How much pump is too much? How much is enough?

Introduction


I just wanted to investigate the phenomena of chasing ever larger pumps, and bring in some figures to explore just when "too much pump is bad".

First let's start with our challengers, all configured for in-line operation.

Eheim 1046, adds around 1.5W of heat to the loop
Eheim 1048, adds 3W of heat
Eheim 1250, adds 9W of heat
MCP600, adds 8W of heat
MCP650, adds 15W of heat
Iwaki MD-15R @ 60Hz, adds 22W of heat
Iwaki MD-20RZ @ 60Hz, adds 31W of heat
Iwaki MD-30RZ @ 50Hz, adds 50W of heat

Let's assume that we're watercooling an exceptionally hot CPU, that's putting 100W of heat into the water cooling loop under load.

Let's assume that we're using a Cascade waterblock, and for our CPU die it has a C/W curve relation that's roughly shaped like this:



Our theoretical loop consists of the Cascade block, 2 meters of 1/2" ID tubing, 1/2" barbs, and a Thermochill 120.2 radiator.

Against each of the above pumps, we arrive at the following flow charts:



For our radiator performance curve, we'll use the following graph, and use the pink line on the graph as befits two moderately powerful (35dBA) fans, being good performance while being fairly loud, but not insanely so.



Okay, so that's all the information we basically need to make our predictions.


Results


For each of the pumps we see that we'll get the following flow rates, and also the corresponding radiator C/W, and waterblock C/W at that flow rate:

Eheim 1046, 3.2 LPM, 0.045, 0.187
Eheim 1048, 4.2 LPM, 0.044, 0.178
Eheim 1250, 5.2 LPM, 0.043, 0.172
MCP600, 5.7 LPM, 0.042, 0.169
MCP650, 5.8 LPM, 0.042, 0.168
MD-15R, 6.7 LPM, 0.041, 0.165
MD-20RZ, 8.7 LPM, 0.040, 0.159
MD-30RZ, 10.0 LPM, 0.039, 0.157

Okay, so the CPU heats up by the CPU wattage dissipated by the waterblock's C/W at that flow rate. The water temperature rises above ambient by the radiator's C/W at whatever flow rate multiplied by the CPU wattage plus the pump heat wattage being added to the water.

This now gives us a predicted correlation for the relationship between the final CPU temperature as affected by the flow rate, but more importantly also after factoring in the pump heat.

Eheim 1046, WB delta = 100W * 0.187C/W = 18.7C, Water delta = (100 + 1.5)W * 0.045C/W = 4.6C, Total CPU temperature = 23.3C above ambient
Eheim 1048, WB delta = 17.8C, Water delta = 103 * 0.044 = 4.5C, Total CPU Temp = 22.3C above ambient
Eheim 1250, WB delta = 17.2C, Water delta = 109 * 0.043 = 4.7C, Total CPU Temp = 21.9C above ambient
MCP600, WB delta = 16.9C, Water delta = 108 * 0.042 = 4.5C, Total CPU Temp = 21.4C above ambient
MCP650, WB delta = 16.8C, Water delta = 115 * 0.042 = 4.8C, Total CPU Temp = 21.6C above ambient
MD-15R, WB delta = 16.5C, Water delta = 122 * 0.041 = 5.0C, Total CPU Temp = 21.5C above ambient
MD-20RZ, WB delta = 15.9C, Water delta = 131 * 0.040 = 5.2C, Total CPU Temp = 21.1C above ambient
MD-30RZ, WB delta = 15.7C, Water delta = 150 * 0.039 = 5.9C, Total CPU Temp = 21.6C above ambient


Discussion


Overall I think the patterns established speak for themselves. All of the high-flow focused pumps lose out significantly due to the wasted strength of their motors in comparison to the flow rates that can be realistically pushed. The high-head, low-moderate flow pumps (MCP600, MD-20RZ) perform the best as the ratio of motor heat to the final corresponding flow rate means that they are operating more efficiently for our proposed water-cooling scenario.

If we doubled the radiator effectiveness, which could only be achieved through exceedingly noisy fans, or using two such radiators (and a corresponding amount of fan noise), we get to the situation where the MCP600, 650, and the MD-15R all pretty much still fall on each other, the MD-20RZ pulls a slightly more significant lead, but the MD-30RZ still fails to catch the MD-20RZ while consuming a boat load of power in the process.

Of interest is the relative closeness of the Eheim 1048 and the Eheim 1250. If instead we chose a single 120mm radiator the difference between the two drops to 0.2C, or next to nothing, and so the Eheim 1048 would have to get the choice every time out of the Eheims, but the MCP600 still would be the first preference if its small amount of extra noise is not an issue. All of the other pumps become rather unattractive in a single 120mm radiator scenario where even the MD-20RZ should be avoided.

