Multi Pump Setups - Parallel vs Series Test Results

[Martinm210]

Detailed Blog/Review Link

Nothing really new for the majority here on performance, but I have heard the debate more than once on the redundancy of pumps piece. I tested both performance in operation of both pumps and with one pump off for your viewing pleasure..

A special thanks to Koolance, XSPC, DIYINHK, and Bmaverick for sponsoring the pumping parts...without their support this test would not have happened..

Summary

Parallel





Redundancy check


Series





Redundancy check


Conclusion

Parallel pump setups are very clearly a poor choice for water cooling. Not only is the performance benefit very small and only some benefit for very low restriction loops, it also barely manages to provide enough pumping power for redundancy. Series setups give you double the head pressure of a single pump and that increased head pressure is tuned ideally for water cooling’s high restriction nature. I will caution that running two pumps in series and the resulting increased flow rate does work your pumps harder, so you should consider additional cooling particularly with DDC pumps. I would also advise that dual pumps is not something that will normally net large measurable temperature gains, it’s generally reserved for those that are in search of that last degree in performance…or for those with extremely large, complex, and restrictive loops.



Bottom line….Use Series setups in multiple pump configurations, do not use parallel pump setups. There is really no advantage to using parallel over series, series is always better for watercooling.

Of coarse this is assuming you are talking simple series type loops. If you start messing around with parallel loops and parallel pumps things could start looking a bit different since you are altering the restriction scenarios. Regardless, I still doubt there is any benefit to going parallel on pumps in just about any condition. For simplicity, I've stuck with the basics on this and hopefully it's of some use for new users on the fence about the two options.

Cheers!
Martin

[Utnorris]

Nice. Any chance of three pumps? I am running two D5's in series on a single loop with EK HF, EK HD6790 x 2, Koolance Formula FC block, 3 x PA120.3 and XSPC bay res and with two in series I get around 1.04GPM. I am thinking of adding 1 or 2 more for either 3 or four in a series to get more flow. Any ideas? Also, can you compare two DDC's versus two D5's?

[Martinm210]

It'll work exactly the same regarding the adding of pressure. If a pump produces 4psi at 1.5gpm:

2 in series = 8psi
3 in series = 12psi
4 in series = 16psi

Flow rate increases will be diminishing, something like 30% more, 15%, 7% etc..

Temp gains even more so with a loss as heat dump becomes more damaging than the flow rate gain.

What particulars are you still looking for?

Are you just looking for which performs better DDCs vs D5s?

[kroks]

ah finally someone tested the redundancy of pumps
nice!

[NEOAethyr]

Cool nice thread .

Can you possibly do this test with 2 rads?
I'de like to see the results of 2 rads in parallel compared to a single rad.
My theory, if the tubing is done with the least restriction possible, is that 2 rads in parallel will beat a single rad in terms of restriction.

But it'll probably take more work to get the air out of it when setting it up.

The idea came to me when I realized that the prob with most loops is that there gets to be to much restriction.
So you need a stronger pump.
But if you cut that restriction in half then perhaps it could work out better.
Plus it would store alot more water without the need of a reservoir.

Granted you wouldn't bother doing this if you had a 3x120mm rad, but my idea was to use multiple 1x120mm rads instead.


Edit:
On your series config, it might be possible to get more out of it if you lower the voltage on the 1st pump.
Just a theory.

[Johnny87au]

Nice post martin always knew it though hehe, so 4psi inbetween each pump you add? Also could you test certain GPU blocks ? I know the swiftech blocks must run in paralell..

[miahallen]

Very interesting stuff Martin

I like your commentary about using two different pump types in a single loop...however, I think your points would only be valid in a series configuration. In the parallel setup, I think using two different pumps would skew your results, and show significantly worse performance than two identical pumps. They would constantly be fighting with each other.

[Church]

Martin: can you retest parallel pumps with one pump switched off and put flow meter in that "dead" pump "subloop"? It kind of interests me how much of flow goes there instead of going whole loop route, is it reverse proportional to flow restriction of whole loop's waterblocks vs restriction of stopped pump?

[Utnorris]

It'll work exactly the same regarding the adding of pressure. If a pump produces 4psi at 1.5gpm:

2 in series = 8psi
3 in series = 12psi
4 in series = 16psi

Flow rate increases will be diminishing, something like 30% more, 15%, 7% etc..

Temp gains even more so with a loss as heat dump becomes more damaging than the flow rate gain.

What particulars are you still looking for?

Are you just looking for which performs better DDCs vs D5s?

Yes. and obviously the DDC's would have after market tops, either per pump or a dual/triple top. I have both, a set of DDC's and a set of D5's (two more on the way, got a deal on them) so I am wondering if I put the two DDC's in series with tops if they will perform better than my two D5's in series.

