CPU and GPU in parallel?

madbrayniak · Jan 25, 2017

Hey all,

Still learning watercooling and just put in my first loop. However, I am already thirsty for more!

I am wanting to run my CPU and GPU in parallel and wanted to know if anyone knew if this could cause any problems? I invision this as simply having a "T" block or Q between the two on both the feeding and return lines. My hope is that this would lead to better temps and also a clean looking setup.

I am also looking into trying to use QDCs so that I can have the loop quickly unhook when I slide the motherboard tray out of the M8 and also quickly hook back up when doing any maintenance.

billbartuska · Jan 25, 2017

What percentage of the total water flow will go through each of the parallel legs? You don't know because it depends on how much restriction the CPU and GPU blocks have.

Also you get much less flow through each parallel part of the loop. That's an issue because there will be much less turbulence in each block and hence less heat transfer.

Watch out for QDCs, some are very restrictive.

steadly2004 · Jan 25, 2017

What he said.... lol

Most people use the loop in series... that allows for the lowest temps on the CPU which needs to be lower in temp to be stable to happen first, then the GPU which can run stable easier next. They also seem to be lower temps in custom loops easier for some reason (I don't know why, but it was in my first loop involving a GPU).

But.. also if your pump is strong enough it "may" provide enough flow to each component to work, especially if you do some research and figure out similar restrictions for each block involved. You may also need beefier pump. It could work, but requires syncronized restrictions to flow properly. If not, I imagine you find that one is hotter than the other because of an imbalanced flow rate, perhaps you can add a restriction to the second more free flowing piece. However this goes against most theory which involves removing as much restriction as possible allowing for lower temps.

"IF" both blocks provide a lower restriction than the radiator and the radiator is placed later in the loop, then the restriction of the radiator is going to be divided equally or unequally over the prior blocks. That's another reason why ultimate equality in restriction is necessary and is difficult to predict.

billbartuska · Jan 25, 2017

Quote:

Originally Posted by steadly2004

"IF" both blocks provide a lower restriction than the radiator and the radiator is placed later in the loop, then the restriction of the radiator is going to be divided equally or unequally over the prior blocks. That's another reason why ultimate equality in restriction is necessary and is difficult to predict.

No. Any restriction the water "sees" is caused by what it is flowing through, not what is upstream, downstream, or parallel to that restriction. But the total restriction the water "sees" is the sum of all the restrictions it "sees" in the loop.

It's a different way of looking at it but that's the way your god made hydrodynamics work.

toolmaker03 · Jan 25, 2017

Quote:

Originally Posted by billbartuska

Also you get much less flow through each parallel part of the loop. That's an issue because there will be much less turbulence in each block and hence less heat transfer.

Watch out for QDCs, some are very restrictive.

this is not right at all.

let me give you a simple example of the difference between a serial water loop and a parallel.

first thing I need to know when building a parallel water loop, is the flow rates of each component that will be on the water loop. I determine this by building a simple test loop with only a pump, a flow meter, and the component being tested.

for example lets say that the flow rates of these components are as follows.

CPU water block 4.0LPM
GPU water block 3.5LPM
radiator1 6.0LPM
radiator2 6.0LPM
reservoir 12.0LPM

with these components connected in a serial water loop the fastest the water loop can flow is 3.5LPM or the flow rate of the most restrictive component on the water loop.

with these components connected in parallel it is possible to achieve the high flow rate for each component in the water loop.
meaning that the flow rate on the CPU can be 4.0LPM, while at the same time the flow rate on the GPU is 3.5LPM, assuming that the total system flow rate is at 7.5LPM.

http://www.overclock.net/t/1573189/serial-vs-parallel-9-6lpm
http://www.overclock.net/t/1615072/cpu-and-radiator-upgrade-on-water-cooled-rig
http://www.overclock.net/t/1584867/tec-chill-box-chamber-build-log

I build all of my water loops about the same way, and they have flow rates from 9.0LPM to 12.0LPM.

steadly2004 · Jan 25, 2017

Quote:

Originally Posted by billbartuska

No. Any restriction the water "sees" is caused by what it is flowing through, not what is upstream, downstream, or parallel to that restriction. But the total restriction the water "sees" is the sum of all the restrictions it "sees" in the loop.

It's a different way of looking at it but that's the way your god made hydrodynamics work.

You are correct sir... I mis-spoke. I means if the downstream is more restrictive than the blocks..... The what's left will be unevenly distributed with less flow than without it. My bad. But I see what you're saying.