The Eheim 1046 is totally behind the knee of the curve in all respects.

So to answer: How much pump is enough, and how much pump is too much? I can make the following general recommendations:

Single 120mm radiator: First choice: Swiftech MCP600 or AquaXtreme 50Z. Silence Choice: Eheim 1048
Dual 120mm radiator: First choice: Iwaki MD-20RZ, Second choice MCP600. Silence Choice: Eheim 1048 or 1250
Larger radiators: Iwaki MD-20RZ all the way, with the MCP600 as a good second option.

I think it could be safe to say that if you didn't know what pump you need, and don't have the dollars to spring for the Iwaki MD-20RZ, then the Swiftech MCP600 (aka AquaXtreme 50Z) would have to be the single safest bet. For the performance freaks, the Iwaki MD-20RZ can't be overlooked. Avoid anything stronger than the 20-RZ though - it is a total waste - your CPU will be hotter - your water will be hotter - and you'll be sucking down more electricity while being worse off - there's no good news here at all. The Eheim 1048 puts in an admirable show, with the Eheim 1250 perhaps giving the best performance for low noise, and the Eheim 1046 should simply be avoided unless you have a specific need for it (low space and low noise).

The Iwaki MD-15R holds a bit of a no-mans-land position, being large, powerful and hot, but not really offering anything much to show for its go. Both the MCP650 and MCP600 are better, cooler, and less power hungry choices. The MCP650 fits the bill adequately as a very close runner up to the MCP600 in all scenarios, and is a worthy successor to the MCP600 once expected pump life-span is taken into the equation.


Summary & Conclusion


Recommended pumps:

Performance: Iwaki MD-20RZ
Jack-of-all-trades: Swiftech MCP600/AquaXtreme 50Z
Silence focused: Eheim 1048

Broad guidelines on pump selection:

• DO NOT select a pump that draws more than 50W of power. Performance will not improve further and the pump will be drawing more power than is needed, effectively being a waste of electricity which just shows up on your power bill. Pumps drawing much more power than 50W will actually make your CPU hotter.
• In general avoid pumps that put more than 16W of heat into the water per 120x120mm of radiator area (or equivalent). Match your pump to your available radiator capacity so that pump heat is not a major player in your water's temperature. In general, try to keep pump heat to radiator capacity down below 12W of pump heat per 120x120mm of radiator area.
• Try to choose a separated armature pump (spinning magnet around the impeller) as opposed to a canned motor (eg. Laing style), or electro-magnet motor (eg. Eheim style) as the separated armature design minimises motor heat transferral into the water
• Avoid pumps with less than 1.0mH2O of pumping pressure at 3.5LPM flow rate (or 3' of pressure @ 1gpm), unless you have a very specific need for some particular pump (space requirements). Such pumps are too weak and performance will suffer noticably.
• Try to avoid "high flow" pumps (pumps with >20LPM peak flow rates). Such pumps tend to have the wrong impeller design and flow characteristics for water-cooling use.
• Choose pumps with at least 6LPM of peak flow rate
• Choose pumps with at least 1.5mH2O (5') of peak pressure
• When trying to decide between two pumps where one pump has more peak flow than another pump, then so long as the rated peak flow rates are at or above 10LPM, always choose the pump that has the higher peak pressure rating, over the pump that has the higher peak flow rating. If the peak pressures are about equal, but both offer peak flow rates above 10LPM, choose the pump with the lower peak flow rate as it will add less heat to your system.

 

Information for 50Hz Iwaki MD-20RZ

The 50Hz Iwaki MD-20RZ was not shown on the above charts or maths. It works out as:

25W of added heat, 7.5LPM expected flow rate.

0.162C/W for waterblock, 0.040C/W for radiator.

MD-20RZ (50Hz), WB delta = 16.2C, Water delta = 125 * 0.040 = 5.0C, Total CPU Temp = 21.2C above ambient.

This still makes it the best pump for dual radiators and larger, and still is the largest pump you should choose of the Iwaki range, whether you have 50Hz or 60Hz power.

 

Thanks Cathar, that is the most pleasing thing a virgin watercooler who has just bought a 1048 wants to hear.

And im going for silence.

Your knowledge of this stuff astounds me, theoretical maths is a strength of mine but this is a really useful application.
Congrats.

 

2 x Swiftech MCP600's in series

I decided to crunch the math for this as an alternate performance oriented approach:

16W of added heat, 7.3LPM expected flow rate.