[Church]

Utnorris: in average generic loop two DDCs most probably will slightly win over two D5 and slightly loose to two D5 strong (if only pure performance at fullspeed is considered). D5 migt be a little notch quieter at full speed. Both setups should give more then enough performance to not worth the upgrade, so you might look on 2ndary differences, which to you might matter more, like that slight performance win or slight noise level win or built in speed controller or overall size/dimensions and thus ease of mount, or max power pump uses or is in specs for controllers you might use for them .. all of that is up to specifics of your preferences/tastes and specifics of your build.

[antiacid]

thanks for confirming what I suspected: serial is better than parallel for a single loop. However, personally, I like having 2 loops (gpu split from cpu/board) because of the workloads that I have. This way, I can start only the radiators in the loop that's actually working and save on noise when I'm not gaming, for example. Also, tubing all that crap into a single loop would be a lot more challenging in my MM case

[Interitus]

Martin,

What kind of cooling would you consider adequate with 2 355's in series? I'm building my first loop soon and will be running 2 355's in series with the EK dual top. I wasn't entirely sure that I would need the extra head pressure, it's more of a redundancy thing as I just wanted the extra safety net for my first loop.

I had planned on mounting my DDC's on L-brackets, keeping the pumps about an inch or two off of the case floor. They would also be mounted right in front of a Yate Loon Medium 120 that would probably run about 1000rpm.

Would that be a good enough cooling scenario in your opinion?

In case it's relevant, here's my parts list:

EK Supreme HF Nickel
Koolance VID-AR587 (x2 eventually)
Enzotech EV-X58 MB block
2x Swiftech MCR320
EK Dual Serial DDC top
2x MCP355's
Iandh Stealth 225 Res

I have a handful of 45's and 90's, gonna try to use them sparingly, but would probably have a mininum of 4-5 45's and 90's.

[bing]

Thanks, this is a really good thread.

[Serpentarius]

It'll work exactly the same regarding the adding of pressure. If a pump produces 4psi at 1.5gpm:

2 in series = 8psi
3 in series = 12psi
4 in series = 16psi

Flow rate increases will be diminishing, something like 30% more, 15%, 7% etc..

Thank YOU, Martin

what's gonna happened if we mix (2 series)x(2 parallel)?
or maybe (2 parallel)x(1 series)
i bet the chart going haywire lol

[Halo_003]

Martin,
I had planned on mounting my DDC's on L-brackets, keeping the pumps about an inch or two off of the case floor. They would also be mounted right in front of a Yate Loon Medium 120 that would probably run about 1000rpm.

That should be fine, the pumps do give off a little heat but not that much. My 18W DDC is mounted on top of a Cooler Master R4 and it's barely even warm to the touch. Really a pump is fine without active cooling based on my experience, you just never insulate a pump.

[Martinm210]

I'll run one more test with like pumps, but I'm sure it's not going to help any. The only difference will be under extremely high restriction it should match max head of one pump. With the slightly different max head of the two it averages them because a small amount starts going backwards on the weaker pump. Parallel is still going to do poor, no matter how you mix it up.

The better choice will still be series in watercooling unless you were modifying the restriction through multiple subloops in parallel to an extreme amount for very very very very low restriction, but it would require signifant up front testing and evaluation to net any benefit. Why bother when series is so simple and works without all the planning and for redundancy.

You need to think of pumps and loops separately though, this test was for parallel pumps, not parallel loops.

Parallel loops do have a place in watercooling (MB blocks and other low priority/high restriction items). I just don't see where parallel pumps can be better than series. Aquariums and other really low restriction applications might see the opposite, but in general watercooling demands pumping that is tuned for high restriction which is very fitting of series pump installs.

I haven't found a scenario where parallel pumps would be the better choice..the restriction is always too much..

[Manicdan]

Martin's testing methodology is so awesome we need him to start reviewing real problems, like if global warming is real, and pro-life vs pro-choice....
....after hes done with water cooling

[Alexandr0s]

Great review Martin, thanks !

I'd like to make a suggestion for a test if that's possible, because I already asked this question to Waterlogged, and he never got back to me.

According to the technical specs, the MCP350 does 24 ~ 26 dBA @ 2', and the MCP355 does 30 ~ 32 dBA @ 2'. Now my question was: What would be better performance vs. noise-wise. Because a doubling in noise means an increase of 3dBA if I remember correctly, so 2 MCP350's should do 27 ~ 29 dBA @ 2', but provide an increase in both flow and pressure, and should easily beat a MCP350. That's the theory of course . I can't remember if anyone has tested it, so I ask you, if you want to test it .

[stephenswiftech]

Pretty good work there, Martin.
Although it does not concern most (if not all) pumps that are currently used in the industry, I would like to add that the results (pumps in series being superior than in parallel) are the consequence of two things:

- the actual pump specifications: particularly their relatively high maximum discharge and their head pressure level. Let's imagine for a second that if a DDC could only do 1.5 GPM as maximum discharge, then putting them in parallel would be a much better choice than in series. So keep in mind that these results are not universal but dependent on the pumps specs. Different technology could imply different results - and we never know what will be coming next!

