Sign in with
Sign up | Sign in
Your question

DIY water-cooled system?

Last response: in Overclocking
Share
July 13, 2010 10:41:20 PM

To start off, my last OC'd system stemmed from a Celeron 300A -> 450MHZ with an ASUS BE-6, then to a pair of Celeron 400's OC'd to 500MHZ on an ASUS P2B-DS. Since then, I haven't messed around with anything for lack of interest.

I'm currently in-process of assembling a C2duo E8500 system (I've had the extra CPU sitting on the shelf since 12-26-2008).

Current detail:
e8500 (pretty sure it's a SLAPK)
XFX 790-IUL9
2x 2x2GB Corsair PC1600C7 dual
Diamond 5770 (though I'll likely replace these with a pair of GTX460, if the e8500 would be powerful enough)
160GB 10k Raptor HD
650w PSU
DIY case made with .050" AL6061 sheet

I can machine all my own mounts / waterblocks / contacts, and likely use AL6061 1/4" tube for all the water-lines oppose to flexible hoses. I'd need to cover both GPU's and GPU RAM, the CPU and possibly CPU capacitors(?), HD, N/B and of course my RAM. I could see materials running about $125 and a small radiator about another $30, with a pair of fans putting at around $200. Machining and assembly time, probably 12-15 hours... Rough cost, $600-800.


SOo my question is -
Would I be crazy to attempt to build my own water-based coolant system? Or should I just stay with a generally-fair air-cooled COTS solution?
I don't even really need to O/C the system, but.... why not?
July 14, 2010 12:22:07 AM

for $600-800 you could just buy a VERY good watercooling system.

if i were you i would buy a cpu/gpu block and pump. you can make your own radiators, reservoirs, fittings, and tubing pretty easily.
m
0
l
Related resources
July 14, 2010 1:15:59 AM

needsumhelplol said:
Hi.

There's a pretty decent tutorial on building your own reservoir here;

http://www.bit-tech.net/modding/guides/2010/02/28/how-t...

DIY Waterblocks

http://www.madshrimps.be/?action=gethowto&number=2&howt...

Another good tutorial including homemade rad..

http://forums.bit-tech.net/showthread.php?t=105157



Awesome, thanks for the links!
I'd be able to go a lil further than the standard DIY stuff; I've got a CNC mill + lathe in the garage which can do 4+1/2 axis work, so pocketting and thread-milling to a few 1/10000" tolerance is realatively trivial ;)  (and to those of you whom know what working in tenths means, I wouldn't say 25-millionths ... because I'm not that crazy hehe)
I mean to ask, if going to the effort of water-cooling, or heck, even refrigerating a system would yield results to be proud about, beyond the sense of self-accomplishment.

Side thought - does anyone know (or could find out) how thick the nickel + copper casing is on top of a Core2duo? I could run a test of copper tube braized directly into an e1500 (no need for a water-block w/grease). Otherwise I guess I could just buy one for $5 and cut it in half :pt1cable: 
m
0
l
July 14, 2010 3:30:48 PM

ive only sanded the nickel layer off the copper but ive never had the balls to sand down to far.

once you start working on ur stuff post up some pictures dude!
m
0
l
July 16, 2010 12:50:48 PM

Air cooling is only for those who want to run at stock speeds, or those in cold countries. Really. Refrigeration works, but is only worth it if you're going for some insane overclocks, due to complexity.

The only drawbacks I've found with watercooling to date (after 5+ years) are the costs, and if you are making your own components you can get the cost down to roughly the same as a high-end air cooler, if not less. Tubing can be either the clear stuff from an aquarist shop (NOT recommended, it kinks very easily, and looks like crap after about three months of having glycol solution pumped through it) or automotive fuel-line (which works pretty well).

One thing, though, is if you are intending on using rigid aluminium tubing instead of flexible rubber hoses, you are going to have a major problem in changing out components - unless you're intending this to be an 'assemble and forget' system, in which case I'd strongly suggest building two or three watercooled rigs along more 'traditional' lines first, for practice and for demonstrating where some of the difficulties are. I've already done a liquid cooled system using rigid tubing, but since it was for an embedded control system in a hot environment with a lot of vibration, being able to easily change the CPU or graphics card wasn't an issue - in fact, there was no graphics card.

It's a cool idea, though, and I have been considering how to do it and yet retain the accessibility to the components. Banjo fittings, maybe, or those hex-nut compression fittings - though it will have to be drained and refilled every time something is removed.

The heatspreader - the metal case on top of the CPU - is about 1mm thick, I think, though I could be wrong. Removing it is not easy, but it is doable, with a pencil torch, flat screwdriver, patience, luck, and extreme balls. It was easier on the old P4 chips, since those used a different interface material inside, and didn't need the pencil torch - just some oil-based solvent like WD40 or Q20, or similar.

I can only think of one reason why I wouldn't try brazing a CPU, and that is the heat involved can theoretically damage the internals, though I have been proven wrong before... try it, let us know.

