remarks, keep them coming.

I'm noticing though the thread that the examples you provide tend to be extremely powerful enthusiast systems. My guess is that they innately have much greater restriction on the loop and produce much more heat than a relatively modest single-GPU watercooled rig. Maybe that's why our perspectives on the issue are so different?

 

[Batboy88]

I'm noticing though the thread that the examples you provide tend to be extremely powerful enthusiast systems. My guess is that they innately have much greater restriction on the loop and produce much more heat than a relatively modest single-GPU watercooled rig. Maybe that's why our perspectives on the issue are so different?

Very powerful enthusiast builds..but like one mentioned a lot external pumps have quite a bit better flow rates, it's not an AIO.

 

[Tsumi]

I no longer have access to the gear I would need to perform the tests to generate said data. That being said, if you wanna pay for the system I'd be more than happy to perform the testing and show you the data that proves my original point, which is that the CPU will run appreciably hotter in front of a high enough heat load and a low enough coolant flow rate as opposed to running the CPU behind that heat load.

Until then, quit your bitching, it's not adding anything to the conversation.

I told you exactly how much hotter it will be. 5.7 degrees C after a 1500 watt heat source at 1 gpm. However, anyone running a build that hot will be running parallel and/or flow rates above 2 gpm, which would drop the delta to about 2.9 degrees or less. Appreciable, yes, but not significant.

When you throw around scientific terms without knowing how to use them, you only make yourself look stupid. If you're resorting to insults when you cannot refute the math, you're only making yourself look petty and ignorant.

 

[Brian_B]

otherwise the system is not in equilibration

Not that I disagree with your content, but is "equilibration" a word? It's like "strategery", only more physicsier.

 

[Brian_B]

I told you exactly how much hotter it will be. 5.7 degrees C after a 1500 watt heat source at 1 gpm. However, anyone running a build that hot will be running parallel and/or flow rates above 2 gpm, which would drop the delta to about 2.9 degrees or less. Appreciable, yes, but not significant.

Just to emphasize Tsumi's post here - 1500W heat source isn't your typical computer. It isn't even your typical enthusiast computer. 1,500W is the equivalent of liquid cooling your toaster, and keeping the part that usually gets red hot and toasty, down to something cool enough that you could snuggle up to and not get burned. It's a very significant heat load, and probably 5-10X what a typical closed loop water cooling system would actually reject from cooled components.

1,500W may be the rating of the PSU that goes into that computer, but that isn't necessarily the same thing as the energy actually consumed by the parts which are cooled via water blocks.

Most enthusiast pumps are rated in liters/hr or gallons/hr. 1-2 gpm is a very credible flow rate for a typical pump in a custom loop.

Another point of reference - there is actually an efficiency standard for shower heads - they cannot be more than 2.5gpm - so the amount of water that comes out of a shower head is roughly the same amount of flow going through a high end custom loop. A tankless water heater - for example - is on the order of 12,000 W to heat that water from the temperature of your cold tap to shower-temperature (about 105F).

I didn't confirm the math in Tsumi's post, but I'm not refuting it or anything - it sounds credible in my experience. Just providing some real life comparisons.

 

[Emission]

I told you exactly how much hotter it will be. 5.7 degrees C after a 1500 watt heat source at 1 gpm. However, anyone running a build that hot will be running parallel and/or flow rates above 2 gpm, which would drop the delta to about 2.9 degrees or less. Appreciable, yes, but not significant.

When you throw around scientific terms without knowing how to use them, you only make yourself look stupid. If you're resorting to insults when you cannot refute the math, you're only making yourself look petty and ignorant.

Alright, I'll concede that I started it, and I apologize for not keeping it more civil.

I also concede that my example is an extreme case and does not qualify as a backing for general advice.

