Multi Pump Setups – Parallel vs. Series

Posted: April 26, 2011 in Pumps

While most veteran water coolers have come to accept that pumps in series is the preferred through practice, occasionally the question of series vs. parallel comes up in forum discussions.  There is quite a bit of good information out there in this regard for industrial pumps, yet I haven’t seen much documented on this question with actual testing in water cooling.  In an effort to provide another resource and to provide some testing to support theory and practice, I decided to test just that.  I proceeded to test two pumps in series and parallel and also evaluate the redundancy result if one pump stops for safety purposes.

Before going to far, I would like to thank my many sponsors including Koolance, XSPC, DIYINHK, and Bmaverick for the pumps and mods used in this evaluation.



To test and evaluate performances, I used conventional methods of testing dynamic head pressure vs flow rate each scenario.  This evaluates not only one restriction condition, but the entire range of conditions possible for a very complete look at pumping performance.  I utilized a Dwyer 477-5 digital manometer for pressure differential and King Instruments 7520 0-5gpm flow meter.  For voltage I used a Cen-tech P98674 measuring at the pump plug, and for amperage I used my Mastech HY3005D power supply amperage meter.  In this round I skipped the RPM monitoring due to the use of two different pumps.

Note: This test is for PUMPS ONLY!  Parallel vs. Series LOOPS is a whole different topic and I’m not testing that here.

There can be some benefit to parallel LOOPS under the right desired condition, but I wanted to look at pump setups under this test.  THIS IS PUMPS, NOT LOOPS..  Many people for example run multiple like GPU blocks in parallel to reduce the restriction caused by the GPU blocks and emphasize more of their pumping power on the CPU block. Nothing wrong with that, but that’s parallel “LOOPS”, and this test is “PUMPS” only, make sure you understand the difference.


I purposely chose to use two different pumps for a couple of reasons.  For starters, there is a “forum myth”, that you need to match two like pumps when run in the same loop.  I believe this grows from an assumption that the impeller speeds need to match and also because of the fear that one stronger pump would “Push” the weaker pump beyond it’s abilities.  Fortunately, this myth is wrong. As long as the net system flow rate does not exceed either one of the pump’s operating range, there is no problem at all with mixing different pumps.  Actually, I would recommend it over two of the same because they have different noise frequencies.  Due to this difference in noise output, the combined mixture of noise is smoother than putting two exact pumps together.  In addition, I have experienced RPM harmonics between two like pumps where if the RPMs are extremely close, but just slightly can get an undulating harmonic noise effect that can be very harsh.

I also chose two different pumps so I could more closely examine the result of pumps in series and parallel vs the single pump results.  In the end, the two mixed pumps worked perfectly together, and the test results using two different pumps did find some interesting details out for me particularly the parallel test results.

Moving on, here are the individual pump curves previously tested:

PMP-450 Detail

DDC-1 Tested at 12V

DDC-1 Detail

Testing Pumps In Parallel

To connect the two pumps in parallel, I used some custom Y fittings fabricated out of copper 1/2″ pipe.  I made these myself to represent the best possible condition for parallel.  Actual usage with more conventional Y fittings would perform slightly worse than what was tested because of the added restriction.  You can see the setup below:

Parallel Testing In Progress

Parallel Detail

Only small gains for low restriction setups, NOT RECOMMEND!

The result was somewhat as expected (Very Poor), but there are some interesting oddities.  When you run two pumps in parallel, the curves somewhat get stacked in the X or flow rate direction with an averaging of pressure.  Unfortunately this means the real gains of parallel don’t happen until you get beyond the useful restriction range of water cooling loops.  On a very high restriction loop you actually don’t gain anything at all.  The parallel curve crosses the single DDC curve at about the same point.  On a low restriction setup you would see some gains, roughly a 22% increase in pressure, but not at all what you might have hoped for.  Parallel simply doesn’t show benefits for the higher restriction levels that we typically see in water cooling and that holds true for pretty much any typical water cooling pump.

As you can see, it really doesn’t matter much which pump, the result simply doesn’t favor parallel pumps.

Beside the expected poor results, it was interesting that the parallel result is not simply adding the X direction as I thought.  Max head pressure for example is not the 6.5PSI of the DDC, instead it is an average of the two pumps(6.1PSI).  In the end it’s more complex than just adding the X direction, but the bottom line is that it’s not good and definitely not recommended. Parallel pump performance is poor for water cooling.

Parallel Pumps Redundancy Check

Besides performance, many folks use two pumps for redundancy purposes.  On occasion, pumps do fail and the idea is that if one fails, you’ll still have one moving adequate water for cooling.  Parallel is a bit unique and has raised quite a number of debates amongst forum members about this exact item.

What happens in parallel when one pump fails or quits for whatever reason?

To test this, I simply disconnected the PMP-450 and retested the parallel configuration.

Parallel One Pump Off Detail

DDC Single pump vs in a parallel setup with second pump off

Rather than the DDC pump pushing through the intended loop, it is now bypassing that intended loop via the PMP-450 subloop.  That’s bad new for parallel loops, as you lose not only the one pump that quit, but you also loose about 83% of the remaining pump’s power.

However….it does still have “Some” pumping power left over.  While it’s not going to be of any good levels, chances are your loop will still get between .3GPM and .8GPM which will function without causing catastrophic failure.  In the end…it will still serve a redundancy benefit..but barely.

Pages: 1 2 3

  1. Rori Woodman says:

    Hi Martin,

    Love reading your reviews, just wondering if you could answer a quick question for me?

    According to your 360 rad shoot out, the Black Ice GTX 360 has some pretty high restriction for a rad, I had read other reviews first and bought the Black Ice GTX 360 before coming across your review. I also bought the same rad, but a 480 to go in series with it, along with 2 EK HD 7970 water blocks, the Koolance 380i CPU block, a flow meter and a few elbows, not many as I have used the rigid acrylic tube and made most of the bends.

    My question is will the XSPC Twin D5 Dual Bay Reservoir with obviously the two D5′s in series be enough to achieve a minimum of 1gpm through the loop? Whats your thoughts? I think there would roughly be 2.5m of tubing all up. Not sure if its important, but the tube size is 3/8 ID and 1/2 OD.

    Thanks and hope to hear from you soon!

    • Martinm210 says:

      Yeah, no problem. One D5 is really plenty for most loops. Even if you drop below 1GPM, it’s generally more than enough pumping power. Some of the AIO kits like the H220 max out around .7GPM, I consider .5gpm the minimum for decent self bleeding performance, 1GPM if you want to be sure it bleeds itself easily, but thermal performance really doesn’t start dropping off fast until you get down in the .3GPM range.

  2. Mafieuso says:

    Hi Martin!

    You have a really nice blog and all the test and experiments are very useful.

    I would like to add that when you put 2 pumps in parallel you always have to put a check valve at least on the output of the weakest pump! It is the reason why the results are “strange”, when you have a pump with a higher head than the other you might have negative flowrate in th weakest pump!

    The theory is: 2 pumps in parallel are use to increase the total flowrate but not the head (you add the flowrate for a constant head).

    Finally, in parallel it’s always better to have 2 indentical pumps.

  3. Kevin Jin says:


    Love your pose, read it twice.
    Based on your articular 2 mixed pump in a serial loop works great?
    It would not create convection?
    How do you calculate the flow rate with two mixed pump in a single loop?