Going to try watercooling. Inevitable questions to ask I'm afraid

JonJ678

JonJ678

JonJ678

Soldato
Joined
22 Dec 2008
Posts
10,369
Location
England
Hey. This is my first attempt, so I'm aiming to keep it simple. Only putting an 8800gt underwater for this loop, would like a reservoir since I think this will make my life a bit easier.

Intended plan is reservoir -> pump -> rad -> card though I doubt the order of rad and card matters. I'll probably be using 11/8mm tubing as the gpu block has compression fittings for this size.

The radiator I'm looking at is the quiet power swiftech 120.1, any guesses how well this will do at cooling the gpu? It'll have a 1600rpm scythe s-flex pushing air through it from the outside of the case but will be mounted internally [I assume case : fan : radiator but know nothing about shrouds]
I'm concerned about the 3/8" barbs fitted to the rad as I would have liked compression fittings, and I'm not convinced by fitting 11/8mm tubing onto 3/8" using clamps. Reckon I'll be alright here or would moving to 3/8" tube be a much better call?

Finally, I've seen this http://www.overclockers.co.uk/showproduct.php?prodid=WC-007-XS
which is a combined bay res and pump at 25 quid. This is very cheap, so presumably quite rubbish, but will it cope reliably and effectively with an 8800gt and a radiator?

I'd very much appreciate any input, a part of me is still screaming that water and electricity should be distant

Jon
 
Strange how quickly plans change. It now looks like I'm going to take a long time on water and attempt to do it properly. I take it soldering brass and copper together won't lead to terrible corrosion problems?

One thing is that the main few blocks out all seem to perform within a few degrees of each other. This suggests to me that a well designed one is capable of transferring however much heat the water can hold. I think this gives me some leeway, but as my favourite design so far appears to be identical to that of one of their north bridge blocks I fear it will be inadequate.

I guess that leads me to what are the important design parameters for a cpu block? Surface area in contact with water maximised, resistance to coolant flow minimised, base thickness minimised?

The current favourite design is to mill parallel slots over a square that corresponds to the size of the processor, and have an inlet and outlet at either end of this. Nice fast laminar flow, but I'm no longer convinced that's what I need from this block.
The overall design I had in mind was blocks on sb, nb, cpu, mosfets, all aimed at minimising resistance to flow. No reservoir or t line. Rational being that it should be possible to send water through this (+120.2 rad of some sort) very quickly to make things very cold.

My best guess is that a high flow rate, low resistance loop will make little noise but not cool things terribly well.

^and that really is the most coherent I can be at this time, g night :)
 
Is the argument against compression fittings purely financial?

The radiator I'm looking at is the 240 feser. Well reviewed, and their site actually offers a 3D model of it for download which I'm very impressed by.

'Decent' block is potentially a problem, as I'm planning on making the blocks myself. Not going to make a radiator though, too time consuming and unlikely to lead to good results.

As such, what makes a block decent? Nb I'm basically going to imitate an EK block. For mosfets, I intend a shape 12 x 12 x 48mm ish with a large hole drilled straight down the middle, tubing connected to the ends. Very low resistance, possibly quite crap cooling effect, but so easy to make its worth trying. Sb needs more thought and an idea of what power it dissipates. Cpu block I'm very unsure about.
Before manufacture I think I'll post models on here and see if anyone shoots them down. I may even get around to learning finite element analysis for this project.
 
I'm reasonably confident with metalwork, but this will be my first time with CNC. I have a fairly standard manufacturing lab available, contains lathes, milling machine, band saws etc. The type with an electronic readout for how many mm in each direction from the origin you are.
Adjacent to this is a rather big CNC milling machine that I'll have to beg for time on, and will find out the limits of on Monday. I'd love to believe it can reliably cut 0.2mm fins at 10mm deep, but I rather suspect it wont. I'll ask the lab technicians about its limits tomorrow.


I think the XSPX edge is exactly what I have been looking for! Thank you enormously for mentioning it. Water block manufacturers with clear tops are wonderful. I don't want to touch acrylic with a stick myself, but I'm grateful that other people use it.