Sounds like you might me able to help me understand if I'm just wrong.
Hypothetically.....
If the radiator flows 4 "units", and one block flows 2 and another flows 3. Will the 2 not get a different flow than the 3?...Not speaking about the total flow? I understand the total flow will be restricted to the 4. Do liquids not flow to the path of least restriction? Wouldn't the block that flow 2 get less cool water than the clock that flows 3?

Basically if a flow is restricted to 4 "lanes" after the blocks and those 4 lanes are distributed over 5 prempting lanes. Then the 2 lanes will get 2/5 of the 4 lanes and the. 3 lanes will get 3/5 of the 4 lanes.

geriatricpollywog · Jan 25, 2017

Is there any compelling reason to go parallel vs series?

I'm thinking parallel would only help if your delta T is under 5 and you are worried about the CPU heating the GPU. Or if your flow rate is so high that the friction of the flowing water causes it to heat.

toolmaker03 · Jan 25, 2017

Quote:

Originally Posted by 0451

Is there any compelling reason to go parallel vs series?

I'm thinking parallel would only help if your delta T is under 5 and you are worried about the CPU heating the GPU. Or if your flow rate is so high that the friction of the flowing water causes it to heat.

yea, it's possible to lower the load temperatures on the components by 8C or more with the same amount of radiator space.

http://www.overclock.net/t/1573189/serial-vs-parallel-9-6lpm

that's what I tried to show with this build. the difference between the load temperatures from the serial loop water loop and the parallel water loop with two pumps is 8C I did not add any radiators to the build or change the fans. all I did was change the configuration of the build, and increase the flow rate on the water loop by adding a second pump to support the hardware.

geriatricpollywog · Jan 25, 2017

Quote:

Originally Posted by toolmaker03

yea, it's possible to lower the load temperatures on the components by 8C or more with the same amount of radiator space.

http://www.overclock.net/t/1573189/serial-vs-parallel-9-6lpm

that's what I tried to show with this build. the difference between the load temperatures from the serial loop water loop and the parallel water loop with two pumps is 8C I did not add any radiators to the build or change the fans. all I did was change the configuration of the build, and increase the flow rate on the water loop by adding a second pump to support the hardware.

Hmm I might benefit from this. I have 2 D5 pumps already. I might be asking you some questions after I buy more tubing.

billbartuska · Jan 25, 2017

Quote:

Originally Posted by steadly2004

Quote:

Originally Posted by billbartuska

No. Any restriction the water "sees" is caused by what it is flowing through, not what is upstream, downstream, or parallel to that restriction. But the total restriction the water "sees" is the sum of all the restrictions it "sees" in the loop.

It's a different way of looking at it but that's the way your god made hydrodynamics work.

Click to expand...

You are correct sir... I mis-spoke. I means if the downstream is more restrictive than the blocks..... The what's left will be unevenly distributed with less flow than without it. My bad. But I see what you're saying.

Sounds like you might me able to help me understand if I'm just wrong.
Hypothetically.....
If the radiator flows 4 "units", and one block flows 2 and another flows 3. Will the 2 not get a different flow than the 3?...Not speaking about the total flow? I understand the total flow will be restricted to the 4. Do liquids not flow to the path of least restriction? Wouldn't the block that flow 2 get less cool water than the clock that flows 3?

Basically if a flow is restricted to 4 "lanes" after the blocks and those 4 lanes are distributed over 5 prempting lanes. Then the 2 lanes will get 2/5 of the 4 lanes and the. 3 lanes will get 3/5 of the 4 lanes.

Let's try this:

The race track effect

No matter what the flow rate is the water spends exactly the same amount on time in any component in a loop.

Consider a race track that is a one mile oval and has two lines painted on it 88 feet apart.
A car going 60 mph around the track spends one second between the lines each lap.
A car going 120 mph around the track spends one half second between the lines each lap.
But in two laps both cars spend the exact same amount of time between the lines because the second car gets there twice as often.

Sooooo:

Why does higher flow rates in a loop yield better cooling? There are two reasons:

1.) At the higher flow rate the temperature differences of the water between entering and leaving the block (or rad) are closer, so there's a greater Delta T between the block (or rad) and the water. This results in more heat transfer than if the water was moving slower.

2.) At higher flow rates there is more turbulence than at lower flow rates resulting in better heat transfer. (We could get into laminar flow vs non laminar flow, but that's not necessary here)

Soooooooooooooooo.

How does this effect serial vs parallel?

While there may be an overall flow rate increase with a parallel loop, at least one block (or rad) will have a lower flow rate, most likely they both will.

