'T' fittings suck because the flow has to make a 90 degree bend. This isn't natural for water to do and causes a great deal of turbulence, loss of energy. The waters momentum wants to maintain is forward velocity. I've actually seen water flow through an open T in 2.5in PVC and not spray out because the flow rate was so great. Keep in mind how water will flow through the pipes of your system. A large manifold will slow the rate down so its *probobly* the best solution. A series of 'T's will kill your flow rate and the flow will be different in each branch.

My solution was to build a CPU block with 1/2 intake an 4 x 3/8 exausts. This way the block itself is a manifold. If you are cooling a dual system them Y the intakes.

Y blocks are almost always better because you are keeping the water as close to a straight line path as possible.

Manifolds, unless they actually have an open area like a hollow waterblock are just as bad as multiple T fittings, because most times manifolds are one large bore with smaller splits at 90°, like so...
What we want for smooth water flow is a Y splitter, like so...

Not really. Performance drops off exponetially at lower flowrates while the heat from a second CPU is linear. Take a look at BillA's data. Unless you have amazing flow, splitting is going to take a lot more then .6C off.

And if you had such flow, then you'd have such a high coolant velocity that the .6C figure would be much less anyway. theres really no reason to ever split CPU flow.

More on Manifolds...
 

Well, I guess that did stir things up a bit, kicked over onto another page and everything :D

Lets see...

True, but the local hardware store has PVC galore, I haven't seen any "Y" fittings. I find I like to have my hands on the hardware when I'm buying things.

The manifolds I'm planning will have a LARGE open space. What I'm planning is a piece of 1.5" PVC pipe, 3-4 inches long, capped on one end and plugged on the other (in case I need to clean it out). The inlet will be whatever size comes off the Rad (probably 3/4") and probably come in via the cap end. The outlets will be around the sides, located to make the best shots at the target devices. The exact configuration will have to wait until I have the hardware in hand. However it sounds like I'm meeting your description of a 'good' manifold.

I am planning to do a high flow rate system, I am thinking in terms of an Eheim 1250 pump or equivalent. (I'm looking at a Danner Pondmaster "Pond-Mag 5" which claims higher flow, better head, slightly smaller size and similiar wattage. Anyone have any experience with these?)

I don't see using the block as a manifold as being a good solution in my case - I want the blocks to be as identical as I can make them, and it would be a challenge to balance the flows so as to make them all even at best. I'd probably have to many branches as well. Also I plan on making the 'drive loop' from much smaller ID tubing than the CPU loop, so putting that in series with the CPU would impose serious restrictions.
Lastly, the mobo I'll be using doesn't have the holes to do a 4 bolt mount, so I have to do a clip-on. This cuts into the amount of real estate I have to put barbs into, so I don't think it's on.

FWIW dept, I am planning to make a high flow block with I/O at opposite corners, doing a serpentine maze with a wider center section containing lots of turbulence pins. Sort of a cross between a Swiftech and a DD Maze 3. I know this looses the benefit of direct jet impingement, but I think it will work OK anyways.

This sounds like you are suggesting I SHOULD plumb the CPU's in parallel, which appears to be the opposite of what redleader is saying - I feel confused :confused:

FWIW again, 3rotor on his website seems to be running his dual rig in parallel though it is hard to be certain from the pictures.

Gooserider

Every system is different so to say you should do something 1 way all the time is not productive. Try mocking your system up in a sink with the blocks in parallel then in series. Measure the flow you get each way. The best way to optimize any system is to experiment with it then fix the problems.

If your pump can generate enough pressure by all means run them in series to get the max flow through each block. But chances are that 2 blocks are too much resistance for most pumps and will cut the flow too much. At first it may seem running in parallel may cut the flow in half. But since the systems resistance will be lower the pump will push more water increasing flow. So in actuality you may see some drop in flow for each branch but the overall system will have more flow. What works for you though will be whatever your experimentation leads you too.

The problem with this is that you do not have the highest flow rate where it counts... namely through the waterblock.

Doing it in the tub....
 

Makes sense, though I'll probably use the tub since it has better capacity for flood control :dome:

The big question as I see it is that the simulation would lack the thermal input from the processors, so it might be a challenge to figure out how much of an increased total flow volume from running parallel it would take to compensate for the decreased volume going through each individual block.