As for results, at the moment my LAN machine (non-overclocked, just a uATX case with a T-line loop in it) with an E5200 CPU runs at about 8ºC above ambient - AT FULL LOAD! The radiator is 150mmx260mmx50mm, and was a car heater core. Pump is some generic thing from an aquarist shop, 1.35M head, 600L per hour, 240V. Relays - what would I do without them?

Since you have access to what is basically a machine shop, you should have no trouble at all in making the blocks. Inside a waterblock, your primary design considerations should be surface area exposed to coolant and as little restriction as possible. If the flow through the block is turbulent as well, the heat transfer is just that much better. A bunch of zig-zag channels would get that effect.

Hope this helps...
m
0
l
a b K Overclocking
July 16, 2010 4:20:57 PM

1. Use Copper C110. Not 6060/6061. This is mainly due to significantly better thermal properties of Copper AND the fact that most/all good rads/heater cores are Copper. If you use Aluminium, you will get galvanic corrosion (unless using 10-50% anti freeze mix).

2. Use flexible hosing. Grab some Tygon or Primoflex LRT tubing online.

3. This should NOT cost you $800 (if you own the CNC that is). The material cost for Copper C110 bar*, heater core/rad, Tygon tubing, etc, should only set you back $200-300ish. That should enable you to make a block for CPU + GPU.

*Note: I said bar, not plate. Copper Plates are significantly more expensive than copper flat bars. 1/4" is good for the work.
m
0
l
July 16, 2010 7:35:53 PM

Hey gang - thanks for the input!
I should add a few things, and clarify on others.

0) Intent: I've put together another machine with spare / extra parts, and thought I'd play around with different possibilities. Short of placing an entire computer into a refrigerator, I plan to try to cover all heat-producing modules within a more permenent heat-reducing system. (Hence, rigid lines, ect)

1) Garage-shop: yes, I own my machines; eventually I hope to build a CNC rapid-protoyping machine and fast 2-D cutter (most likely plasma 'cause waterjets are messy); I also have access to an orbital welding machine, though I don't think it will be needed.

2) Cost: I listed ~$800 of my own efforst where time=money, and $600 at ~15hours of assembly is underestimated by about 4:1 in total time involved ;)  LOL

3) Tubing material: I believe aluminum lines would work best, because I would want to remove the heat from within the CPU container, and copper would radiate that heat back into the casing... Aluminum would reduce this a fair amount, allowing more heat distribution to the external condeser / radiator. I do have plenty of 1/4" Copper tube on-hand, but its' zinc lining might disqualify it for use...

4) Material source: I primarily use rem-stock from a few of my metal providers; nothing I'll be working on will require material certification, so the cost is considerably lower :) 

5) Material (more): Shadow703793, yep, I figure about $200-$250 range for materials; I've seen some 1/2" o2-free remstock at $3/lb, and a lil better for CP; gonna check today on availability of different stock (.020" sheet, 1/4" bar, 1/2" plate, and maybe some 1"x2" bar)

6) Disclaimer: I don't mean to imply I think this could be done quickly :(  I'm sure it may take a month or more, as I'm only in the early planning phase and want to gather as much inf as possible about what's already been done.

7) CPU casing: I figured it would probably be about .020" thick (~.5mm), but the heavier, the better.

8) Braze vs Solder vs Expoxy adhesive: Copper braze takes affect at around 1800F (I believe); far higher than the Silicon can survive. 5% silver solder would require a heat-soak just above 400F, which shouldn't destroy the chip, but 455F would. Though, IF the case could be removed, then it could very easily be brazed with tube (or, even replaced with a match-machined casing... possibile?) I've been told about a few types of adhesives, and just now found that Arctic Silver actually has a product which bonds at room-temperature.

9) Geometry: Squared honeycomb patterns are supposed to provide the greatest surface-area for thermal transfer - either air or water. Water-blocks made with .025" internal veins could be ideal (2-half assemblies).

10) Refrigeration vs Radiator-cooling: I only know basic refrigeration concepts - I'm open to thoughts and suggestions.
m
0
l
a b K Overclocking
July 16, 2010 8:28:44 PM

Quote:
9) Geometry: Squared honeycomb patterns are supposed to provide the greatest surface-area for thermal transfer - either air or water. Water-blocks made with .025" internal veins could be ideal (2-half assemblies).

For a water block, you want a square pin matrix pattern. If you want, I can send you some designs that I have tested via CFD but no real world testing. PM me if interested.

Imo, there really is NO NEED to make the tubing rigid. Just use regular Tygon or Primoflex LRT would cut down your work considerably.

There is also no need to remove the IHS of the CPU. Testing have shown you will only achieve about a 2C cooling gain in most cases. This is not worth the damage to the CPU. (There was a thread at Xtremesystems about this, ~2007ish.

m
0
l
a c 324 K Overclocking
July 19, 2010 7:26:46 PM

Making your own reservoirs is easy, and once you figure out what you need/want, you can start to get creative. Just make sure you get good acrylic and practice a few times. Google AC-Ryan; I have emailed him ( tribaloverkill[at]gmail.com ) about some of his designs, tips and ideas to help out. I made my reservoirs out of 2 1/2" acrylic tubing and a bunch of other stuff from your typical hardware store.
m
0
l
!