0.163C/W for waterblock, 0.041C/W for radiator.

2 x MCP600 in series @ 12.0v, WB delta = 16.3C, Water delta = 116 * 0.041 = 4.8C, Total CPU Temp = 21.1C above ambient.

So if a single Iwaki MD-20RZ blows the budget, but you're looking for something to get you by for now, and also to upgrade your system later, then a single MCP600 followed by a lter MCP600 is both cheaper than a single MD-20RZ, and offers the same level of cooling performance, and would be better across all scenarios.

So perhaps to extend the "Best Performance" category:

Best performance across the board: 2 x Swiftech MCP600/AquaXtreme 50Z.

Cheaper, less heat, and acts as a natural fail-safe if a single pump fails. Only compromise is that more space is needed, although probably not a lot more than a single Iwaki MD-20RZ.

 

um .....
is the 1250 much noisier than the 1048?
in my setup i'm lookin to cpu, gpu, nb and only a 120mm rad ... will the 1048 be enough? the 1250 is only $15 more ...
i'm not looking towards the swiftech ones at all, a mate had a few problems (i dont want any problems ...)

 

Cool.. Nice comparison..

So how does one go about finding the amount of heat a pump transfers to the fluid..?

I read that some ppl do a guestimation.. Is this like 95% or 50%..?
It looks like the MPC650 dumps around 60% of its rated wattage..

 

Thats a great demonstration Cathar, if its not too hard to include I am interested in where that little March pump fits

 

wow and I was about to trade in both my MCP600s for an Iwaki. Thanks for the money saving tip.

 

Great comparison Cathar... just what I was looking for in finalizing my pump decision.

The only good addition I could think would be to compare relative noise levels of each setup.

 

Thanks Cathar. Very informative.

My rig is exactly the same with the one you suggest for performance wise. That is: cascade, about 2m 1/2 tubing, Thermochill 120.2, Iwaki MD-20RZ (50Hz).

But, I got the question here. As my understanding, the heat the pump dumped into the loop is basically the fucntion of how much resistence the pump have to overcome to push the water. The higher the resistence is, the higher the heat will be added. Thus, I am wondered if the heat (25W) added into water by 50Hz version is measured on the same basis (same resistence in the loop). As you mentioned, after factoring the heat added into water by pump, the result will be greatly changed. Do you think I am correct to think like this?

Thanks for you reply.

 

thank you for that dose of common sense. I feel that you and the editor of overclockers.com would get along just fine. Perhaps an extension of this analysis ould be to consider which cpus benefit from watercooling.

For example, northwoods dont benefit near as much as prescotts. In the amd camp, i am running a 3200 newcastle in a dfi 250 gb mobo with 512 m bh5 pc3500 mushkin ram with the following results:

speed stock (11x200) mild oc(11x225) mod oc(10x240)
volts auto auto+4% 1.55+13%
air (xp120) 45*C idle/58 prime95 50/68 50/shutdown

stormG4 39*C idle/48 prime 95 41/55 41/61(only 1st test)

the max i can boot windows is 10.5x245 (though the ram will pass memtest+ v1.27 at the following speeds- 8x256 or 10x250 or 11x246). At this speed and at the moderate overclock above, it is only prime stable for the small FFT test for a couple of hours..My point is that to go above a mild overclock with this chip, one needs a lot more volts and the gain is 200 mhz with a loss in prime 95 stability. Water cooling makes the temperatures much lower but does not really allow a much greater overclock due TO OTHER LIMITATIONS.

Now, on my athlon xp mobile chip, that thing really loves the water because it is much more heat limited. I imagine a prescott would would also really benefit from a storm watercooler as well.

 

dual mcp600, i cant hear the pumps.

 

The E1250 does have a noticable rumble to it, unlike the E1048 which cannot be heard normally unless it's rattling against something. It depends on one's thirst for absolute silence. If, when you turn your computer on, you want to be able to NOT tell whether it's on or not, then the E1048 would be the better choice over the E1250.

I've measured most of the pumps there by working out how long the pumps took to heat up an insulated body of water, and approximated the results, probably within +/- 10% or so accuracy, which is still good enough for what we're working out above. For the Iwaki pumps I extrapolated from what I measured for my MD-30RZ (which I think is going to get relegated to acting as a pump on a rainwater tank for watering the gardens pretty soon).

Each pump type dumps differing levels of their power draw into the water as heat, and dump differing levels of their power draw out through the casing.

As soon as I generate a PQ curve for it, I'll include it. It is a little weaker than the MCP600 though, and adds about 6W of heat to the loop. It is a good contender for over-volting with a 13.5-13.8V PSU to perhaps be more attractive than the MCP600, but then again, over-volting the MCP600 also makes it more attractive too.

