dual processor cooling

[chrubb]

Hi people!

As a first water cooling project I am looking at installing WC in a dual AMD MP1900 tower running my penguinware.

The main reason for this post is to see if any of you have setup a dual processor setup before.

I am concerned about the heat loading between processor 1 and processor 2 - should I :

1. have two seperate wc circuits, one for each processor
2. install a heat exchanger between processor 1 and processor 2

or

3. not worry about it

I am planning for a water cooled PSU as well and this is going to dump a fair amount of heat into the system - do you need to increase the volume of liquid in the system according to the amount of heat you are trying to shift?

sorry if these are bleedin' obvious questions, but as I said this is my first wc project

[adwhitworth]

If your Radiator is going in your case? What is the max size it can be (1x120mm, 2x120mm or 3x120mm). If you have a double one it should cool it on one Rad,

[Darv]

I don't think you would have a problem of having the CPU blocks in series. You could use Y splitters to have them on separate lines but you would probably loose performance because of the lower flow.

[chrubb]

wow! quick response...thanks

I was toying with the idea of external radiator(s),pumps etc as this is going to be a (semi) permanent installation. I think I'll try running the circuit in series and see how it pans out - part of the fun of this sort of thing is the experimentation :D

I'm a bit concerned about the noise level on the pump (I have a couple of fish tanks and the pumps on them can be noisy) - any recommendations ?

Thanks for the help.

[Knipex]

External Pumps and rads can be a good idea in some situations but you have to be carefull how far away the pump will be from the case. I know Pug runs an external rad setup using a semi permenant bracket but I have to admit that my prefference is for a fully self contained unit, thats just a personal thing but make sure the pump and rad are not far from the case.

I agree with you on running your CPU's in series, in my humble opinion when it comes to watercooling running circuits in parallel is a big NO NO. In ideal circumstances you would be halfing flow to each CPU which is in its self not desireable. This could be counteracted by using bigger pumps and pipes but as the reall world is not ideal, flow resistance in each side of the circuit would not be identical so flow could actually be split 60 / 40 or perhaps there is a tight bend in one side of the circuit so flow could be split 70 / 30 or even 80 / 20. Something to keep away from.

On the pumps front. My current pump is damn near silent, the 120V papst fan @ 7V is definately louder. Its an Eheim pump by the way which is pretty much the standard in watercooling. Just mount them on foam to eliminate vibration and you will not hear it a few feet away. There are some nice new 12V pumps about but I have never used these so I cannot comment on either there ability to move water or noise levels.

What do you plan on watercooling ??
Just the CPU's or is it going to be a bit more ??

Either way welcome aboard and have fun..

[Nexxo]

Don't worry about the extra load of a second CPU. Difference in water temperature before entering and after exit from a waterblock tends to be a few degrees Celcius at most. Many Dual CPU setups simply link the waterblocks in series, and still find the capacity to watercool a graphic card using only a BIX for radiator.

If you want to see a dual WC rig in progress, visit the thread of my project "Metaversa" :here: (http://forums.bit-tech.net/showthread.php?t=29556&page=1&pp=20). B)

[Monkeyharris]

i mount my ehiem by suspending it from rubber mounts that you can buy especially for pumps, usually you attach them as feet but i found suspending them was easier for my particular setup. i forget where i got them from but i now cant hear my ehiem 1250 at all over my 7v papst 120mm fan, so its as damned near silent as i will get i think? B)

Pat

[chrubb]

amazing response guys.

nexxo: what can I say except WOW! lots to think about there (especially the 5" lcd screen - hadn't even vaguely occured to me)

loads of info from you guys - I think I'm ready to get on with the detailed planning now and then the painful bit - buying the gear.

I'm sure I'll be posting some stoopid questions soon.


once again thanks for the help all of you - wizd is a great forum

:D

[dutchcedar]

I understand and agree with the comments favoring a system in series rather than parallel, however, there are opportunities to use a parallel system. I've found that when using a large supply line off of the pump, it can indeed be split into two or even three smaller lines off of a manifold quite efficiently. This can be an especially efficient method, for example, when the waterblocks use small tubing (3/8" or smaller) and the supply at the pump is larger (1/2").

