Flow and pressure in serial/parallel loops
 

Hey guys,

I am new to this forum. I've been watercooling for six months now, and my first computing forum was extremeoverclocking.com (ECOF). I joined it by mistake :shrug:

My roommate uses Folding@home for tom's hardware folding, but we use EO for their graphs and plots of finished computing. I thought I was joining Tom's Hardware Forums at the time.

Anyhow, I have a few questions.

The current watercooling setup I have is cooling a P4 3.2 E0 Prescott and an Nvidia 6800. For the CPU I am using the DangerDen TDX 3/8"ID fittings(at the time the Swiftech 6002 was out of stock) and GPU the ThermalTake Bigwater 1/4"ID fittings. My original setup used the Via Aqua 2600 pump and a 9"x6" heatercore. Initially my blocks were in serial, but that caused the impellor to create too much noise due too much head.

Since then my original pump has failed (outlet barb leaked due to overtorqing)and I am working on a new case. I'll be using a Hydor L30 pump with a Dual 120mm heatercore.

I would like my CPU to get the best flow, should the blocks be in parallel or serial? Is the loss of flow due to being in parallel greater than the loss due to the restrictiveness of the 1/4" ID size of the GPU blocks?

Approximator to the rescue!
I've used a MCW5002 as the GPU block. Don't have PQ for TT Bigwater, nor C/W. Anyone?
CPU = 70W and GPU = 30W.

Now all of this below is plus/minus one trillion degrees because its all interpolated graphs and guesswork and it is combining testing results with different assumptions and bunch of other stuff that has BillA groaning and Les vomitting blood, and ScienceWiz slitting his wrists. Don't care, it gives a good guide as per above.

xbitlabs indicates a 6800 is nearer 40watts
http://www.xbitlabs.com/articles/vid...v-power_4.html

I would think that it is far more helpful to calculate the prsure drop by analyitcal methods. I would of thought that the big water is a maze type block with loads of turns. I'll have a look at it this evening.

The 1/4 inch tube is going to make a huge difference isn't? Espeically the borda head loss in the y splittler.

tried to calculate the pressure drop in the gpu block form first principals but getting a high answer. Guesimating k=0.2

head loss from pipe fiction seems low (or order 2mm) but bends add shed loads of resistance.

The GPU block is here http://www.xoxide.com/thermaltake-bi...ter-block.html

I made a different mounting bracket for it. First and last time I'm buying a cheap block.

Right now I'm running in parallel with the L30, load temps with 100% CPU and GPU (running World of Warcraft GPU/100% CPU/50% and Folding@Home CPU/50%) temps are CPU 44C, GPU 42C, Water 25C, intake air~20C

Idle Temps are CPU 29C, GPU 35C, Water 23C. +/- 1C on temps, using external probes for CPU, Air, and Water, Onboard for GPU.

Thanks for your help guys, just what I was looking forward to when joining the Procooling forums. :)

Here's a pic of my current temporary setup
http://img256.imageshack.us/img256/3...mporary8ku.jpg

You're right. In series the 1/4" GPU block will restrict the flow to the CPU block. But, in parallel the 1/2" CPU block will get most of the water.

The solution isn't a bigger pump (though that almost always is a positive).

The solution is a new GPU block which will function efficiently (ie. cool the most) with the flows you anticipate with the new pump.

Ya kinda have to look at pump curves and block restriction curves and see where you're at and where you want to go. Don't forget to figure in pump losses for tubing, fittings, rad, and ras

As i see it, parallel/serial isn't the question.

My figures our out because I forgot to add in the restriction caused by 3/8" fittings on the blocks and rad.

That said, something is wrong with my approximator. The water temp is either too high for serial or too low for parallel.

Sigh.

My radiator has 5/8" fittings. The 1/2" tubing is stretched over it.

It looks like I just might have to shell out $70 for a high flow GPU block, Perhaps the Dangerden Maze4? This was what I was trying to stay away from seeing as how I'm strapped for cash.

problem is that the gpu blocks restricitiness can not be easily analytically quantified. I attempted using pipe pressure drop equations (darcy with rectangular to cylinderical diameter correction, calculated re=2800 @ 1 lpm so used colebrook for a punt) but the bends in the block proved to be where the restrictivness came from. All gpu blocks are going to be like this as they are maze style and nobody has done any testing on them.

Rather than test doing seris with a big ass pump to remove any vagueness is a sensible approach that can't fail to produce satisfactory results.

Git (if i may call you that) have you considered that a) the block approximation you are using is wholy inaccurate and your calculator does not include bends in pipes which have a k factor of 0.19 per bend. Also all these tube connectors are bound to cause some kind of pressure drop.

Besides your approximator is approximate anyway and you can always blame crap mobo temprature sensors.

http://overclockers.com/articles373/wbsum.asp
http://overclockers.com/articles373/basum.asp

Do you have TDX PQ data or are using another block for estimation?

Here's my guess for a series loop:
System1:
--TDX was derived from BillA WW data shifted to coincide with JoeC head loss data @ 1 GPM
--1.5 m straight 1/2" ID tubing
--MCW55 with 3/8" OD (~1/4" ID) fittings (assuming fittings from Swiftech data)
--77' Bonneville HC with 12.83" ID fittings ....slightly longer than a Caprice (~2"), but not much variation in head loss.

Too lazy for parallel data.

Bingo - I just use BillA's testing results, and yes I used the LRR WW and not a RBX. Should of said so. I don't model blocks, but instead take measured results from other's testing. I can, however, model a conversion to 3/8" barbs pretty easy.

I would argue that the restrictiveness of the GPU block is what is keeping the CPU temps so good. Swap in a less restrictive GPU block, and more flow will be avoiding the CPU block and its temps will go up...