Overclock.net banner

Series Versus Parallel Liquid Cooling Loops

10K views 11 replies 8 participants last post by  rebelranger 
#1 ·
Which cools better?

There are many overclocking sites that overclock and provide benchmark results. Some run single loops between the CPU and GPU cooling blocks. Some run two loops between the CPU and GPUs. The benefit of the two loop set-up allows for the hot water exiting the cooling block to go directly to the radiator. Single loops send the hot water exiting the CPU block to the GPU blocks. Therefore, the GPUs are getting hot water which is used to transfer the heat off them.

Now lets look at the GPU cooling, regardless of the inlet water source. With many multi-GPU configurations, the graphics cards are also placed in a single loop for liquid cooling. This means that the hot water exiting card 1 is sent to card 2 to cool it. In 3-way SLI, the 3rd card is getting hot water from BOTH card 1 and 2.

Overclocking a processor causes more heat to be generated. As the heat increases, the processor's stability decreases. Therefore transfering the heat away from the surface of the processor allows for stable overclocks. So it is important to get the heat transfered efficiently. Water cooling does this well and provides good temperatures at increased loads.

So my theory is:

To get the best heat transfer and coolest operating temperatures, you need to have the lowest inlet temprature for EACH device using liquid cooling. Thus you can overclock your processor HIGHER with better stability.

So I got my old Thermodynamics book from college. I found the equations used for conductive and radiant heat transfer.

Q=-kA(Ts-T)

Q: heat transferred (the higher the more heat removed)
h: heat transfer coefficient (constant for H20)
A: area (surface area of block)
Ts: temperature of the surface
T: temperature of the medium (water)

So without doing any math, we can see the relationship between the amount of heat transfered to the difference of the surface and medium temperatures. The higher (Ts-T) is, the greater the Q will be. So if we have a CPU overclocked to a set frequency, the amount and temprature of heat needing to be transfered is constant. So the variable we can control to increase the Heat Transfer (Q) is T, the temperature of the inlet medium (water).

Let's do some math to see how this theory would factor against the heat transferred (Q). A and -k are constant so we don't need to analyze their effect.

Series
(Ts-T): 50C-40C=10C

Parallel or indepedent water source per device
(Ts-T): 50C-35C=15C

With a simple 5C variance of the inlet water temperature, the Heat Transferred (Q) increases by a factor of 10 to 15. We can derive a formula that calculates the factor in Series versus Parallel Heat Transfer since A and -k are constant.

Qp: parallel heat transfer
Qs: series heat transfer
added s and p for the corresponding temperatures of the parallel versus series quantities

Qp/(Tsp-Tp) = Qs/(Tss-Ts)

thus,

Qp/Qs = (Tsp-Tp)/(Tss-Ts)

So using our 5C temperature variance from above we get:

Qp/Qs= 15C/10C

thus,

Qp = Qs(3/2)

Hence, the amount of heat transfered in the parallel loop is 3/2 times the series loop. With a 5C temperature drop in the inlet temperature, the parallel solution gets you 1.5 times the heat transfer. If we had lowered it another 5C the factor would be 2.

In conclusion, running series cooling loops lowers your heat transfer ability and limits your maximum overclock with stability.

It would be interesting to see if someone would try an experiment on graphic cards using this theory. They would have to split the tube coming out of his resorator and run seprate inlet lines to the cards. Then bring them back together as the water returns to the resorator.

Thanks for reading
 
See less See more
#3 ·
Quote:


Originally Posted by ltulod
View Post

I'll try in on Monday.

Thanks and please post your results. Please note the number of devices you are cooling in your loops. It would also be interesting to see the temperature of the fluid before and after it passes through the radiator.
 
#4 ·
I'm sorry but it seems totally wrong to me.
Throwing a bunch of formulas doesn't mean you arrive at the right result for sure.

What about flow rate?
You assume that splitting will keep the exact same surface temperature on the component to cool down but somehow magically generate less heat, even with reduced flow rate after splitting.

If you split something then later merge the waters, you lose any advantage that could have been gained with splitting. The way you explain it, it's like using a single res for 2-3 loops. No advantage whatsoever. The water will mix in the res.

If the water is exposed to 1x400W of heat load, it's the same as 2x200W. Splitting doesn't reduce the heat load. The only way to gain an advantage is with completely parallel loops, each with one pump/res/rad for EACH CPU or GPU.

