Solved

Should i create 2 seperate water loop for multi GPU to get better overclock performance?

Hi all,

Below with the listed parts I've on-hand:



I'm planning to use 3 radiators & not able to make up my mind since I'm not sure how much performance gain for 2 separate loop for GPU & CPU.

So should i go for 2 water loop in order to squeeze the best performance out of this build?

Estimate how much gain I'll have?

Also should i mix the nickel plated with the aluminum radiator since Mayhem concentrate contain chemical to prevent corrosion?
20 answers Last reply Best Answer
More about create seperate water loop multi gpu overclock performance
  1. I wanted to use 2 radiators and make a figure 8 single loop. Since the CPU and GPU don't normally run 100% at the same time i fgured which ever one was running harder would get the cooler water coming out of the other ones cooler. Just a thought. My idea was with a single loop any excess cooling capacity would be available. But with 2 radiators nothing would be in line behind the other. Both would draw straight from a cooler. 3 radiators? I can see why you're confused.
  2. the only practical reasons to do separate loops are:
    1. improve liquid flow if the pump is too weak.
    2. for the ease of swamping components.

    now to your loop. get rid of the aluminum rads or get aluminum everything else.
    Aluminum is not compatible with copper/brass. if you mix them in one loop, you will get galvanic corrosion very soon.

    once you figured out the above, for 2x980ti you need at least 4x120 rad surface (more is better) and about 2x120 rad surface for CPU loop. The rads are used more efficiently in a single loop.
  3. if you have a single loop with 3 components and a single radiator then 2 of the components will be getting heated coolant from the one(s) ahead of them in the loop.
  4. ^ that's very untrue. in a loop with the flow of 1gallon per minute or more, the difference between hottest and coldest spot of the loop is about 1C. components order really does not matter.
  5. What I said is definitely TRUE. It may not "matter" if the loop is as efficient as you say it is. But it is True.
    Moron this.
    http://www.overclockers.com/guide-deltat-water-cooling/
  6. There has to be a temperature rise somewhere of there would be no cooling. In the single loop the temperature of the whole loop will go higher if you add components to the loop. The GPUs can run at higher temp. than the CPU, and they(2) produce more heat. So putting them on a 2nd loop with a bigger radiator there is the best solution.
  7. n0ns3ns3 said:
    the only practical reasons to do separate loops are:
    1. improve liquid flow if the pump is too weak.
    2. for the ease of swamping components.

    now to your loop. get rid of the aluminum rads or get aluminum everything else.
    Aluminum is not compatible with copper/brass. if you mix them in one loop, you will get galvanic corrosion very soon.

    once you figured out the above, for 2x980ti you need at least 4x120 rad surface (more is better) and about 2x120 rad surface for CPU loop. The rads are used more efficiently in a single loop.


    OP, I would follow this advice. Also, Aurora is a show coolant and probably is not the best choice for daily use, especially if this is your first loop.
    https://mayhems.co.uk/mayhems-aurora-guide/
  8. The concept of delta-T and liquid temps is that the loop is always running at a load/equilibrium state. There aren't really sections where the coolant temps spike heavily between one or another...and if you are seeing this (which would require you to actually have temp probes on the coolant flow) you have much bigger issues than you're discussing here.
  9. After a certain point, more radiator surface does not equal "best performance".
    Only if your parts are actually hitting their thermal limit.

    The performance difference between 60C and 55C is exactly squat.
  10. The Delta T has to happen or you don't have cooling. It happens at the radiator water to air interface. This makes airflow very important. In a single loop you will have to cool everything down to CPU temperature limits. This will require more radiator surface and more airflow than if you looped the CPU and GPUs separately. If the GPUs are looped alone they can run at their higher limit which requires less cooling. The CPU can then be cooled to it's lower limit. The 1*C Delta in the coolant loop makes this impossible in a single loop.
  11. I'll try to explain. when your CPU is at full load and runs let's say 75C, the coolant will be about 40C and the air temp would be 25C.
    So the delta T is there and there is nothing wrong with it (component - coolant, coolant - air are way more than 1C). and yet, since in a "healthy" loop the coolant moves fast - really fast, the temperature of the coolant is almost the same across the whole loop. If you slow down the flow, the differences might get larger, but that makes cooling much less efficient. It's (not so) simple thermodynamics.
    So if you have a loop with CPU and 2 GPUs with let's say 6x120 rad surface, it will cool all the components same or better than if you split it to 2x120 and and 4x120. And the reason for that is simple - a rad can dissipate a certain amount of heat with some airflow at a given delta T (coolant - air). The cooling capacity of the rad scales linearly with the surface - read double the rad size, you can handle double the heat at the same airflow and delta-T. And the components (CPU and GPU) also has this parameter of how much heat they can transfer to the loop at specific delta T (component - cold plate of the block). That's why it's completely normal to see different components at different temps (for example CPU at 75C and GPU at 45C in the same loop).
  12. There are a few things that contribute to delta-T of a loop:

