"parallel flow best with 1046" - hunh?

[bobkoure]

from this thread at silentpcreview

Originally Posted by snutten

If you want Procooling numbers on why parallell loops are the better choice, read Cathar's article on pumps and flow.

Heck, I thought I knew a little bit about watercooling for low-noise. Clearly I still have a lot to learn.

Now, based on this thread here at ProCooling

Originally Posted by Cathar

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

I'd guess that a 1046 should be fine for someone looking for as-quiet-as-possible and temps just below the CPU mfgr's specified max.

However, I have to wonder what happens to the waterblock C/W when the rate is reduced by running in parallel loops.
I realize that the flow isn't reduced by half, first as the resistance isn't the same in both sides and the lower-resistance side will "steal" flow, and second, actual total flow should go up as total system resistance is reduced.
Anyway, just supposing for a moment that these two factors cancel out and the flow is simply halved, how would a un-flow-hungry block, like the 6000 do with, say, 1.6 LPM? That's not lots less than the half-gallon a minute I've always thought of as a bare minimum, so maybe it's fine? Actually, given the restriction difference between the Cascade and the 6000, maybe he actually will see a half GPM.

I'd already questioned a parallel setup in that thread and am now trying to figure out if I'm wrong.

[Etacovda]

Id imagine by putting the flow in parallel that you'd be adding more restriction, by adding greater tube surface area (and extra fittings, more importantly)

Surely you're using all the head pressure the pump can give, or else it would be flowing at full manufacturers rated lpm, eh?

I guess Bill can answer the 0.5gpm question - what would be interesting would be to know the pressure drop on the 3 commonly thought of 'performance' blocks (maze, mcw50, sp fusion)

at any rate, for a silent set-up id be more concerned on radiator surface area + fan speeds than anything else.

[BillA]

the MCW6000 curves include 0.3 and 0.5gpm data
all wbs hake a progressively greater hit as the flow drops below ~0.7gpm, the MCW6000 is just less (comparatively)
AND dropping the coolant velocity in the tubes is not beneficial at all (parallel rads), it just drops the head loss and the dissipation
- it is extremely difficult to get system gains from a rad at the expense of the wb
run some numbers, its clear

[Cathar]

Huh? I was and never have been a proponent of parallel flowed system setups.

Strange how people will often interpret what they want...

[Nickd]

I've read many threads similar to this, and generally the advice is to do the math.. The math I don't have a problem with, except for the fact that I'm finding it incredibly difficult to find dP Vs Q curves for different waterblocks.
I've found Bill's Thermochill testing on OC, and C/W Vs Q charts are available for just about every waterblock ever conceived, but to make any sense of the numbers I need dP_Q curves for the WBs as well. Could anybody suggest where I can look to find such info please???
The only place I've found at all is Cathar's thread about " Common Pump Flow vs Pressure vs Heat Comparison" which gives curves for the WW and Cascade. Any data in tabular form that I can graph myself or do I have to estimate the data from these charts. Can I find any such data for MCW6000s etc???

/*Edit Found BillA's Thermal-Management-Testing data. */

[Risky]

Originally Posted by Etacovda

Id imagine by putting the flow in parallel that you'd be adding more restriction, by adding greater tube surface area (and extra fittings, more importantly)

Well by that logic, bigger hose would be more restrictive than smallbore

I would imagine that parallel, or part parallel would reduce restrictrion, but also reduce cooling as the coolant flow through the blaocks would be reduduced. In my case I run parallel from the CPU to the res (dual outlets), and may a pair of disk blocks on these parallel out-lines, but then I wouldn't be aiming to maximise cooling on the drives, and would be more concerned about keeping up the flow rates through the CPU block.

[bobkoure]

From that thread over on silent PC review

Originally Posted by bobkoure

Let me try so re-say that so you can tell me that I understand what you are saying.
Say you have a set length of tubing, two radiators, two blocks and one 1048.
Setup entirely in series you get some flow rate, say X (no idea what it really is)
If you set this up in parallel, not only is total flow higher than X, but the flow in each branch is greater than X, meaning your total system flow has gone to >2*X.
Do I have this right?

He tells me that I do have this right: flow in each branch is over 2X.
I have never tried this experiment and am withholding any judgment until I do. I'm wondering if anyone here has actually tried this one - with a low powered pump...?
I've tried a number of experiments with very-undervolted fans and sometimes got results that didn't fit my "mental model" of what should occur, so there may be some unexpected things here, too - and this might be particularly interesting information for any waterblock manafacturer working on capturing market share in the "ultra low noise" (AKA "German") market... And maybe it's cold fusion (meaning some un-thought-of-effect fools experimenters).

[HammerSandwich]

Originally Posted by bobkoure

He tells me that I do have this right: flow in each branch is over 2X.