To save typing;
Op = total output of parallel hookup/time;
Os = Total output of series/time;
Bp = volume through ONE block in parrallel/time;
Bs = volume through either block in series/time;

Now correct me if I'm wrong, but Bs = Os in a single branch system as I understand it.

In a parallel system, Bp = OP/# branches, assuming that all branches have equal flow resistance. I.e. in a two branch system, each block would see 1/2 the total volume pumped in a given time.

Also, within diminishing return limits, the higher the volume flowing through an individual block, the more heat it will remove from the block. However, the hotter the water is entering the block the less heat it will remove per unit of volume.

The issue is mostly a question of in a series setup how much hotter the water going into the second block would be on account of it's trip through the first block, and thus how much hotter the second CPU will run. In addition there is a question of how much the flow would be reduced by having all the water having to pass through both blocks.

The series advocates maintain that the temperature difference between the two blocks would be nominal since each volume of water would only be picking up a little heat in the first block. In addition they feel that making all the water pass through both blocks means that each block will see more volume and be cooled more than the flow would be increased with the lower resistance parallel setup.

The parallel advocates maintain that the greater total flow due to lowered resistance will make up for the fact that each block will only see part of the total.

Now, boundary conditions...

Case 1: Op < Os - The series folks win clearly, as there is no increase in flow with the parallel setup, and Bp = 1/2 Bs, which would obviously mean less cooling.

Case 2: Op > 2 x Os - in this case the parallel folks have it easily, since the flow through each block would be the same or better than the series flow, and the increased volume should mean the whole system runs cooler.

I would be suprised to see EITHER of these cases, but would expect to see something in between, where Op = 1.? x Os. This leads to the main question, how does one estimate the point where one crosses from case 1 to case 2? :confused: :eek:

I know I've been talking about having an additional low flow branch for the hard drives, etc. I think it is safe to leave that out of the overall calculation since it would be the same in either hookup. (actually it might get LESS flow in the parallel setup since the lowered resistance of the main branches would make the high resistance side branch less 'attractive'...)

I don't quite get your point Skulemate, there will be two blocks, and I don't know where else the water would go besides through them (barring leaks :cry: ) I am NOT talking about having a 'no cooling bypass loop'! If I split the flow, I would be splitting the cooling loads as well.

Gooserider

I wrote up a really long explanation with RHUs(Random Heat Units) and RVU(Random Volume Units)s but realized it was way to complicated to make much sence so I found a power curve chart. Lets imagine for a second the Maze 4 equals a foot of head. So figuring everything else in your loop equals 1.5 feet of head and you have an Eheim1048.

1 Maze4 115GPH Total Loss 2.5ft
2 Maze4s Ser. 85GPH Total Loss 3.5ft
2 Maze4s Par. 125GPH Total Loss 2.0ft
62.5GPH per loop

Cut the flow for the parallel system in half and you get 62.5GPH through each loop. A difference of 22.5 from series. Now I choose 1 ft of loss for the Maze4 at random. It would be a close call at this point what would do better parallel or series. Lets take this further through.

Lets say I decide I need a bigger radiator and now my base head loss is 2.0ft instead of my previous 1.5ft. So my new numbers are.

1 Maze4 100GPH Total Loss 3.0ft
2 Maze4s Ser. 62GPH Total Loss 4.0ft
2 Maze4s Par. 115GPH Total Loss 2.5ft
57.5GPH per loop

Now that the curve is flattening out we see a huge drop in flow for a highly restrictive system. The difference is a mere 4.5GPH. A little more restriction and the parallel may actually get more flow. Given the great increase in flow with parallel I choose it simply because now I have 53 more gallons per hour to cool other parts of my computer.

I've figured out 2 things in the past 5 minuts. Its crucial to check the curves for figuring out the layout of a system. And it would be freakin awesome if places posted numbers on headloss through their blocks.

Good show, Sevisehda.

Most pumps are used at the far end of the curve, but it's different for everyone. The average frow rate for a typical rig is ~60 gph.

With blocks in CPU you still want most of the flow to go through the CPU block, and that means throttling the flow to other blocks. By doing so, the total pressure should end up being slightly less than if the CPU block was by itself, which will result in a bit more flow, but still split.

Blocks in series put the pump further out of their energy efficiency range: gph for head, you end up loosing further than you already have.