Perhaps these pictures will explain:

edit: Turbbulence is spelled wrong in the pic, but I'm to lazy to fix that.

madbrayniak · Jan 25, 2017

very interesting stuff guys!

From what I am reading it is sounding like the idea I had for a "T" block might not work very well.

The majority of parallel setups that I have seen are like the one that Toolmaker showed in his build.

The reason I wanted the "T" was because I wanted to have the coolant enter the the blocks below the GPU and above the CPU. I have a reverse MOBO layout because I am hoping that such a setup helps reduce the risks with watercooling in case the GPU block ever leaks.

Another thing that may be of useful information is that I am a single GPU kind of person. Going to SLI/Xfire setups cause just as many problems as they solve so I would rather save my money to get the next hot GPU than get a second of the one I have. However, I do think that going forward we are going t o start seeing more waterblocks on NVMe SSDs which I would be interested in.

I use the EK DDC 3.2 PWM pump so I have something that can do a bit more pressure and would like to add a second one for redundancy purposes as well as increase in performance.

geriatricpollywog · Jan 25, 2017

Quote:

Originally Posted by billbartuska

Let's try this:

The race track effect

No matter what the flow rate is the water spends exactly the same amount on time in any component in a loop.

Consider a race track that is a one mile oval and has two lines painted on it 88 feet apart.
A car going 60 mph around the track spends one second between the lines each lap.
A car going 120 mph around the track spends one half second between the lines each lap.
But in two laps both cars spend the exact same amount of time between the lines because the second car gets there twice as often.

Sooooo:

Why does higher flow rates in a loop yield better cooling? There are two reasons:

1.) At the higher flow rate the temperature differences of the water between entering and leaving the block (or rad) are closer, so there's a greater Delta T between the block (or rad) and the water. This results in more heat transfer than if the water was moving slower.

2.) At higher flow rates there is more turbulence than at lower flow rates resulting in better heat transfer. (We could get into laminar flow vs non laminar flow, but that's not necessary here)

Soooooooooooooooo.

How does this effect serial vs parallel?

While there may be an overall flow rate increase with a parallel loop, at least one block (or rad) will have a lower flow rate, most likely they both will.

Perhaps these pictures will explain:

edit: Turbbulence is spelled wrong in the pic, but I'm to lazy to fix that.

There are competing forces working against eachother.

Higher flow rate generates heat. When leak testing, I ran both my punps at 100% with the fans off. Within 20 minutes, my loop temp went from 18C to 25C due to the friction of the water flow.

At work, we have an RO water loop running at a high flow rate. The temperarure of the loop is 30C and the water is not heated.

I'm not trying to contradict your points. Its just a matter of whether the extra cooling efficiency of turbulence and flow rate is enough to outweigh thr increased temperature due to friction.

Running rhe loop in parallel makes sense because it creates a shorter path for the water and slows it down.

billbartuska · Jan 25, 2017

LOL...

There's no heat build up from friction. What you're seeing is heat dump form the pump(s), typically 10 - 30% of the wattage rating of the pump.

Most people wrongly don't consider that in their Delta T calculations.

Ashcroft · Jan 26, 2017

Yes, Bill is correct.

Friction is not the primary cause of heating at high flow. Water pumps are not all that energy efficient. If you compare the actual work done in the water they move vs power consumption then it works out that most of the power is actually waste heat that goes into the loop. A 24W pump can gnerate nearly 20W of heat.

On the Australian forums a guy set up his pump with a rad and res to flush the rad or something and just by having the fans turned off the fluid got very hot before he noticed. If the rad had not been there at all I would hate to think what could have happened.

Simple Parallel loops like the one talked about in the OP will work fine with most modern watercooling gear. Since accurate testing became more common and available the restriction of water blocks has evened to a great extent. It used to be the case that some European blocks were triple the restriction of most others and some CPU blocks could be half the restriction of the average.
Nowadays it is much more common for CPU blocks and GPU blocks to be very close in restriction so using them in parallel gives us very close to even flow.

There are problems with larger parallel setups though in that to get a decent flow rate through each block it requires ever higher flow in all of the loop that is not parallel. In a common 4 block serial system it is not unusual to get 4LPM or 1 GPM. If that system was setup is parallel it requires a total system flow rate of 16LPM just to get the same flow rate through each block and that just becomes impossible even when doubling or more the pump power.

madbrayniak · Jan 26, 2017

So with all this talk about how to do parallel correctly this raises another question.