The difference between the hottest and coldest point in a huge loop (CPU+chipset+GPU+GPU) is less than 2 degrees.
Try it for yourself first, instead of suggesting to others to try it.

TiGa
 
#7 ·
Thanks for the comments. I started this thread because I am WANT to build a new system and plan to use watercooling. Also, I do not have water cooling parts to test with on my current system. I was hoping to gain information from this community to help myself make a good decision before buying. That being said, I would love to test and verify my initial calculations. Building and testing is fun!

So far, it has been noted that the temperature of the water leaving the radiator is only a few degrees less than the temperature entering the radiator. My initial assumption was centered on the fact that the radiator would remove heat, thus lowering the water temperature. Therefore, if the colder water was distributed evenly to different components (parallel) then the heat transfer per component would be greater. Conversely, my initial theory was that series loops moved hot water over the sequential components in the loop (compared to parallel). Thus not providing maximum heat transfer in series.

Mass flow rates would be an effective way to increase heat transfer per component, but I do not see this idea, or pump overclocking, being done much. Perhaps reducing the tube ID would increase flow rate and improve heat transfer. Again another experiment.
 
#8 ·
Quote:

Originally Posted by utengineer View Post
Thanks for the comments. I started this thread because I am WANT to build a new system and plan to use watercooling. Also, I do not have water cooling parts to test with on my current system. I was hoping to gain information from this community to help myself make a good decision before buying. That being said, I would love to test and verify my initial calculations. Building and testing is fun!

So far, it has been noted that the temperature of the water leaving the radiator is only a few degrees less than the temperature entering the radiator. My initial assumption was centered on the fact that the radiator would remove heat, thus lowering the water temperature. Therefore, if the colder water was distributed evenly to different components (parallel) then the heat transfer per component would be greater. Conversely, my initial theory was that series loops moved hot water over the sequential components in the loop (compared to parallel). Thus not providing maximum heat transfer in series.

Mass flow rates would be an effective way to increase heat transfer per component, but I do not see this idea, or pump overclocking, being done much. Perhaps reducing the tube ID would increase flow rate and improve heat transfer. Again another experiment.
A couple of degrees would be very conservative, it's actually usually tenths of a degree.
Remember this one:
Q = M X CP X dT

Make it a bit more friendly for emperial units:

Q in watts = 263.43 x (Flow rate in GPM) x (dT in C)

solve for dT

dT in Celcius = (Q in watts)/(263.42 x (Flow rate in GPM)

So a 200 watt heat load at 1.5 GPM will give a dT of about .5C, so most people don't even see a full degree of difference throughout the loop.

Parallel might do ok if you had a Tri-SLI configuration where you ran the GPUs in series. Not so much for better GPU cooling, but to relieve pressure drop for the system and maximize flow rates on the CPU. Also GPUs tend to handle the heat better so a little higher heat on the GPUs due to lower flow rates running parallel might be a worthwhile tradeoff for minor flow increases on the CPU.

Testing that is pretty tricky though. Most folks have ambient temps that fluctuate a few degrees and the processor DTS sensor is only good to 1C resolution anyhow. We're talking differences of less than a degree, so you'd need very extensive logging of temps with good sensors plastered throughout the loop and air to get anything solidly tested.

Even with good logging tools, I've found flow rate effects to be fairly tricky to measure. For one, the differences are so small they are very hard to measure. Even with good digital sensors, you typically have at least .2C to .5C of relative error. You'd have to build an array of sensor for each measurement point to even get down under the .1C resolution area. Unfortunately the tools capabile of measuring better than that are far to expensive to even consider. Secondly, the pump heat dump increases with flow rate, so it's counterproductive to the increased flow rates. It does depend on the heat exchange system, but regardless the pump heat dump generally quickly becomes an issue for gains.

Regardless, it's still fun trying to measure. Watercooling loops are pretty hard to screw up these days, so the recreation in it is alot of tinkering after it's built. The science and math is a good basis for understanding where too look, but it'll never replace some good measuring and tinkering. Things like mounting and remounting, lapping, measuring air temperature into the radiator...etc. These can all add up to small gains that in aggregate..add up to something. If you enjoy tinkering, you'll never leave it alone anyhow, so go ahead and build it...then tear it apart...again...and again...


Oh, and pump overclocking. Pump heat dump begins to play a very counterproductive role in our loops when you factor in the very minimal flow rate gains. Also most of the pumps have protective circuitry. The DDCs for example won't start beyond about 13.1V, and the D5, will start at 24V, but it will turn it's rpm down to about 13V.