I haven't tried such a setup, but it would seem to defy the P-Q curve.

EDIT - complete brain fart on the numbers. Need coffee. Stand by the above and will rework example in a bit.

[bobkoure]

Originally Posted by HammerSandwich

I haven't tried such a setup, but it would seem to defy the P-Q curve.

My thought too - but if he's right, there's something interesting going on. It may be a special case in that it only applies to low-flow, low-head pumps, which would explain why folks like Ph haven't observed it (although it would still be very useful info for the low-noise crowd) and it may be bad data, I dunno...

[BillA]

what evidence was provided ?
when an impossibility is described why continue ?

[HammerSandwich]

Okay, trying again...

Suppose we have dual CPUs cooled with identical blocks, and we change from a series to parallel arrangement. If total system flow doubles (so same flow per block), pressure loss across the blocks drops to 1/4 of series'. No surprise: per-block flow will increase, pump allowing.

Doubling the total flow again would give the same pressure drop as with blocks in series. At this point, the pump would provide 4x the flow at the same pressure loss when compared to the series setup. Only with a magic pump...

So Snutten's claim of >2x flow in each block can't be correct. And that ignores increasing the radiator's pressure drop by 16x!

[pdf27]

Originally Posted by HammerSandwich

So Snutten's claim of >2x flow in each block can't be correct. And that ignores increasing the radiator's pressure drop by 16x!

Not sure you aren't misinterpreting him there - my reading of his claim was that the flow through either water block in his system is higher when the two are plumbed in parallell rather than in series.
This will of course depend on the pump and block P/Q curves, but is presumably possible for some combinations. It would certainly be interesting to play with some numbers to see if it should theoretically work (ignoring the seperate practical problems!).

Note: his SPCR profile states he's Swedish, so there may be some language difficulties here.

[HammerSandwich]

You're right. I just went back to the SPCR thread to confirm. I blame bobkoure's "He tells me that I do have this right: flow in each branch is over 2X."

[Nickd]

For two blocks in parallel, the pressure drop across them must be the same mustn't it? Flow rates will be different depending on the resistances of the blocks. So for a given pressure drop you can calculate the flow through each from each individual block's dP_Q curve. Total flow across the blocks in parallel will be the total of flow through A plus flow through B.
Same as voltage differential across two resistors in parallel and the currents being added to get the current in the circuit - right?

[pdf27]

Pretty much, yes, but the battery/resistor analogy is a bit too simple for the overall situation. You probably couldn't do the resistors in paralell calculation as the pump doesn't have a constant head over a wide variety of volumetric flow rates.

[bobkoure]

Originally Posted by HammerSandwich

I blame bobkoure's "He tells me that I do have this right: flow in each branch is over 2X."

Here's a direct quote:

Originally Posted by snutten

You said it better than me, bobkoure. Flow in each of the two loops superseed the flow in the single original loop. Total flow is more than double.

I was very explicitly trying to not mis-quote him. This was in response to my re-stating his earlier post (in terms of serial flow rate as X). How did I misrepresent what he said?
Anyway, don't trust me - go read the thread for yourself (and please don't give me a hard time about presenting procooling as a bunch of reasonable guys )

[Etacovda]

Originally Posted by Risky

Well by that logic, bigger hose would be more restrictive than smallbore

Good point, doh
Hmm, dont know what i was thinking there; I think i was fixating on adding Wyes + extra tubing.

[HammerSandwich]

Originally Posted by bobkoure

How did I misrepresent what he said?

The word "each" got me. No problems once I caught up with the SPCR thread.

[Nickd]

PDF, I was trying to model the systen P_Q curve for two blocks in parallel. Can then compare this to the P_Q curve of the pump and calculate flow rate through the system, and from that work out the pressure drop across the parallel pair. Once we know the dP across the blocks we can go back to the individual WBs curves and find the flow through each branch. Obviously most of the flow would go down the least restrictive path, and this would be shown by the above method.
Can someone confirm that I'm going about this the right way???
I thought it would be interesting to actually model the above scenario to see what it shows. We can take a WB, length of hose and a rad and model this as one restrictive block (same as we work out a P_Q curve for a series system) and then model two of these units in series and in parallel.

I think there might be something in BKs reference to "with a low powered pump" though. Might be that having all the above in series is sinply far to much for the pump to handle as it is operating very inefficiently at the end of its P-Q curve. By running these loops in parallel you'd be working the pump much closer to its point of maximum useful work (where P*Q is max - hydraulic power) and so get the best out of the pump. He'd be way underpumped for running two waterblocks, two rads and metres of hose with a 1048. Find it weird that his arguments were based on Cathar's "How much Pump Is Enough..." article. Methinks he was way underpumped for a full series system (as the article shows!).

