Is it possible to tune water flow in a parallel loop to suit different restrictions?

[Liam_G]

Is it possible to tune the flow of water in a parallel loop via a 3 way manifold to suit different

blocks restrictions? EDIT: Solved


I am seeking to fully watercool an ASUS Rampage IV Extreme; ie, Mosfets, Chipset and RAM, as well as

an i7 3960x CPU and 2 x GTX580 GPU's or equivalent. I am including the RAM and Chipset because

the system will be in an entirely passive chamber of my custom case, basically sealed with no airflow whatsoever and I

anticipate that the RAM (8 dimms) and Chipset will require some cooling to function well.


So, the details:

I will have one large custom RES, followed by an Swiftech MCP35X2 (ie 2 pumps) which will then flow

through 2 x XSPC RX480 rads, followed by the CPU with an EK Supreme HF Block, all in serial so far, now it gets tricky though, after the CPU I want to

split the flow into 3.

the goal is for me to have one inlet to a manifold and three outlets,
Outlet # 1 - 2 x EKWB GTX580 Fullcover blocks in parallel,
Outlet # 2 - EKWB ASUS RIVE Mosfet block
Outlet # 3 - 2 x EKWB RAM block and EKWB RIVE NB Block in series

That will then all return to an outlet manifold; ie 3 into 1 which will then return to the RES.

it will all look something like this:

.................................................. ................./ > Mosfet------------\
RES > MCP35X2 > RX480 > RX480 > CPU>Manifold--> 2 x GPU in parallel-----> RES
.................................................. .................\ > RAM > NB > RAM --/

Ignore the full stops, couldn't show spaces without it?

(the uneven distribution of blocks is due to tube routing, which is a nightmare on x79 with full WC'ing)


So basically the question is, can 2 x swiftech mcp35x's provide enough flow and pressure to be split

three ways and can that flow/pressure then be tuned to suit each blocks cooling needs and respective

restriction.


Obviously I want the most flow to the CPU and GPU blocks, and the least to the Mosfets, NB and RAM

blocks, which is converse to their respective restrictions AFAIK. So is it possible for me to tune

the flow through a custom manifold by sizing the respective ports, ie, if i have one inlet point, off

that i will have three outlets each varying in size to supply the required flow rate to each component

taking into account their respective restriction. You could put it like being a 3 way manifold, each

outlet has a G 1/4" port and then each port has a ball valve attached and the ball valve is opened a

certain percentage to increase or decrease it's flow relative to the blocks need.


This is not a question of whether I will get better temperature or flow, simply a question of will it

work. So, what do you guys think? Bit of a rant I know but something that I feel hasn't been addressed in

the usual Parallel vs Serial Loop question. This is probably a question for Martin or Skinnee but i'm

happy to hear all opinions/ideas/bull$#*t etc :P

Edit; so I did some flow analysis in Solidworks with FloXpress, looks like it will work, thanks for all the input

[Waterlogged]

If I understand what your asking correctly, absolutely possible. . .but, you'll need to toss a valve and pressure gauge on each branch of the loop to fine tune the pressure balance.

[Liam_G]

If I understand what your asking correctly, absolutely possible. . .but, you'll need to toss a valve and pressure gauge on each branch of the loop to fine tune the pressure balance.

That's what I thought, the principal of it sounds feasible just the details that need to be worked out :P but was hoping to avoid a huge amount of testing and prototyping of the manifold by running some simulations first

Maybe Tiborr can throw all those EK Blocks together and see what the flow rates are out of each sub loop so that i have a basis to go off after all blocks are connected, hint hint, nudge nudge :P

[Martinm210]

You would want a flow meter on each branch, not a pressure if you really wanted to tune. Normally flow in each parallel branch will adjust until pressure is equal in each line (Lower restriction branch increases flow until pressure drop of that line is equal to that of the higher restriction line, etc.)

I wouldn't put the CPU in parallel though, I'd leave that in series up front then split out.

[PiLsY]

Use an Apogee HD cpu block and you can save yourself half that headache.

Id also keep it simple and just use 2 branches (assuming using the Apogee HD), one for gpu's and one for mb/pwm/ram. You will get all flow past cpu, majority flow past GPUs and whats left through mb/ram etc - so it puts your flow where you need it. A single 35x would give enough flow for this loop.

