Does the flow rate matter - Tested single D5 pump at 1, 3, 5 and triple D5 pumps at max :)

[Tam3n]

Hello, it's experiment time again

 

How much does the flow rate improve when adding more pumps in series, and how much does it affect the thermals? Those were some of the questions I had in my mind.

So, I decided to have some fun, and added two more D5 pumps in my loop to find out. I was curious how it would affect the cooling performance, so I did a little comparison test.

 

Here's my water cooling loop, which should have some restriction in it:

- Pump(s): 3x Laing D5 Varios (1x EK-D5 Vario with X- RES 100 combo + 2 x EK-D5 Vario with EK-XTOP Revo Dual D5 Serial)
- Radiators: UT60 360 mm + UT60 480 mm  + Xtreme Nova 1080 mm
- Blocks: CPU: 1x Alphacool NexXxoS XP³ Light  Acetal,   GPUs: 2x EK-FC R9-290X Acetal+Nickel (Rev.2.0) full cover

- Like 5 m of 10/16 mm tubing

 

I was interested in the results at very high OC with high power draw and heat, since that's where the results might matter for me - though I wasn't expecting huge difference in any case.

 

I did the tests as follows.

First, before installing the additional two pumps in the loop, I ran the tests with a single D5 at speed settings: 1, 3 and 5 (max). I should mention, that the D5 kind of stalls (at least mine does) to a very low speed and vibrates more when you set it to as low as it goes (it almost goes below 1) so I ran it at something like "1.2", where it runs nicely and quiet, as the lowest setting.

Next, I added the two pumps with the EK dual top in series after the original pump with res combo and ran the tests with all three pumps at 5 (max).

 

I tested my CPU and one of my GPUs (other card was not powered on) separately.

For the CPU (i7 3970X Sandy Bridge-E six-core) I dialed in settings 5.0 GHz @ ~1.5 V and ran Prime95 with AVX small FFTs for the maximum heat output.

For the GPU (AMD R9 290) I dialed in settings 1300 core, 1600 memory @ ~1.5 V (1250 mV + 250 mV with reduced Vdroop bios) and ran Furmark @ 1080p for the maximum heat output.

I used the digital PSU Corsair AX860i to log the temps and other data. Max system power draw (coming in the PSU) for the CPU only test was ~650 W and GPU only test ~1000 W.

I ran each test for 3 min and logged the data with intervals of 30 seconds. Also, at each corresponding 30 seconds I manually wrote down the water temperature to excel file. This way I got six data points for each individual test with CPU and GPU temperature, power and water temps data. For the CPU temperature value, I used the average of the six core temps. Finally, I calculated the difference to water temperature for each of the six data points, and took the average of that for the final result.

 

Here's what I got:

 

Now something to note, is that the ambient temps (around 20-22 C) were not controlled or measured which affect the Tmax values, that represent the max temp that CPU or GPU reached during each test. The test were done in order from red to blue and I think the room was a bit colder during the triple pump test. Also the water temp probe is a basic cheap inline one with resolution of 0.1 C, but the actual accuracy and response time is questionable.

 

Anyway, the theory behind flow rate improving temps is the same as with more air flow improving temps. Basically, you make the boundary layer on the cold plate surface thinner and move more cool liquid in proximity of the cold plate surface (and heated liquid away), which improves the convection of the heat. Besides the cold plate, this should also result in the radiators working more efficiently for the same reason, resulting in additional reduction in temps (but you don't see that in component temp delta to water temp). However, consider you have infinite radiators and flow rate, which would keep the water temp and the cold plate inner surface at the ambient temp. You would still have certain temperature difference between the component and the water, which depends on the component heat output and thermal resistance from the component to the inner surface of the cold plate.

 

Depending how close you already are to this situation, where you can keep the cold plate and radiators inner surface close to water temp, determines how much more you can improve the temps just by rising the flow rate. As you can see, temps didn't improve too much for the CPU going from 1 pump to 3 pumps, but the GPU still saw some improvement. Also the CPU and GPU improved approximately in the ratio of their heat output. Obviously, the result will be greater the bigger heat concentration you are dealing with.

 

I've been thinking of doing some oc runs with cold water... So one thing to consider, is that the water viscosity is 2 - 3 times greater at below 5 C compared to the normal 30 - 40 C one would expect in a loop. Higher viscosity directly decreases the flow rate, which additional pumps will help to overcome.

 

Bonus - Had fun bleeding the system with triple D5, I think this is the first time I didn't need to tilt my radiators/case at all to get all of the air easily out.