- the other thing is that no matter how restrictive a loop is, pressure drop always increases with the square of flow rate. In other words, no matter how low restriction a loop is, at 3 GPM and more, we will always face MAJOR pressure drop. This only means that the best thing one should do is to try and enhance pressure in the actual working range of flow rates (1 to 2.5 GPM - see my latest "article" about this) and for that, all the pumps having a max discharge > 3 GPM will be benefit much more from serialization.

[Church]

Taking in mind charts of block performance relative to flow, i'd reduce actual working range to 0.9-1.25gpm. One shouldn't care about getting even more flow nor do/change/spend/optimise anything for above that range. Too little returns, mostly wasted effort/funds.

[bmaverick]

Martin,

This testing is for parallel in the same loop correct?

What if a setup has two loops in parallel, one for a CPU loop and another for a GPU loop. Both loops pull coolant from a RES and return fluid to the same RES at the end of the loop. The reason I ask, that's my loop. Of course, my RES is a 1-gallon glass pickle jar.

Great work again for the A+ effort you put into these reviews!

[Martinm210]

Pretty good work there, Martin.
Although it does not concern most (if not all) pumps that are currently used in the industry, I would like to add that the results (pumps in series being superior than in parallel) are the consequence of two things:

- the actual pump specifications: particularly their relatively high maximum discharge and their head pressure level. Let's imagine for a second that if a DDC could only do 1.5 GPM as maximum discharge, then putting them in parallel would be a much better choice than in series. So keep in mind that these results are not universal but dependent on the pumps specs. Different technology could imply different results - and we never know what will be coming next!

- the other thing is that no matter how restrictive a loop is, pressure drop always increases with the square of flow rate. In other words, no matter how low restriction a loop is, at 3 GPM and more, we will always face MAJOR pressure drop. This only means that the best thing one should do is to try and enhance pressure in the actual working range of flow rates (1 to 2.5 GPM - see my latest "article" about this) and for that, all the pumps having a max discharge > 3 GPM will be benefit much more from serialization.

That makes sense, I suppose the closest thing to that would be a pair of DDC3.2s with factory tops. Maybe I'll use that for my retest..

I like your christmas tree better through..

That's awesome..

Source:
http://www.swiftech.com/stephens-corner.aspx#tab3

[Martinm210]

Martin,

This testing is for parallel in the same loop correct?

What if a setup has two loops in parallel, one for a CPU loop and another for a GPU loop. Both loops pull coolant from a RES and return fluid to the same RES at the end of the loop. The reason I ask, that's my loop. Of course, my RES is a 1-gallon glass pickle jar.

Great work again for the A+ effort you put into these reviews!

This is pumps only, loops is a whole different beast all together. You're basically running separate loops with a shared reservoir, nothing wrong with that. Also nothing wrong with going parallel on lower priority items as many do with GPU blocks etc.

This test was specific to pumps only.

[stephenswiftech]

That makes sense, I suppose the closest thing to that would be a pair of DDC3.2s with factory tops. Maybe I'll use that for my retest..

I like your christmas tree better through..

That's awesome..

ahah

if you've already defined/measured your "low restriction" loop then, just put the estimated graphs for 2x DDC1 in series and 2x DDC1 in //. The intersection is the precise sweet spot where Series equals Parallel.

If your "low restriction" loop is higher than this point, then Series is better.
If it's lower, then // is better.

[Martinm210]

ahah

if you've already defined/measured your "low restriction" loop then, just put the estimated graphs for 2x DDC1 in series and 2x DDC1 in //. The intersection is the precise sweet spot where Series equals Parallel.

If your "low restriction" loop is higher than this point, then Series is better.
If it's lower, then // is better.

Alright, the best possible condition I can think of for pumps I do have is two DDC-1 with OEM Laing Tops. Thanks to Bmaverick!

It's about as steep a pump curve as anything, and this is how the theoretical curves pan out. Of coarse this completely ignores friction losses in the fittings and my Y fittings are lower in restriction than anything you can buy practical. I suspect the fitting losses will degrade parallel to a point where even at extremely low restriction there is little value in the parallel setup. You have to also consider that in an extremely low restriction loop, most people wouldn't bother with two pumps (There is really no need). Also I think there is a fairly high percentage of users with DDC pumps that put aftermarket tops on them. Once you do that...you dramatically increase max flow rate which then strongly leans to series yet again.

Most people don't go to dual pumps unless there is a "High Restriction" reason, where series is still the better option. A single pump would produce 1GPM or better for average to low restriction loops. It's only under Very High to high restriction loops where flow rate could dip below 1GPM in which case series is the better option.

But....in spirit of science, I'll test this (DDC-1s with Laing Top) and see how the actual compares to theory. That's probably the one pump curve that would be the most promising in possibly showing some benefit to parallel in the extreme low restriction area. I just don't see the value even if the test performed as good as theory though since there is no need and questionable benefit for dual pumps under that condition.