Even if you dont use the Apogee HD then stick to 3 branches, dont split your mb and ram/pwm. Its just not worth it, uneccesary complication . You would probably need a pair of pumps in this case though to make sure the cpu got enough flow.

[antiacid]

+1 on the Apogee HD, save yourself a lot of trouble for arguably very little gain!

[Liam_G]

You would want a flow meter on each branch, not a pressure if you really wanted to tune. Normally flow in each parallel branch will adjust until pressure is equal in each line (Lower restriction branch increases flow until pressure drop of that line is equal to that of the higher restriction line, etc.)

I wouldn't put the CPU in parallel though, I'd leave that in series up front then split out.

Ah, that makes sense, thanks for clarifying..... And I have just done some sketches and having CPU in series before manifold will work and putting the CPU in first in series seems to allay my fears that the CPU will be worse off, I think that takes a lot of the problems out of the equation for me.


I've also tried to fit the Apogee HD into the loop and from what I have been trying and how it helps me with loop order it won't work for me, still might use my Apogee XT but I think restriction is too high, even with Rev 2 XT

[Liam_G]

So, the loop will be ordered something like this. The manifold will be under a midplate, actually it may be part of the midplate, but I think this should work, one in, one out, with all components getting what they need without being in series.

The motherboard and RAM blocks will all have custom shaped copper tubing to allow for the tight bends between blocks and midplate and also to restrict flow a little, the CPU and GPU's will have 1/2 ID 3/4 OD tubing.

[Waterlogged]

You would want a flow meter on each branch, not a pressure if you really wanted to tune. Normally flow in each parallel branch will adjust until pressure is equal in each line (Lower restriction branch increases flow until pressure drop of that line is equal to that of the higher restriction line, etc.)

I wouldn't put the CPU in parallel though, I'd leave that in series up front then split out.

Thanks for the correction Martin.

[prava]

Outlet # 1 - EKWB Supreme HF CPU block,
Outlet # 2 - 2 x EKWB GTX580 Fullcover blocks in parallel,
Outlet # 3 - EKWB ASUS RIVE Mosfet block
Outlet # 4 - 2 x EKWB RAM block and EKWB RIVE NB Block in series

As Martin stated, pressure will be the same on each brand and, thus, flow will be different, according to each branch pressure-loss. So, if you did what you are planning, you would have the least flow on the cpu block which, as it is, is the most important thing as its flow sensitive.

I'd definitely go for Swiftech's solution:



Use your cpu block as an IN and, then, OUT it to the other components. I'd aim for:
-out1: gpu1
-out2: gpu2
-out3: chipset and everything else.

I believe this would net you very decent temps and you'd have a lot less clutter.

[Liam_G]

@ prava

Thanks for the input, but as i sketched out on my last post, i have taken Martin's advice and i will be putting the CPU in before the manifold, so it will be in series, then it will split three ways through the manifold. This ensures my CPU gets enough flow and makes it easier for me to tune the other three... Also, as shown in my sketch you will see i have a rather less cluttered loop than what i could achieve with the swifty HD.

With the swifty HD the outs would not allow me to get where i'm going, which is a degree of symmetry in the loop, you might agree that x79 lends itself to this setup. I'm approaching my case design as a piece to be shown off/admired (mostly me just staring at it :P) so i'm really trying to find the best loop order/tube run possible, and i think i have cracked it now... I still have a lot of designing to do though so i will be constantly playing with my design until i start manufacturing the case, that includes swapping hardware if something else comes along, i can tailor my design to suit whatever i go with at the time, I won't be purchasing anything until my case is ready to be built to ensure compatibility. feel free to sketch up how you think i should run my loop if you think you have something that fits what i'm after.

[prava]

@ prava

Thanks for the input, but as i sketched out on my last post, i have taken Martin's advice and i will be putting the CPU in before the manifold, so it will be in series, then it will split three ways through the manifold. This ensures my CPU gets enough flow and makes it easier for me to tune the other three... Also, as shown in my sketch you will see i have a rather less cluttered loop than what i could achieve with the swifty HD.