What I can tell you is that the system in question looks something like this, and this is the basis for my speculation regarding the loop order matter:

- 4930K, OC'd to ~4.3-4.4 GHz, approximately 200W TDP at typical operating voltage and temperature.
- 4x Titan Z's, approximately 1500W at factory stock clocks, roughly 65C operating temperature on average.
- Loop height: ~2ft
- 2x EK D5 Vario pumps operating at speed 3, running in series: https://www.ekwb.com/shop/ek-d5-vario-motor-laing-d5-vario-motor
- 1/2" ID tubing
- 3x Radiators: 1x 420x140mm, 2x 280x140mm

IIRC loop order was something like this: reservoir -> pumps -> 420 rad -> CPU block -> GPU blocks (semi-parallel) -> 280 rads in parallel -> reservoir

Going off of the above, the flow graph EK provides suggests 4 GPM for a single pump. The pumps sit directly next to each other at the bottom of the loop, so I'm not sure that there is a performance gain over 1 pump (I believe that's usually only the case when the pumps are spread out height-wise through the loop), the design choice was primarily for the sake of redundancy in case of a pump failure.

So for the sake of calculation, lets say we're working with 4 GPM base flow. I'm not 100% sure that you're still going to see 4 GPM by the time that fluid works its way upward through four dual-GPU blocks, but you're welcome to correct me.

 

[Brian_B]

Qin = Qout (First Law of Thermo)

So heat in the blocks (the TDP of the cooled components) = heat rejected by the rads to the air.

For each component:

Q = m * c * dT

Q = energy (in watts, or BTU, or Joules depending on your unit of choice)

m = mass flow rate (not the same as volumetric flow rate which we've all been talking about here... m = density * volumetric flow rate) Water, in this temperature range, has a density of about 8.3 lbs/gal.

c = specific heat capacity (for water, it's 1 BTU/lb-F)

dT = differential temperature ( ABS(Inlet - Outlet) ) for the particular flow path you are running the calculation for.

For components with multiple flow paths: radiator for example, you have one of these equations for each flow path, and the energy transfer (Q) is equal across both sides. You would probably assume that the ambient temp stays constant (so that you have an ultimate heat sink, and that would be the temperature that system would ultimately reach quiescence with)

If you really wanted to get down into the weeds, you would also add the energy input from the pumps themselves (that 4gpm that you suspect you wouldn't get - your correct, and that energy that is lost as flow is gained as thermal input, but, if you actually run the numbers, you'll find it's pretty insignificant).

And if you want to, you can model the heat transfer across the blocks as well - it uses a similar equation as above, just use the heat transfer coefficient for copper and your preferred thermal compound), and instead of mass flow rate, it's surface contact area.

From there, you got all the data to do whatever calculations you please. That's thermodynamics 101. Draw your system out, set your ambient temp to something arbitrary, and run it through each component to find your ultimate outlet temp. In however level of detail you wish to go, to whatever decimal point you want to calculate it to.

Also, for your "extreme example" - that's pretty extreme right there... 4 SLI cards, 8 GPUs total, and an OCed CPU is pretty aggressive, and your likely powering that system on 220 V and assuming there is even a motherboard that could support it. You could see scenarios similar to that in a data center though, but there your looking at hard plumbed and actively cooled loops for that kind of energy density.

 

[Tsumi]

Not that I disagree with your content, but is "equilibration" a word? It's like "strategery", only more physicsier.

Sorry, the correct word is equilibrium. Equilibration buffer is a common term in the pharmaceutical industry (what I am in), so I mixed up the two.

Alright, I'll concede that I started it, and I apologize for not keeping it more civil.

I also concede that my example is an extreme case and does not qualify as a backing for general advice.

What I can tell you is that the system in question looks something like this, and this is the basis for my speculation regarding the loop order matter:

- 4930K, OC'd to ~4.3-4.4 GHz, approximately 200W TDP at typical operating voltage and temperature.
- 4x Titan Z's, approximately 1500W at factory stock clocks, roughly 65C operating temperature on average.
- Loop height: ~2ft
- 2x EK D5 Vario pumps operating at speed 3, running in series: https://www.ekwb.com/shop/ek-d5-vario-motor-laing-d5-vario-motor
- 1/2" ID tubing
- 3x Radiators: 1x 420x140mm, 2x 280x140mm

IIRC loop order was something like this: reservoir -> pumps -> 420 rad -> CPU block -> GPU blocks (semi-parallel) -> 280 rads in parallel -> reservoir

Going off of the above, the flow graph EK provides suggests 4 GPM for a single pump. The pumps sit directly next to each other at the bottom of the loop, so I'm not sure that there is a performance gain over 1 pump (I believe that's usually only the case when the pumps are spread out height-wise through the loop), the design choice was primarily for the sake of redundancy in case of a pump failure.