I ran the calcs once for LiquidRulez, for pressure drops of a nozzle, and found that the rad/heatercore alone will create a significant pressure drop, throwing anything else way up on the curve.

Your RUA (random unit analysis) is a bit bogus, since the flow resistance of the various components will be dependent on the flow rate through the system.

And Ben, running a pump on one extreme end of the curve or the other means that it wasn't chosen properly in the first place ;)

True.

I'm gonna stick to the old rule of thumb: pumps in series, blocks in parallel, but the pump has to be considered in the equation.

why in hell would you do this?
Blocks in parallel need MORE flow. (flow is split)
pumps in series provide more head, not more flow.

Did you miss my last post. Sometimes more head means more flow.

2 pumps may push alot of coolant in parallel but in a restrictive system both will be held back severely. Placing them in series will allow them to 'throttle up' and push more water.

There just for explantions sake. Perfect Units for a perfect world.

Now I'm not a MechE but I've seen some formulas for drag and I take it the formulas for resistance will be somewhat similar. So since headloss is related to flow they could provide a graph simlar to the pumps power curve.

I'm not an idiot, but he implies it is a "rule of thumb" - in fact, it is more the exception to said rule

Good discussion sevisehda, your examples make sense, and look like reasonable real world numbers. I suspect (can't prove yet) that I'll be able to get higher flows than your example, but I think your reasoning still holds.

Pump - I'm currently planning on a Danner PondMag 3. It is smaller and less expensive than an Eheim 1250, and has better flow rate and head numbers if one is to believe the advertisements. I've searched the forums here, and haven't found anything really bad said about it.

Blocks - I'm going to make my own, but I'm planning to go for a very high flow design (at least as I understand how to get high flows) to get good cooling via 'brute force' of lots of volume, rather than fancy flow juggling tricks (like Cathar & Co.)

Rad - From the reviews I've read, discussions, etc. I've gathered the impression that most of the WC purpose designed rads have pretty high restrictions, and tend to be on the small side. I'm planning to use a heater core (supposedly less restrictive) big enough to put at least 2 92mm fans side by side blowing through it. I've just ordered an AMS CK1100B 20 bay cube case from Servercase.com, (195.00 w/o PSU) and figure on hanging the rad off the back of the case which I'll mod to have the two case exhaust fans blowing through it.

I'm thinking now that testing for comparative flow volumes with the hardware I'm actually using will be my only way of knowing for sure what will work best, and that will have to wait a while until I get it in hand.

Gooserider

You can look at it in more than one way ;)

Blocks in parallel *allow* more flow

Pumps in series allow more pressure and, as a result will give you more flow.

But you still have to throttle the blocks so that most of the flow goes through the CPU block.

this is true.
i just wasnt imagining blocks so restrictive that this would actually be a benificial scenario.

One other thing:
Blocks in Parallel do *allow* more flow - but thats TOTAL flow...so flow per block is not ness. higher.

Keep in mind that its not a "rule" only a possibility. A set of unrestrictive blocks would probobly fair well in a series system that wasn't getting close to the max head. However a look at the most modern blocks show a tendency to go toward a high flow rate at the cooling point. This means narrowing the path to increase flow however this also increases resistance.

Its important to either A) do the math on the best setup for your system or B) experiment in your sink/tub to figure what method will work best for you.

My method of testing head resistance is imperfect but its costs nothing and is fast. I hook up my pump(inline or submerged doesn't matter) outside. I make marks on the wall every cm. I connect about 10 feet of hose. Then I raise the hose until the water stalls. I mark this point as Max. Head. Then cut the hose at about a foot and connect that to my test piece and the other 9ft to the outlet. Then I raise it to max head. Take the difference and Bam.

I know this is imperfect because since the flow decreases so does the resistance so then flow increases(makes sense if you think about it). So my number is a little low for each block. I guess I could meause the height at which the pump puts out a gallon per minut(60gallons per hour) which would be closer to my normal flow rate, and hence give me a more accurate number but this would also take alot more time and I'm just to eager to build my sytem and not to test it for hours.

Trying to come up with a general rule of thumb for multiple blocks seems fairly pointless to me. It depends on the blocks and the pump.

Saying that blocks in parallel is always better seems particularly dumb.

Here's a graphs showing some pressure drop vs flowrate curves.