Would it be best to have the "T" barb closer to the higher restriction waterblock to help drive the water into that block while also helping divert the water to the lower restriction blocks?

Just a nickname · Jan 26, 2017

I remember reading a while ago about this very subject (comes back time to time). The advantage is that you can get better temperature on the gpu. This issue is the loss of flow in both branches. There is a threshold where more flow results in minimal performance gain (see image). However, if the flow is too little, you will have a linear regime (image: beginning where x is small).

Edit: log function (that's what I was looking for).

SpringY1989 · Jan 26, 2017

What if you have 2 seperat loops for GPU and CPU would this change anything to why you should either choose parallel or series? Also if you have multiple graphics cards?

toolmaker03 · Jan 26, 2017

Quote:

Originally Posted by madbrayniak

So with all this talk about how to do parallel correctly this raises another question.

Would it be best to have the "T" barb closer to the higher restriction waterblock to help drive the water into that block while also helping divert the water to the lower restriction blocks?

you can put the T's wherever you would like to the key is getting the flow rate up enough to support the component water blocks. to do this you need to know what the flow rate through them is.

if your planning to place the components in a parallel configuration, than the radiators will need to be in parallel configuration as well. other wise the radiators in series will restrict the total flow rate of the water loop.

for example.
lets say the CPU has a flow rate of 4.0LPM, and the GPU has a flow rate of 4.0lpm, than the total flow rate on the water loop that you would be tiring to achieve is 8.0LPM.

the issue is that most radiators flow from 4.0LPM to 6.0LPM, so in order to get the flow rate on the system to 8.0LPM the radiators will need to be paralleled as well.

the radiators do not need to be the same, nor do the water blocks have to be the same. the only thing that really matters when building a parallel loop. is that the system has be configured in such a way, to support the total flow rates possible through the water blocks that are in a parallel configuration with each other.

billbartuska · Jan 26, 2017

Quote:

Originally Posted by toolmaker03

Quote:

Originally Posted by madbrayniak

So with all this talk about how to do parallel correctly this raises another question.

Would it be best to have the "T" barb closer to the higher restriction waterblock to help drive the water into that block while also helping divert the water to the lower restriction blocks?

Click to expand...

you can put the T's wherever you would like to the key is getting the flow rate up enough to support the component water blocks. to do this you need to know what the flow rate through them is.

if your planning to place the components in a parallel configuration, than the radiators will need to be in parallel configuration as well. other wise the radiators in series will restrict the total flow rate of the water loop.

for example.
lets say the CPU has a flow rate of 4.0LPM, and the GPU has a flow rate of 4.0lpm, than the total flow rate on the water loop that you would be tiring to achieve is 8.0LPM.

the issue is that most radiators flow from 4.0LPM to 6.0LPM, so in order to get the flow rate on the system to 8.0LPM the radiators will need to be paralleled as well.

the radiators do not need to be the same, nor do the water blocks have to be the same. the only thing that really matters when building a parallel loop. is that the system has be configured in such a way, to support the total flow rates possible through the water blocks that are in a parallel configuration with each other.

Yes, to size the pump you have to know the total flow rate you need.

Putting the rads in parallel just compounds the problem though. Not only do you have to know the flow rate through each rad too, you have to choose fans for each that will give you the air flow you need to remove the BTUs that each rad has to remove.

For a nOOb series is just sooooo much easier to figure and there less chance for getting a bad Delta T from to much guestimation.

There's no substitute for experience and I think most people get it figured out with their third try. But lotsa good info here......

madbrayniak · Jan 27, 2017

Please explain what you mean by 'radiators in parallel".

Not sure I understand this. I was going to have the loop go CPU+GPU >Rad>Res>Pump>CPU+GPU.

I am using the EK DDC 3.2 PWM which looks to be able to run 5LPM at about half speed.

Not seeing the flow rate of the WB for the EK 970 block but I didn't look long and may consider waiting until I upgrade my GPU again.

CPU block is Supremacy Evo which is advertised as being low restriction.

Rad is EK XE360 and looking to add another XE 360, XE 240, or PE series of same sizes.

CPU and GPU in parallel?

madbrayniak

billbartuska

steadly2004

billbartuska

toolmaker03

steadly2004

geriatricpollywog

toolmaker03

geriatricpollywog

billbartuska

madbrayniak

geriatricpollywog

billbartuska

Ashcroft

madbrayniak

Just a nickname

SpringY1989

toolmaker03

billbartuska

madbrayniak

Top Contributors this Month

Recommended Communities