There is some gain to be had by multiple pumps in series or so some test have shown, but it's not as much as you might hope for. The few tests I have run, the pump heat dump of the second pump in series usually was so significant it completely overcame my net gains from flow rate. This does depend on the block though. Some blocks like the GTZ or Koolance blocks appear to be very happy with more pumping power.

For most folks/setup, series works well. The pumps are strong enough to handle the pressure drop and operate the pumps at a good operating point without putting too much stress on the pump.

I have made the mistake of overdesigning a pump in a commercial facility. Nothing like adding valves later to add some artificial resistance because the pumps are overheating and drawing more current than you can legally set breakers for. I havn't seen that with watercooling pumps, but the same principals apply...not the best thing for them to operate at the far right (low/no restriction) side of the curve...
 
#9 ·
Quote:

Originally Posted by utengineer View Post
Mass flow rates would be an effective way to increase heat transfer per component, but I do not see this idea, or pump overclocking, being done much. Perhaps reducing the tube ID would increase flow rate and improve heat transfer. Again another experiment.
There's nothing wrong in trying stuff for yourself, things like pinching the tubing or using splitters upon splitter upon splitters.
But there is such a thing as "re-inventing the wheel". I don't want to sound mean here but you should start by doing some reading on the forum here or elsewhere. The FAQs and stickies are very good reading material. It's really the best way to learn about watercooling. Asking some questions to the right people is the way to go too.

Some nice people like Martin do the kind of methodical testing that everybody would want to do themselves but lack the equipment to do so.

Before investing some money in your first setup, be sure you are ordering the right parts to save you from ordering again and again.
Example:
A good question to ask on the forum would be: "Is flow rate better with 1/4" or 1/2" tubing?", before buying some 1/4" tubing and learning the hard way that the flow rate would have been better with 1/2" tubing.

TiGa
 
#11 ·
Quote:
Mass flow rates would be an effective way to increase heat transfer per component, but I do not see this idea, or pump overclocking, being done much. Perhaps reducing the tube ID would increase flow rate and improve heat transfer. Again another experiment.
I used to over-volt my old high head 18W DDC that I had in series all the time, it didn't help temp's on my CPU only loop any, didn't hurt them either, I massively over-rad and over-fan. It was fun to play with though..


Going from 1 DDC to 2 in series dropped my temp's a C or two with a just a single block CPU loop, the redundancy is always a great thing with series pumps. If I had a higher restriction loop it may of helped more I don't know.

I've always gotten a little better temp's with larger ID tubing, not much but it all adds up if your pushing your clocks. I'd at least use 1/2" barbs with 3/8"-1/2" ID tubing on them. The only time I ever use all 1/2" ID tubing is when I have long external runs, I want every bit of flow I can get then. Usually I run 1/2" barbs with 7/16" tube. But they gains to be made from 3/8"-1/2" tube is very small with a normal length loop. The ID on a standard 1/2" barb is 3/8-7/16". Some say that the smoother tube/barb union that you get with 3/8" or 7/16" tubing on the 1/2" barb will help flow a bit since it has a smoother tube barb transition then the 1/2" to 1/2". Makes sense and it is a smoother union, I doubt it would be much gain, but any free gain is good.

WC'ing cooling gear isn't generally as flow dependant as it used to be, the older type blocks loved high head and flow. Take a look at the old "Swiftech Storm" with its jet and cup array, the more flow you could get through it the better.

I still use a jet block and high head pumps, I love them. But the newer blocks and rads just don't seem to beneifit as much as they once did. Which is really a good thing, saves some pump money that can be used for something else.
 
#12 ·
This an interesting debate.

A couple of things seem to be missing perhaps? Ambient air temps and its effect on the system/rad/blocks etc and the type of surface--cooper, Brass etc. Possibly the dissipation of heat (effected by air flow outside the loop) at each exposed surface.

I agree with the OP in theory, and although it seems that there may be variables that present a challenge to the results my past results really prove this as correct.

That being said, if I had a best "case" scenario and the room (inside a case) to build an optimal liquid cooling system, I would without a doubt want separate loops for each device being cooled.

I remember having an old P-4 "prescott" chip in a loop and it gave off so much heat that it affected my GPU enough to cause concern. I ended up running a separate loop and both the temp of the CPU and GPU were reduced.
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top