1 water cooling loop for both CPU and GPU...

[LjSpike]

I was having a bit of a think about using 1 water cooling loop for a GPU and CPU. One good pump with a nice 360mm radiator or such, then the pipes split so the CPU and GPU are in parallel. One way valves where it splits and joins back up to prevent back-flow.


I'm likely not going to do this, but just curious about it's feasibility.

 

[SeñorDonut]

I'm no waterflowologist, but I would think that the coolant would take the path of least resistance, so only one component would get cooled unless each loop has its own pump.

 

[Josh | NFC]

Hello, Lj.

My theory is as follows:

1. You will weaken pressure to both blocks significantly.
2. The higher density block will likely not be cooled well as Senordonut pointed out.
3. Best case scenario even without any issues a properly setup loop will be the same temperature throughout, so you would just be adding complexity.

However what I am interested in is hearing you out on why this could be an effective idea. I might be (probably) missing something.

Peace

 

[loader963]

I wouldn’t see the advantage over a simple series route. Over on oc.net, the wc section has a few parallel builds though so maybe there’s something to it. But I haven’t seen any with a backflow valve....

 

[Josh | NFC]

I wouldn’t see the advantage over a simple series route. Over on oc.net, the wc section has a few parallel builds though so maybe there’s something to it. But I haven’t seen any with a backflow valve....

Yeah there are instances where parallel flow is desired (in a slightly different form than the OP suggested), but usually its for triple/quad GPU setups. Personally I think it's best to do series and just have a strong pump. High flow is is the name of the game seems like.

EK has a nice little blurb on it--they end up recommending serial too to keep things simple:

https://www.ekwb.com/blog/does-loop-order-matter/

 

[loader963]

Regardless that’s still a rare situation in lff builds, much less sff lol.

 

[BernardoZ]

I'm no watercooling expert either, but I saw this today.
It seems like this guy watercooled both cpu and gpu on a single 92mm rad inside a Dan A4 lol

 

[loader963]

I'm no watercooling expert either, but I saw this today.
It seems like this guy watercooled both cpu and gpu on a single 92mm rad inside a Dan A4 lol

That is nuts !!!

 

[thewizzard1]

Just run the loop from the CPU (lower power and more temperature sensitive) in front of the GPU, to the rad to the pump. That'd work fine.

 

[bichael]

+1 for the above, happy to see someone else say that. For any loop with a small pump or small rad then having the cpu after the rad is definitely the way to go.

A lot of the watercooling rules of thumb are built around getting the absolute best cooling with massive pumps and massive rads etc. Definitely possible to get away with a lot less and still get good results. It makes me cringe slightly whenever I hear that the temperature all around the loop is the same, that's clearly not possible as heat is going in and coming out, it's just a reasonable approximation given the massive flow of most water cooling pumps compared to the load they are handling.

In terms of parallel cpu and gpu it should work fine, go for blocks with somewhat high/similar resistance and it will find a balance point that while not 50/50 should still work fine for both (massive flows remember). Including backflow valves would probably help due to the added resistance in each path but shouldn't be needed to actually prevent backflow. Tees should be arranged so that flow is split and recombined evenly (ie in at the bottom of the "T" outs at the top at either side). As others have mentioned though there's not really anything to gain from doing so in the vast majority of cases - so simpler to just go series.

As an aside parallel is the norm for cooling and heating circuits in buildings, the difference being there you are usually connecting up 10's to 100's of coils, not 2 or 3. So it's easier and more efficient in terms of pumping to add the flows for all the coils together and keep the pressure low. Moreover the temperature differentials are much higher (economics), say 6oC or more between water in and out of a coil, so combined with the large number of coils parallel provides a means of getting water at the temperature you want to all the coils.

 

[loader963]

^ my setup goes from :

Pump-sens1-240-cpu-gpu-sens2-T-360-pump in a series. Sensors 1 and 2 have never been more than 2C apart while gaming. I have only built 1 wc rig but in my limited experience coupled with what appears to be a majority of seasoned enthusiasts over at overclock’s wc section seems to be that there isn’t much difference in a series at decent flow rate or better.

Now I admit idk with regards to a parallel loop but the few I’ve seen have went series from the cpu to parallel GPU blocks, back to serial and on the the rest of the loop.

 

[bichael]

Yeah the rules of thumb are definitely valid in most cases. It's just that I'm a mechanical engineer so I enjoy getting into the theory
The main point is just to note that builds with less rad or smaller pumps than typical may well be workable.

Below won't be relevant for 99% but couldn't help myself so I did some water temp numbers anyway...

Assuming 100W for cpu and 250W for gpu with 30oC water going in.
Note 2lpm is around equal to 0.5gpm which is the often quoted minimum to aim for, many pumps would easily achieve double or triple of that though (which is why this is mostly a moot point).
At 1lpm having the cpu first gains 3.6oC on the temp going into the cpu (compared to having gpu first) which would be worthwhile.
Parallel included for the sake of it to show that even if it gets a bit more flow overall due to lower pressure it will perform if anything a little worse due to lower flow across the blocks (again though put a D5 on it and would probably be back to negligible differences).

1 lpm. > rad 30 > pump 30 > cpu 31.4 > gpu 35.0
2 lpm. > rad 30 > pump 30 > cpu 30.7 > gpu 32.5

1 lpm. > rad 30 > pump 30 > gpu 33.6 > cpu 35.0
2 lpm. > rad 30 > pump 30 > gpu 31.8 > cpu 32.5

1.2 lpm. > rad 30 > pump 30 > tee > cpu (0.6lpm) 32.4 & gpu (0.6lpm) 36.0 > tee 34.2
2.4 lpm. > rad 30 > pump 30 > tee >cpu (1.2lpm) 31.2 & gpu (1.2lpm) 33.0 > tee 32.1

 

[loader963]

That’s pretty slick math good sir! I guessed that there would be minor hit and cold spots but everyone says it s close. Have you ever seen a cpu and a single GPU run parallel though?