With the swifty HD the outs would not allow me to get where i'm going, which is a degree of symmetry in the loop, you might agree that x79 lends itself to this setup. I'm approaching my case design as a piece to be shown off/admired (mostly me just staring at it :P) so i'm really trying to find the best loop order/tube run possible, and i think i have cracked it now... I still have a lot of designing to do though so i will be constantly playing with my design until i start manufacturing the case, that includes swapping hardware if something else comes along, i can tailor my design to suit whatever i go with at the time, I won't be purchasing anything until my case is ready to be built to ensure compatibility. feel free to sketch up how you think i should run my loop if you think you have something that fits what i'm after.

There is no doubt that if you are aiming for aesthetics the Swiftech block I showed may not suit you as you state.

Anyway, please create a worklog so we can follow your progress

[Liam_G]

I will start the worklog when the first piece of the case hits the CNC :P

Also, quick question, does anyone here believe that even with a totally passive chamber and no airflow over the hardware and no cool air entering the case (it's basically sealed air tight to keep dust out), that the RAM and chipset do need watercooling? bearing in mind i want to OC the CPU (i7 3960x) close to 5ghz, could i get away with stock heatsinks on motherboard (ASUS RIVE which has a fan over southbridge) and RAM, or possibly put dominator fans on the RAM, thus providing air movement over the RAM and a little over chipset without it actually being cool air.

If that would work i would just watercool CPU and GPU's greatly simplifying the loop. The main thing i want to cool on mobo is mosfets/Vreg's (whatever they are called)for better CPU OC and the ram because there will be 8 dimms and i think they will run hot that densely packed.

Noise is an issue too, i don't like small fan's spinning at high rpm, the rads will have 8 gentle typhoon fans running either 500rpm or 800rpm so if the dominator fans and the chipset fan are too noisy i won't put up with them and will have to watercool at least the chipset, maybe not the ram if it is fine without fans or watercooling. Any thoughts?

[PiLsY]

Dominator type fan would do fine for the ram. Chipset/vrm though would need watercooling if you have no airflow at all through the case.

[Liam_G]

Dominator type fan would do fine for the ram. Chipset/vrm though would need watercooling if you have no airflow at all through the case.

Yeah, I might go with the Rampage IV Formula, stick to 4 dimms and just water cool the VRM's, the RIVF has a two piece heatsink so could leave the stock chipset heatsink on.

In the meantime, I found out Solidworks has an inbuilt tool called FloXpress, it's a simple one in and one out flow simulator, instead of the full blown Solidworks Flow Simulator which i don't have. So i ran some basic tests with it to see what the outcome might be.... This is by no means an accurate depiction of flow, none of the blocks are accurate as regards restriction and pressure drop, I will work on that in time, but it gives me a basis to work off in designing my manifold, which as you can see is part of my midplate, which will be a rather thick piece of acrylic or acetal etc. Ignore the velocity (m/s) scale as well.



So far i am quite happy with the split of flow out of the manifold, i have played around a bit and it is very easy to tune a different flow with minor adjustments to position of flow path and size of flow path in manifold. So i know once i have the blocks and can actually test this that it will be easy for me to get it right.

[PiLsY]

Yeah if you can split the VRMs from the chipset you should be just fine with the stock heatsink there.

Have you tried messing with smaller diameter tubing and barbs on the sections you want ower flow on? That may even help it look like less of a confusion of tubing - kind of help the eye make some order of things.

Your drawing skills have come on a lot - very impressive!

[Liam_G]

Yeah, RIV Formula could be an option.

If i go with RIV Extreme though it will look less cluttered than this solidworks model in real life, i will be using nickel plated copper tubing custom shaped for all of the motherboard and ram blocks, with those little bitspower sli crystal link fitting's to connect it to blocks and the CPU and GPU's will either be 5/8 OD or 3/4 OD with compressions or barbs, and i will either use a dye or coloured tubing to emphasize them over the nickel plated copper. So i don't think it will hurt the eye too much, x79 is a bit crazy but thats why i'm trying all this manifold stuff to get the loop lookin good.

Haha, don't judge me on my Paint skills, Solidworks is more my style

[OldChap]

I have a few rigs set up in parallel, cpu only, and would offer the observation that with dissimilar blocks the flow just finds its equilibrium.

Now, to the point of this post; reducing the flow on any leg (I have valves installed) does improve the flow in others a little but even a VERY reduced flow, in the region of 0.25gpm, still results in good enough cooling. Not optimal, sure but certainly adequate.