So for the sake of calculation, lets say we're working with 4 GPM base flow. I'm not 100% sure that you're still going to see 4 GPM by the time that fluid works its way upward through four dual-GPU blocks, but you're welcome to correct me.

Pumps in series increases head pressure but does not increase maximum flow rate. In a loop with no restriction, one pump will have an equal flow rate to two in series. Of course, watercooling loops are not no restriction loops, so that is how pumps in series add to flow.

D5 pumps are known as high flow low pressure pumps as compared to the other popular pump, the DDC. The DDC can produce more head pressure but has a lower maximum flow rate. The D5's flow rate would drop off fairly quickly with added restriction. Putting things in parallel reduce overall restriction and increases overall flow rates. I would not be surprised if your actual flow rates were 1/2 or 1/3 the rating with what you have.

It's impossible to accurately calculate without pressure-flow curves of every component, but a ballpark estimate of 1/2 the rated flow is pretty good, especially with 2 in series. At 2 gpm, you have 12 foot of head (pumps in series add head, pumps in parallel add flow), which is about 5 psi. CPU waterblocks have a pressure drop of 1-2.5 psi at 2 gpm, less than 1 for a nonrestrictive radiator. For parallel radiators, the pressure drop is around 0.25 psi for 1 gpm (flow split in 2), and parallel GPUs are around 0.5 to 1.0. Altogether, a 2 gpm flow rate seems plausible for your setup.

Loop height actually doesn't matter, and only loop length adds to the restriction. This is because of the closed system nature. The water going down the loop pulls the water going up the loop, so your net height change is 0. Essentially, it's the siphon principle.

Again though, as far as the thermodynamics equations go, it is as simple as what I listed above. This is because what goes in must come out. If the CPU was creating 200 watts of heat less than 200 watts of heat was going into the water (ignoring environmental gain/loss), the CPU temperature would rise because heat is being kept in the CPU. The specific heat of water is precisely known. The only variable is the flow rate of the system. This is classical Newtonian physics, not general relativity or quantum physics.

Also, apology accepted.

 

[Crackinjahcs]

I've always been curious about loop efficiency and planning. Nearly all the data I've seen only refers to component temperature and not coolant temperature. How much warmer is the coolant coming of the CPU block while under load, a couple degrees C or 8-10? What about for GPU blocks under load? Most multi-radiator loops I've seen run their rads in series, would there be any benefit in running a rad between the CPU and GPU (obviously not if the answer to my first question is only a degree or two)?

Total capacity of water cooling loops also only appears to be referenced by fin area in most cases and not coolant volume. Most recommendations I've seen were 120mm/140mm per component plus 120 mm if overclocking (fan/airflow vs. noise dependent, of course).

 

[Tsumi]

I've always been curious about loop efficiency and planning. Nearly all the data I've seen only refers to component temperature and not coolant temperature. How much warmer is the coolant coming of the CPU block while under load, a couple degrees C or 8-10? What about for GPU blocks under load? Most multi-radiator loops I've seen run their rads in series, would there be any benefit in running a rad between the CPU and GPU (obviously not if the answer to my first question is only a degree or two)?

Total capacity of water cooling loops also only appears to be referenced by fin area in most cases and not coolant volume. Most recommendations I've seen were 120mm/140mm per component plus 120 mm if overclocking (fan/airflow vs. noise dependent, of course).

I literally answered all of that already.

 

[Crackinjahcs]

I literally answered all of that already.

Sorry, I got distracted by the flow rate calculations and other bandy. I reread the thread.

I failed to see your 200W CPU calculations show that at 1gpm there is a delta of .76 degrees C.