I'm still highly leaning towards 2 x MCP600's, perhaps with both running at 13.5-13.8v via an external cheap PSU, as the best possible solution. Give me a single pump with that PQ curve, power and heat characteristic, and make it quiet, and that would be my "ultimate" water-cooling pump.

As hard as this may be to accept, but centrifugal pumps actually add less heat to the water in more restrictive setups. As the motor slows down due to the increased load there is not as much electrical current flowing through the pump, and so the pump both draws less power, and adds less heat into the loop. I have measured and observed this many times and it is true.

However, the differences are not huge. The difference between a very high restricted pump, and totally free flowing, is around a 20-30% difference when it comes to pump power and heat. The Cascade system I selected above is perhaps a fairly moderate/middling sort of resistance when considering adding extra blocks and reservoirs and the like, so I would not expect pump heat to vary by more than 15% either way over what I've stated. This pushes the figures around a bit, but still not enough to change the rankings as every pump is affected in roughly the same way by changes in resistance.

 

bwana, the overclocking limitations on Intel CPU's is due to their SOI implementation, as opposed to AMD's implementation. OC.com flagged this as an issue a while back after some information from Intel that their SOI CPU's generally don't respond all that well to extra cooling (unless you're super-cooling them).

However, this may only be a transient issue for Intel. Making broad scale over-clocking judgements is not really the place for the pump comparisons above, purely because it is a CPU dependent issue. I do agree though, P4's in general are strange beasts when it comes to overclocking and applying stronger levels of cooling. Some respond well to extra cooling, but in general most do not respond as well as the AMD CPU's.

As you say, water-cooling's major benefit lies with the ability to hand over-volted heat loads, still hold the CPU cool, and enjoy the extra over-clocks that the extra volts provides. Sadly Intel's SOI solution does not benefit from extra volts as much as AMD's, and so the gains are somewhat limited.

 

Rd-30?

 

Thanx..

Would be nice to see some of these pumps with cooling fins on the motor, like some of those 3phase multi HP monsters..

There maybe a project for some one in that. Expose the motor on a pump and design a pasive thermatic-syphon cooling loop, for the motor..

 

Cathar, thank you for your time and effort. Outstanding service to the whole WC community.

Now I just have to figure out how it applies to my setup using a 6002A and fedco 2-343 core plus 18 feet of 1/2 ID clear, and the rough equivalent of 40 feet of 3/4 copper used in my ground-loop (twin 30 ft 3/8 coils connected in parallel connected to 2 5 ft 3/4 copper risers sections).

Any thoughts on how best to size the pump choice with that plumbing monstrosity?

 

Estimates for RD-20, and RD-30, based on 66% of current draw entering water as heat due to internal configuration.

RD-20, 40W of in-line heat added
RD-30, 51W of in-line heat added

RD-20, 9.4 LPM, 0.040, 0.158
RD-30, 11.5 LPM, 0.039, 0.155

RD-20, WB delta = 15.8C, Water delta = 140 * 0.040 = 5.6C, Total CPU Temp = 21.4C above ambient
RD-30, WB delta = 15.5C, Water delta = 151 * 0.039 = 5.9C, Total CPU Temp = 21.4C above ambient

Both the RD-20, and RD-30 work out about equal. As you raise the radiator performance both pumps will begin to close in on the MD-20RZ, but really both the RD-20 and RD-30 are somewhat in the category of being overly powered.

 

My idea of a "perfect watercooling pump" for almost all scenarios (excluding very small and quiet radiators) would be:

12VDC powered pump
~25-30W typical power draw
Iwaki/MCP600 style separated magnetic armature (least heat added to loop for power draw)
Peak pressure head: ~8.0mH2O
Peak flow: ~13LPM.

Really just a DC powered pump that is much like the 60Hz powered Iwaki MD-20RZ, with a bit more peak flow to widen the PQ curve slightly to cater for the lower PD blocks nicely.

2 x MCP600's @ 13.0-13.5v each would fit that order/scenario perfectly, but we just need those characteristics in a single DC pump. People with a PSU that allows the 12v rail to be adjusted to 13.0v could enjoy such today with 2 x MCP600's.

If Iwaki made something like an RD-15, and at 12v, then that would be the "ideal" single pump.

 

That's a pretty special case sort of scenario. Again, probably the Iwaki MD-20RZ would be the best choice for you, but perhaps given your extraordinarily large amount of additional heat dissipation capacity, then yours would be the one very rare instance where the MD-30RZ might be considered as being worth-while.