 

[AngryBeaver]

I think one problem you have run in to with the way you did this test is you didn't let the water temp hit it's maximum deltaT.

 

You should have run these tests for a minimum of 20 minutes to let water temps reach their maximum. The reason for this is because when water temp is sitting at ambient (due to being turned off for example in between tests) then the flow will make a bigger difference as it is circulating cool water already in the loop. However, once your water temp settles under load the flow rate hits a point of diminishing returns pretty quickly... the water will move away from the blocks faster which helps get heat away more quickly, but it also spends less time in the radiator per cycle. So in the end you hit a point where adding more flow does not actually help.. and normally that is around 2 GPM or so.

 

So I guess my point is that the results are not going to be reliable unless the tests are run for significantly longer than you have.. and also having a fluctuating room temp will also not help.

[daimonie]

Nice going! Somewhat longer times might be good, but I really don't know what the thermalisation time of a high-flow rate system is. The only way to make sure is to check



 

[AngryBeaver]

Something else to note.. is that as I was saying earlier you hit a wall of diminishing returns. The reason for this is that you aren't actually moving more heat from the system. You are just circulating the water more quickly... the time spend in the radiator and block will still be the same amount of time over a specified time frame... it just changes how many full loops of the system occur in that time.

 

So once you hit this point the only way to actually reduce temps further is by adding either more radiator surface area or increase the air flow across those surfaces. Feel free to test this theory, but remember you cannot have accurate results without first allowing the temps to level out.

[Mick Naughty]

Ive noticed speed directly effects my temps in any rig running a parallel block or in parallel. Aside from that I can get more then 1-2c.

[Tam3n]

1 hour ago, AngryBeaver said:

I think one problem you have run in to with the way you did this test is you didn't let the water temp hit it's maximum deltaT.

 

You should have run these tests for a minimum of 20 minutes to let water temps reach their maximum. The reason for this is because when water temp is sitting at ambient (due to being turned off for example in between tests) then the flow will make a bigger difference as it is circulating cool water already in the loop. However, once your water temp settles under load the flow rate hits a point of diminishing returns pretty quickly... the water will move away from the blocks faster which helps get heat away more quickly, but it also spends less time in the radiator per cycle. So in the end you hit a point where adding more flow does not actually help.. and normally that is around 2 GPM or so.

 

So I guess my point is that the results are not going to be reliable unless the tests are run for significantly longer than you have.. and also having a fluctuating room temp will also not help.

Well I think missed the point a little bit. Hitting the stability of the delta T between the component and water takes only about a minute, if even that long. I can see this easily by running lets say Furmark with pump at 1, then turn the pump quickly to 5 and the temps of the gpus drop like ~4 C instantly (water itself remains at the same temp during this time period).

 

Yes if I wanted to make a conclusive test of what the delta T to air temperature would be, then I would need to wait 20-30 min for the water to slowly heat up and reach equilibrium, and also precisely monitor air temperature at the same time. But I wanted to mainly test how water flow affects the effectiveness of removing heat from the component, which delta T between water and component tells essentially... The absolute water temp doesn't matter in that regard, except affect water viscosity. If I wanted to, I could put my rad to intake cold Finland winter air from the outside and get same component delta to water if viscosity effects are disregarded.

 

That whole "water spends more/less time in ..." is irrelevant. All that matters is the effectiveness of the convection that increases with the flow rate. Of course you get diminishing returns once the radiator and blocks inner surface is close enough to the water temp, as I tried to explain earlier. One way you could think of the flow rate a bit like thermal paste, with high flow rate you will get better thermal conductivity and with low flow rate you have lesser thermal conductivity, which will mean higher delta between your component and surrounding air temp. Though even if you had infinite thermal conductivity in part of the path of the heat flow, heat will still have to transfer through the cold plate copper material itself etc etc... Heat transfer always needs a temperature difference... But with better conductivity it will be less.

[Tam3n]

1 hour ago, Mick Naughty said:

Ive noticed speed directly effects my temps in any rig running a parallel block or in parallel. Aside from that I can get more then 1-2c.

Yeah, I would assume the effect of the flow rate would be more noticeable in parallel gpu blocks for example. They split the "original flow" in half, so you only get half of what you would get if they were in series (though then higher pressure drop -> less flow). On the other hand, in parallel the pressure drop over the gpu blocks will be less, which slightly improves the flow rate and compensates for it...

[AngryBeaver]

I can assure you that you are not saturating the temp of your water in a few minutes. 

[Tam3n]

9 hours ago, AngryBeaver said:

I can assure you that you are not saturating the temp of your water in a few minutes. 

Yes, I agree.