The 'Sim' curves, are graphs of equations that approximate the PQ curves of Iwaki MD20-R and MD20-RZ pumps.

The other four curves show hypothetical cooling loops consisting of a 2-342 heatercore, (single pass, low restriction) 6 feet of 1/2" ID tubing (resistance of the tubing is based on one continous straight piece) and one of the following:

2 White Waters in series
2 White Waters in parallel
2 MCW-5000's in series
2 MCW-5000's in parallel

No attempt is made to account for the added flow resistance of 'tees' or 'wyes' required for a parallel setup.

http://uffish-thought.net/wc-gifs/p-vs-s.gif

Two White Waters in parallel have about the same curve as two MCW-5000's in series, so the two curves nearly overlap.

For the Sim20-R, per block flowrate is:
WWS 7
WWP 6.15
MCWS 12.3
MCWP 8.75

For the Sim20-RZ, per block flowrate is:

WWS 8
WWP 5.4
MCWS 10.8
MCWP 6

In all of these cases, series blocks always yields better flowrates through each block. In the high pressure pump case, (similar to cheap pumps in series?) the flowrate advantage of putting the blocks in series is even greater.

Keep in mind that these numbers don't include the flowrate hit for 'wyes' in the parallel cases. Variations in flow resistance in the tubing is not accounted for either.

Looking at the C/W vs flowrate curves for the blocks, it appears that the gain in flowrate would frequently offset the higher water temperature seen by the second block in a series setup. The first block in the series combination would always gain performance from the higher flowrate of course.

References:
Bill Adams' White Water test data.
Bill Adams' MCW-5000 test data.

Edit: This is incorrect. See below.

The above data matches my thoughts on the matter - which is why i thought that your rule was bass-akwards.
I still do.
I think the "rule" is the exact opposite of what you said, bb2k, and the exception is what you have stated.
However, i guess you need to do some maths and plot some curves to find out whats best for your system.
But i bet you were thinking a WW would have been restrictive enough..but it looks like its not.

Interesting curves,
 

Seems to make the case stronger for the series blocks, but it probably still makes sense to do testing as well to get the outputs for the setup one is using. This would seem to me like it would be especially applicable in a case like mine where I'm making my own blocks and won't have any reliable reference data to look at.

I have another question I'm wanting to ask which is somewhat off topic, so I'm starting a new thread here:
http://forums.procooling.com/vbb/sho...&threadid=7079
At least part of it is somewhat related, in that I was wondering if anyone had data about how long it would take the CPU's to overheat if a pump failed so there was no longer water circulating in a system (No pelts or extreme cooling, just normal WC to ambient setup)

Assuming I had a flow detector that would detect the failure and send a signal to the system, would there be time for a graceful shutdown (ie 'shutdown -h now') or would I be safer to have the sensor trigger a relay that killed the power and slammed the system off (and worry about any resulting data problems later...)

I know that there are many variables, but lets assume a medium to bad scenario - Athlon, stock to moderate overclock (80-100 watts?) small to medium size copper WB, some airflow, but just what you might get from 1 80mm and the PSU, not enough to significantly cool things down. The shutdown must be fast enough to prevent damage to any hardware so that you could just replace the pump and power things back up.

How 'BAD' is a rotor type flowmeter?
 

Another quick question; I've seen comments to the effect that a 'rotor type' flowmeter is "highly restrictive", but no data as to 'how bad' it really is.

Can anyone post or point me at a source for data as to just how much restriction they really offer?

My understanding is that those units are the least expensive way to get data on flow in a format that can be handled by a digi-doc or mobo fan sensor.

Are there other alternatives that are comparable in cost and less restrictive?

I want to have some way of monitoring flow status in my system, at the bare MINIMUM I want 'flow / no flow' detection, ideally I would like some idea of how much flow is happening at a given momemt. (reference my prior post on disaster prevention)

Thanks,

Gooserider

Unless your over curious don't bother. Keep in mind that if the rad should fail the water will slowly increase in temp. There are plenty of hardpoint to add a thermal probe. If I were to put a probe anywhere it would be on the top of the CPU cooler(only possible on an all copper block, plexi is to insulative). This spot should be the same temp as the coolant inside the cpu block. If you have a fan failure the temps will rise slowly then you could shutoff your comp. Or if the pump fails the water inside the block will rise somewhat quickly and the probe would also detect this and shutdown your comp.