    thermal load
    coolant flow rate
    air flow rate over heat exchangers
    surface area of heat exchangers

    Also, since no one else has addressed it, I would vote 'no' on using the aluminum radiator if possible. You're still looking at a 240+360 radiator in terms of surface area.

    Also, when you say 'best performance'...what performance do you mean? Cooling performance? Overclocking performance? Running a single loop provides you with the best cooling ability out of the components, where if you split up the radiators, you cannot take advantage of all rads cooling all components.

    I would say 1 single loop, but you did not list your pump, which is rather important.

    Single loop, run your GPUs in parallel, not serial.
  13. krells said:
    n0ns3ns3 said:
    the only practical reasons to do separate loops are:
    1. improve liquid flow if the pump is too weak.
    2. for the ease of swamping components.

    now to your loop. get rid of the aluminum rads or get aluminum everything else.
    Aluminum is not compatible with copper/brass. if you mix them in one loop, you will get galvanic corrosion very soon.

    once you figured out the above, for 2x980ti you need at least 4x120 rad surface (more is better) and about 2x120 rad surface for CPU loop. The rads are used more efficiently in a single loop.


    OP, I would follow this advice. Also, Aurora is a show coolant and probably is not the best choice for daily use, especially if this is your first loop.
    https://mayhems.co.uk/mayhems-aurora-guide/



    Thanks for the info man, now i know why my previous Ryzen 7 build coolant all aurora pearl dissapear must have been stuck into the radiator...
  14. n0ns3ns3 said:
    I'll try to explain. when your CPU is at full load and runs let's say 75C, the coolant will be about 40C and the air temp would be 25C.
    So the delta T is there and there is nothing wrong with it (component - coolant, coolant - air are way more than 1C). and yet, since in a "healthy" loop the coolant moves fast - really fast, the temperature of the coolant is almost the same across the whole loop. If you slow down the flow, the differences might get larger, but that makes cooling much less efficient. It's (not so) simple thermodynamics.
    So if you have a loop with CPU and 2 GPUs with let's say 6x120 rad surface, it will cool all the components same or better than if you split it to 2x120 and and 4x120. And the reason for that is simple - a rad can dissipate a certain amount of heat with some airflow at a given delta T (coolant - air). The cooling capacity of the rad scales linearly with the surface - read double the rad size, you can handle double the heat at the same airflow and delta-T. And the components (CPU and GPU) also has this parameter of how much heat they can transfer to the loop at specific delta T (component - cold plate of the block). That's why it's completely normal to see different components at different temps (for example CPU at 75C and GPU at 45C in the same loop).


    Thanks Will & Nonsence & Rubix for the details.
    Well i just notice that i mistaken some info whereby I'm having 2 x 360mm rad + 1 x 120mm rad, basically I'll put it like back 120mm + top 360mm + front 360mm.

    But still I'm not comfortable to put a 360mm rad at the front reason being currently I'm having a Ryzen 7 build which having the front with a 360mm rad as well. I i notice the air it pulls into the case is fairly warm which bugs me since I'm afraid it will cause some bad temp to my components (M.2 SSD especially). Take note my environment temperature is around 29 degree Celsius - 34 degree Celsius since i don't have any aircond at my place.