/*Edit Thinking about what he said, he said that Cathar's article proved that parallel was best - so maybe he had worked through the simulation and realised that if he was stuck with his 1048 and couldn't get a bigger pump then he would in fact be better running his loops in parallel */

[Risky]

Originally Posted by bobkoure

From that thread over on silent PC review

He tells me that I do have this right: flow in each branch is over 2X.
I have never tried this experiment and am withholding any judgment until I do. I'm wondering if anyone here has actually tried this one - with a low powered pump...?
I've tried a number of experiments with very-undervolted fans and sometimes got results that didn't fit my "mental model" of what should occur, so there may be some unexpected things here, too - and this might be particularly interesting information for any waterblock manafacturer working on capturing market share in the "ultra low noise" (AKA "German") market... And maybe it's cold fusion (meaning some un-thought-of-effect fools experimenters).

I don't see there this comes from.

If the flow rate in the series sections of the loop (Lets say A and B) is X then the flow rates in the two parallel sections (lets say A and B) will sum to X. If the flow rate is the same in the two parallel section then it will be half that in the series section e.g:

A + B = X
If A = B, A = X/2 not 2X

Just remember flow is volume of water per unit time, not the velocity of the water.

[pdf27]

Looks about right to me.
For the block P/Q curve you should be getting total flow rate equals the sum of the flow rates for the two blocks at whatever dP value you're using.
From this, you can set the operating point of the pump (point where the two P/Q curves intersect). This then gives you the actual dP value for the blocks (after allowing for radiator, tubing effects etc.), which in turn gives you the flow rate through each block.
You presumably can already calculate the flow rate in series (just add the dPs rather than the flows), giving you the block flow rates in series and parallell for this pump. Getting performance figures from pH's testing is dead easy at that point. It's also probably worth doing as there are a number of hardcore silencers about who would accept a fairly minor temp rise in exchange for a marginally quieter pump (which it might be if you're no longer operating at one end of the flow curve).

Just be careful with your "one restrictive block" analogy - you may well end up running two rads in parallell, which is unlikely to model most situations accurately.

[Nickd]

Thanks PDF.
That's what I thought. Looking at the SPCF thread they are talking about running two rads in parallel. From what BillA said about reducing flow through a rad though I'm not sure its a good thing to do. Will have a play with the numbers.
Reason I'm thinking about it at all is I've got a WW and GPU and NB blocks from D-Tek. Was planning to run the NB and GPU in parallel from the two outlets of the CPU block. CPU would get full system flow then NB and GPU half each. Wanted to prove to myself that this is the right way to proceed.
Cheers, Nick.

[bobkoure]

Originally Posted by HammerSandwich

The word "each" got me.

Yep - you're absolutely right - sorry.
Should have been either total flow over 2X or each branch over X. D'oh!

[pdf27]

Yeah, I'm rather unconvinced about the rads in parallell. I'd guess you'd get a better performance/case volume ratio for bigger single rads, and if you get any air flow in series the second radiator will be pretty close to useless.

[Long Haired Git]

Originally Posted by Nickd

Thanks PDF.
That's what I thought. Looking at the SPCF thread they are talking about running two rads in parallel. From what BillA said about reducing flow through a rad though I'm not sure its a good thing to do. Will have a play with the numbers.

Check out BillA's radiator test article on his site and do the math. Rough'n'ready testing by Cathar as per postings at OCAU showed serial was marginally better for his setup of the time.
Someone else also did some basic testing (for dual 80mm rads by memory) and measured that there was zero overall flow rate difference between rads in parallel and rads in series once a G4 was plumbed into the system. As logic says velocity inside each rad is halved when in parallel, he wisely chose serial.
Easier to plumb too.

Originally Posted by Nickd

Reason I'm thinking about it at all is I've got a WW and GPU and NB blocks from D-Tek. Was planning to run the NB and GPU in parallel from the two outlets of the CPU block. CPU would get full system flow then NB and GPU half each. Wanted to prove to myself that this is the right way to proceed.
Cheers, Nick.

This is what I was planning to do too. Don't know PQ curves of those blocks, but my home-made NB and GPU blocks are not restrictive at all compared to the LRR. Main reason to do it, though, was to avoid the mega-tight-turns to get from CPU to GPU and then to NB. With dual paths after the CPU, I can take the furtherest-away-outlet to the NB and the closest outlet to the GPU and have a more relaxed radius for both.
Cooling for both is impacted by reduced flow rates, but I don't give a toss and it may raise flow rates for the CPU slightly by reducing overally head.
I was going to make a rough-and-ready HDD cooler, not because it needs cooling, but simply because I will use this to balance both "sides" of the flow to ensure the LRR flows evenly on each side.
Join backup up with individual inlets to a res, then a nice short 16mm tube to the eheim 1250. Oh, rad between pump and CPU.

However, was going to do plenty of flow testing first.