You can see how I applied this methodology here: http://www.athlonxp.com/modules.php?op=mod...ews_BaDass.html (http://www.athlonxp.com/modules.php?op=modload&name=Diner_Wrapper&file=index&req=ShowFile&file_wrap=html/reviews_BaDass.html)

In my new build, I'm using a pump line of 5/8" I.D. off of an Eheim 1250 going into the radiator to a 1/2" NPT fitting (5/8" barb). The radiator has three outlets (1/4" NPT) and I have 3/8" and 1/2" barbs for them. It looks like I'll only be using two outlets with 1/2" barbs with each going to a CPU. The third might be used to cool a graphic card later, but I doubt it. Since I'll be using waterblocks with dual return lines, they won't have to use "Y" adapters, because they will merge in an anodized aluminum manifold with four 3/8" NPT inlets (3/8" or 1/2" barbs) and a single 1/2" NPT (5/8" barb) outlet to go back to the pump. An extra outlet in the manifold will be hooked to the reservoir.

I hope this doesn't confuse matters... but don't write off parallel systems immediately. IMHO, they're more fun to build, if nothing else. :rolleyes:

[Pug]

Regarding the three options you gave, I'd pick 3.
3. not worry about it

To reiterate what's been said, a series loop will probably only show a degree & a half or so for one proc over the other but as you say, experimentation is key (& fun :D).

As Dutch rightly pointed out, a parallel split often works well when integrating matching small-bore blocks into a large bore loop.
It's an approach I advocated to a guy from Spode's Abode a year or so back wishing to use a pair of original 10mm barb Evos with a 1/2" setup -
http://freespace.virgin.net/water.cooled/Pics/MattDual/_dualCase.jpg (http://freespace.virgin.net/water.cooled/Pics/MattDual/dualCase.jpg) http://freespace.virgin.net/water.cooled/Pics/MattDual/_dualEvos.jpg (http://freespace.virgin.net/water.cooled/Pics/MattDual/dualEvos.jpg)

It's a tricky approach for anything other than perfectly matched blocks on exactly the same level (which is rare for an SMP board) with exactly the same tubing length on each branch though, due to the inherent tendency of water to follow the path of least resistance, like electricity. The worst case scenario for this is that one block may be comparitively starved of flow, creating problems. :blink: (I use serial, personally)

PS. Welcome to WizD. B)

[Pug]

Slightly irrelevant but interesting all the same, I heard an interesting theory (http://forums.bit-tech.net/showpost.php?p=353242&postcount=7) from a heating engineer in a thread on Bit-tech as to running household radiators in parallel. It was a first in, last out (FiLo) setup but did rely on three or more components being present so I don't know if the principle would still apply (but doubt that it could hurt to try to adhere to).

It's something I'd definitely like to look further into with time. :)

[Pug]

Regarding your last point,
do you need to increase the volume of liquid in the system according to the amount of heat you are trying to shift?

This is unlikely to have as much effect on peak temps but will lenghten the time taken for thermal equilibrium to set in. In other words, if your comp is generally crunching away 24/7, you won't see the difference but if it's normally idle with occasional bursts of CPU activity, you may get lower temps overall.
{Fwiw, I run about 1/2litre in a 240 Rad loop (or however much it takes to fill the system with an Aquatube res), whether I have 2 or 6 blocks present.}

Hth. :)

[scopEDog]

I agree...dont dismiss parallel yet, as there are certain applications where it could be useful. I see alot of people using 2 cheaper pumps in the future for a nice parallel setup. PRIME was my first crack at it...turned out pretty good me thinks :) Link (http://forums.bit-tech.net/showthread.php?t=32644)

[Nexxo]

Yeah, parallel works, but you have to be WC savvy to pull it off... ^_^

Regarding 12V pumps, I can supply a bit more info, as this has been an area of special interest for me, and the subject of way lots research...