    So now let's get back to the suggestion you guys provided:


    Sample of my existing Ryzen 7 loop (All Aurora pearl gone & disappear)


    7700K Build my 1st hard tubing build with back 120mm rad + top 360mm rad + side 240mm rad
  15. Best answer
    Don't worry about the air intake through the rad. It will raise the internal temperature of the case but not by much. And SSDs are funny - the controller on SSD does not like to get hot, but the memory chips are actually do much better and live longer at higher temps. there are heatsinks for SSDs if you are still worried
    I also live with ~33C room temp for the most part of the year and both front and top rad are pushing air into the case - everything is perfectly fine.
    The best thigh would be to have them all pulling air out of the case, but that's impossible in my case due to very limited space.
    The push/pull thing is really pointless/wasteful on thin rads with low fin density. So if you are going to use 30-40mm thick rads with ~15FPI, just use pull.
    the reason for pull is ease of cleaning. In push config (or push/pull) , the dirt/fur/dust accumulates between the pushing fan and the rad. eventually the airflow gets blocked. and you will have to disassemble the fan to clean it. it might require to take the radiator out of the case.

    about two loops, i'd still say you'd be better with single loop if you care about the noise. 360 is a bit too much for a CPU while 360+120 is barely enough for 2x980ti. the math is simple every 120 rad section is providing about ~150w heat dissipation at ~2000RPM of decent fans. Overclocked 980Ti is easily over 300watts under load. so as you can see, 360+120 for GPUs is kinda to little and very loud. CPU will need about 1/3 to 2/3 of the 360 rad depending on the load. in a single loop the 360 rad will be mostly used to help cooling GPUs (while gaming for example) and will be mostly used to cool the CPU while doing something CPU intensive. I can't thing of real world scenario where both CPU and GPUs are getting to full power draw simultaneously.

    Check if you can use radiators for 140mm fans. 280 is roughly the same performance as 360. if you can use even one 420 (3x140) radiator, it will significantly boost the cooling capacity. or consider the single loop.
  16. I still think a single loop is the best idea in this scenario. There isn't much of a case that tells me that 2 loops are definitively better than just running 1 for this setup. I ran a highly overclocked Q6600 + SLI GPUs for years on a single loop with 2x360 rads and the loop ran great with a single D5 pump. What most people don't realize is that if you can remove as much restriction from your loops as possible, you can run a single large loop with good flow rates and great temps if you do it correctly. Flow rate of coolant is a huge factor, which is why many people end up splitting loops up - it is just 'easier' to build 2 loops rather than designing 'one' well.
  17. Thanks again nonsense & rubix!

    I've make up my mind whereby:
    - Radiators position wise will be back top & front
    - Fans setup will be all push since i don't want to waste my fans RGB lightning
    - Well i do vacuum clean those filters weekly from the PC front top also inner portion
    - 2 GPU will use back & top rad whereby CPU will use front rad since it produce lesser heat compare with 2 GPU
    - Purely planning to re-use whatever on-hand so no 140mm or 280mm or 420mm stuff for me (Limiting my budget)
    - Will plan with a single loop & try to route the best & simplest route i can
    - Current route in mind will be Reservoir Pump -> GPU2 -> GPU1 -> 120mm rad -> 360mm rad -> CPU -> 360mm rad -> Reservoir Pump
  18. your plan sounds reasonable.
    few comments:
    1. no matter how often you clean the filters, dust will get in and after few months there will be dust buildup between fans and rads.
    2. loop order, it seems like moving the CPU right after GPUs will be good idea. the reason is that you might want to avoid going up and down with the loop. this will make filling and draining the loop much easier. think about it as if you open the drain port/valve, you want liquid to get out on it's own as much as possible.
  19. n0ns3ns3 said:
    your plan sounds reasonable.
    few comments:
    1. no matter how often you clean the filters, dust will get in and after few months there will be dust buildup between fans and rads.
    2. loop order, it seems like moving the CPU right after GPUs will be good idea. the reason is that you might want to avoid going up and down with the loop. this will make filling and draining the loop much easier. think about it as if you open the drain port/valve, you want liquid to get out on it's own as much as possible.


    1. Yup you’re right so I can’t escape my annual water cooling maintenance
  20. unless you figure out how to setup the fans as pull and yet have the RGB fun :)
Ask a new question

Read More

Overclocking Water Cooling Intel i7 CPUs GPUs Performance