This is an interesting approach, I had thought of doing temperature based alarming, but thought flow based would be better (plus it gives me an excuse to get a flowmeter :D ) My reasoning was that I wanted as much of an early failure warning as possible, so that I would have time to do a graceful shutdown as opposed to having to slam the power.

You seem to at least imply that there would be a slow enough rise to allow for a graceful shutdown regardless of failure mode. Assuming the pump does a total crapout, do you know if there are any kind of numbers as to how long from failure to melt down?

As a second thought - In a dual CPU setup, would it be necessary to put failure detection on BOTH CPU's? Or would doing one be enough on the theory that the temperature is going to go up at about the same rate on both?

Gooserider

Calculating the pressure drop of two CPU blocks in parallel seems rather pointless to me.

The actual application would be composed of a CPU block, and a GPU and/or NB block, both of which should be far less restrictive, which dictates the use of a valve. I'll admit however, that the GPU and NB should be in series.

As I said, this also requires taking a look at the pump, because the "rule of thumb" isn't always right. Would like to see the same graphs with an Eheim 1250, which is more typical, and in line with the kinds of pumps that most water coolers use.

If you look at the curves for a second, it looks like you can substitute pretty much any centrifugal pump curve for the Iwakis... you're going to have different flow results, but the conclusions should be the same. I'd guess that the Eheim results are similar in character to the Iwaki 20R results... i.e. a less pronounced advantage to having the blocks in series, but an advantage none the less.

Some people have modded there pumps to have a small res right at the intake. The theory is centrifugal pumps need a large source to pump most effiecently. However since the pump output would equal input in a loop does this make sense? Does attaching a res directly to the pumps intake really increase flow or is it just a space saver?

[Edit]
The graph in this post is incorrect. See below.
[/Edit]

I don't know. I've seen the subject come up several times from people with dually boards.

I'd guess that a GPU block in series with a NB block, may well be more restrictive than a MCW5000. What's your basis for saying they, "should be far less restrictive"?

Only if you don't HAVE two CPU blocks! The system I'm working on will have a Tyan 2468UGN (Thunder K7) mobo. This is a dual Athlon MP based board, so I will have two CPU's to deal with...

Sorry BB2k, you need to go back up and re-read my original posts again. Perhaps I wasn't clear enough, though others seem to have gotten it.

The ACTUAL application I was asking about will consist of TWO CPU blocks, and several smaller blocks (exact number unknown as yet) for hard drives, possibly NB, probably not GPU (I'm not into fancy video, so won't have a GPU that needs it).

I was asking about putting the CPU blocks on one or two loops, with the small blocks on an additional loop. I only had one loop in mind for the small blocks, but if that doesn't work well, I might split them up a bit (probably 3-4 small blocks per small block loop)


What makes you think the smaller blocks would be less restrictive? I am assuming, and basing my design setup on the idea that they would be far MORE restrictive!

Here's why:

• My CPU blocks will use 1/2" plumbing. My small blocks will use 1/4" or 3/8" plumbing.
• The loop(s) for the CPU blocks will be as short as I can easily manage. The loop(s) for the small blocks will be far longer.
• The CPU blocks may end up in series or parallel, but they will be the only items in their loop. The small blocks will probably all end up in one or two loops depending on how many of them there are. (If I split them into smaller loops, there will still be 2-3 blocks / loop.)
• The CPU blocks I'm planning to make with passage cross sections (1/2") at least as large as the plumbing for low restriction. The small blocks will likely be made with 1/4" copper tubing, and lots of bends.

I can't see ANYTHING that would make my small block loop less restrictive than the CPU blocks!

Gooserider

Since you'll be making your own NB and GPU block, the actual flow resistance isn't known, but in what's commonly available, it isn't hard to see that even a DD Z-chip block would be less restrictive than a lot of waterblocks. If you design your NB and GPU blocks specifically to be restrictive, then that'll change everything.

I don't see 2 * Eheim 1250 in series, on that last graph, but I can see that Since87 did account for a certain loss, in the "serial" application, which is correct.

Otherwise, I have to agree: in a dual setup, putting the CPU blocks in series appears to be best.

You could run some flow/pressure tests on those blocks you're making, and recalculate what would perform best.