If you choose blocks that are the least restrictive in the first instance I suspect that, provided you have the means to measure temps for each component to prove the effectiveness, you will find that you can achieve an aesthetically pleasing layout that works well enough.

Try running you ideas out on the bench using just tube and fittings maybe

[PiLsY]

Yeah Crystal Links would tidy that up a hell of a lot . If you like the nickel look take a look for TFC variable SLI connectors. Same as crystal links just nickel plated. They're a narrower ID too which may be of help for you increasing restriction in the mobo/ram/vrm circuits.

[BrokenArrow]

Love this plot. Years ago, plumbing GPUs in parallel was quite the hot topic. The same color contours show basically equal flow through the GPUs which was a hard concept for people to get their brains around. Nice!

[Liam_G]

Yeah Crystal Links would tidy that up a hell of a lot . If you like the nickel look take a look for TFC variable SLI connectors. Same as crystal links just nickel plated. They're a narrower ID too which may be of help for you increasing restriction in the mobo/ram/vrm circuits.

Yeah, i think it will look good, the whole nickel plated copper tubing and bitspower crystal link fitting thing i stole from Murdermod MKII, it looks good, only problem i foresee is retaining the tubing, it's only held in by an o-ring, so when i have a tube with no opposing force, ie the RAM Blocks to the Chipset Block above the CPU, it may just pop out, I will have to come up with something there, maybe slightly oversize the tubing or add an extra o-ring.


@ Oldchap, yeah it actually seems like it won't be that hard to tune this the way i want at all, like you say it's not too bad just left alone, i just want to ensure the GPU's get the lions share of flow and not the VRM/RAM and Chipset blocks, don't really care how much they get, they don't need a lot, Plus two swiftech MCP 35x's in series is pushing some serious flow, i think i'll be in the realm of 2 to 2.5 GPM, so if my GPU's get 1.5 of that then I'm sweet.

@ Broken Arrow, Yeah I'm loving this FloXpress tool, so cool to actually see the effect of running in parallel vs series etc... I know the GPU's will recieve basically equal flow to each other, i just need to ensure they get enough of the overall flow, hopefully some of my testing carries over to the real world, this is really just a basic indication for now.

[Liam_G]

(quote removed by Shazza)

Frankly buddy I think your misplaced in your assumption that i'm a noob. Just because I'm putting something out there for some opinions doesn't mean I don't know what i'm doing (This will be my fourth WC build, although my first adventure into parallel outside of just GPU's). What you have failed to realize is that I'm actually doing this to simplify my loop, I'm talking everything here too, from tube runs to draining to looks to division of flow between what components need it. Part of the reason I'm running RAM blocks is to get my tubing from chipset block to my midplate without it looking like total sh#@, which with so much going on on x79 loops is easy to end up with a crappy looking loop, I actually haven't seen a WC loop on x79 I'm happy with yet... Another thing, what isn't simple about parallel, can't you get your head around it mate? From my testing I'm actually seeing an increase in flow in my loop by decreasing restriction, furthering the whole GPU in parallel thing, maybe your too much of a noob to understand that though.

[PiLsY]

Liam dont bother explaining dude. A multi branch parallel flow manifold is far from noob. Just report the post for trolling and forget about it . I hovered over report myself, but at the end of the day its your thread so its up to you whether youre irritated enough to bother .

[Liam_G]

Liam dont bother explaining dude. A multi branch parallel flow manifold is far from noob. Just report the post for trolling and forget about it . I hovered over report myself, but at the end of the day its your thread so its up to you whether youre irritated enough to bother .

Haha, no i won't report him, not really that bothered, just annoying when ppl comment before they take the time to comprehend what is being discussed in any thread. I can tell he has probably read my title, looked at my first post (admittedly long) and decided it's not worth reading and then proceeded to flame away, thus totally missing the point that this is beneficial to my loop and possibly other people's as well. XS is not about just slapping some parts together in the simplest array possible to get a system running, It's about taking that extra time to plan and experiment and try new stuff. I have been really encouraged by everyone else's comments about what I'm trying to do and that's why i love this forum. It should always remain about pushing the boundaries, whether that be OC'ing, custom mods and cooling etc.

[shazza]

Good detail in your discussion Liam. I cleaned up some of the posts in the thread, even though the poster apparently read the info after being rude).

Let's keep the conversation to the technical discussion, as directed by Liam.