There are several types, most of which are good:

Johnson CM10P7-1 a.k.a. Jabsco 59500-0012: In-line only. 1/2" barb fittings. Continuous rated. Quiet, small, sturdy, reliable magnetic drive pumps. Get warm in operation (53 C) but do not pass this on to the coolant. Good pressure and flow. Life-span of the motor brushes is 2500 - 4000 hrs. Verdict: Perfect, but life span could be longer.

Johnson CO10P7-1: the mechanical drive version of the above. Much cheaper, but prone to leaking. Verdict: avoid.

Johnson CM30P7-1: bigger brother of the top ones. 3/4" barb fittings. All the goodness, more of the power! Hideously expensive but indestructible. Alas, lifespan of motor brushes 2500 - 4000 hrs. again. Verdict: great, but price v.s. lifespan would make me want to look elsewhere now more options are available.

Swiftech MCP 600: In-line only. 1/2" barbs. Small, ridiculously powerful and good flow, quiet, aluminium housing (think: polishing...), magnetic drive. Does not get very warm. Brushless motor means a lifespan of 10.000 hrs.+ Early models had some impellor problems but these have been resolved. Verdict: my current favorite.

Eheim 1046 12V version (both Aqua-Computer and Innovatek do these). A range of quality barb adapters already exist due to the popularity of their 240V brothers, which have been the pump of choice for most watercoolers. Early versions appeared to have some problems which are now resolved. Quiet, reliable, not the smallest, do not get warm. Some magnetic interference has been noted (due to their plastic rather than metal motor housing), but metal surrounds are available. Best for small diameter setups. The Aqua-Computer version has a useful electronic flow monitor, and can be interfaced with the Aqua-Aero which (once it comes out... :angry: ) will be the holy grail of watercooling anywhere. Verdict: old faithful that will never let you down.

Hope this helps.

[Pug]

Another range of 12Volt pumps to add to your list Nexxo -

http://www.wizarddesigns.co.uk/imgs/ap900.jpg

and its brethren, the AP700 and AP1500

Verdict - Solid construction and er, it works?
Unfortunately, this (http://www.wizarddesigns.co.uk/miscPics/ap900test.jpg) is the extent of my testing with 'em so far but I'll bung a thread up in product enquiries when I get the chance to set it up against the AquaStream.

[Nexxo]

Hmmm... Grolsch... -wait, we've been there before, haven't we? :p

But yes, that looks like another capable little pump, a bit like the Eheim 1046 but more compact.

[Kickass]

Ok people this will probably help you as ive got a dual cpu set up wtaercooled

My specs: Dual mp 2400+ 512mb ecc mem 550w psu
External Koolance Al Exos system

And basicly its external 2 tubes from the box on top and i am currently running in series! basicly ive been told this is really bad as im seeing a 5 degree temprature difference when i give it load

Idle 30 : 34
Load 34 : 39

Can get lil higher it has 3 settings i can keep it down to about 32 : 37 on level highest one!

Well anyway i would reccomend running with Y splitters or manifolds m8 as i been told mine dodgy. Also with having a Watercooled psu thats gonna have to go After the two CPU's! well i would post a pic for u but i cant my webspace is down :S

message me on msn xploda22@hotmail.com if u got any questions

[Nexxo]

There's two points with your temps though:

One is the question of how they are obtained. Mobo integrated temp sensors are not very reliable, and the observed difference may be mainly measurement error (i.e. due to individual sensor differences, rather than actual temp differences). This measurement error is likely to be smaller with probes, but even then you need to calibrate them, and small differences in mounting position can affect reports.

The second is to do with CPU load. Contrary to popular belief, SMP does not neatly divide the workload 50-50 between the two CPUs. Generally most of it goes to CPU1, with the overflow of tasks being mopped up by CPU2. This is particularly so with programs that do not explicitly exploit dual CPU processing. So you may want to check which CPU runs hotter: my bet is it's CPU 1 (the one which your mobo manual will state MUST be in place for the setup to work, even if running on one CPU. The sockets should be numbered on the mobo). CPU1 generally does more of the work, so it would run hotter.

[Pug]

*Moving to general watercooling*
Mind de doors...