Does anyone know of a good reason why every waterblock on the market has the inlet and outlet on the top, rather than on the sides? Greater compatibility with different boards, but introduces a lot of tubing loops. I'm more inclined to put the barbs on the edges of the blocks, and join the blocks together with shorter, nearly straight tubes.

Second, if anyone knows where I can find the chipset mounting hole locations for my board I'd be thrilled, as I've only got a ruler here. Not solved the problem of how to mount blocks yet, but would rather cut the holes from measurement than by using the current block as a template.

Cheers
 
Finding mounting holes online doesn't look possible, some point this week I'm going to have to draw around the existing heatsinks.

Couple of things. If anyone has a 10/8mm compression fitting they'd be so good as to measure it would be very helpful. Specifically I need the distance it extends out from the surface of the block, and diameter at widest point. I'm starting to understand why Ek don't make a block for my board, theres about 12mm of height to play with in total.

Second, any thoughts on where I can buy copper bar? Its definitely not going to be a competitor to ocuk so should be fine to post. My local engineering shop only sells aluminium, and is closed today so I can't ask them. Google is not helping much, though I think I found a place for 4 meter lengths of it. If I can't source it then this project suddenly became much more difficult, heat exchangers have no place in a computer

edit: looking into scrap with limited success and have contacted a family member who stands a good chance of having accumulated some over the years
 
Last edited:
Good effort on the picture, thats exactly what I was hoping for.

Copper sourcing may be a bit hit and miss. My local stocks 1"x1" bar and various plate. I'll hear back from them today as to whether they can source 2"x1" for me, which I think is probably the size I want to work with. No word from the family member, but he doesn't check email often so I'm not too surprised.

Had a word with the manufacturing guys. I think I'll struggle to get time on the cnc for something so clearly unrelated to academics, but the mills available are pretty solid. We'll see, perhaps doing well by hand on the simpler ones will persuade them.
There are unfortunately fairly inflexible limits on what one can do with a slot cutter. Depth of about 3/2 times diameter being the big one, where I can go deeper but stand a good chance of breaking the bit. This probably limits channels cut to 4mm wide by 8mm deep, so I think I sympathise with ceasing manufacture of the Edge block.

Lovely ductile copper is of course a bitch to machine. Cpu block is going to take a hell of a lot of designing, I think I'll put the northbridge/mosfets (and southbridge if I somehow work out the height issue) under first in the spirit of training myself. I'd better become quite good at this, because I think I'm going to have to order half a meter of copper at a time so need to make many many blocks for years to come. Might even do some mathematics while designing.

Cheers guys
 
This might qualify as a bump, but it's a pretty large update. Small horror in that my psu has died, but aside from that today went reasonably well.
Here we have a list of available materials

It's not financially viable to buy 2" by 1" bar, as the shortest I can get is 500mm at £100 + whatever I'm charged for them buying it in specially. So I have been forced to look at other options.

2" by 1/4" at £49 a meter features heavily in the current design. Think four layers stacked, each with a shape cut out. This neatly solves the problem of milling machines not being able to cut very narrow and deep at the same time, while introducing a nightmare for leakproofing.

Acrylic comes in at £38 per 1/4 square meter. So thats 125cm by 5 cm at 0.5cm thick. Compared with copper at £49 for 100cm by 5cm at 0.6 thick, its got to be the copper really. The problem of how to stick acrylic to copper is a very big one and pretty much unsolvable without introducing rubber gaskets.

Next up, we have copper pipe. 15mm diameter which is rather larger than I'd hoped, and 3m long which is ridiculous for a computer, but at the price of £7. This is the narrowest available, and it was sold out today so I don't know the bore. I'm assuming 11 or 12 mm.

Then, a choice of two readily available solders. £15 for 250g of silver solder, £10 for 250g or normal solder. Both lead free of course. Solder is likely to feature heavily in this build. Silver solder melts at 220 ish, the other at 230ish. This is bad news for me as I had hopes of using an oven to melt the solder. Damn.

Finally. a G1/4" tap is bloody difficult to find locally in England, but I think I've found one close enough in 11.5mm, as opposed to 11.445mm.

So, in summary I still don't want to use acrylic. I'm currently inclined towards 4 layer blocks for the chipsets, drilling 15mm holes in the faces of these and soldering painstakingly bent copper pipe between them. The end result hopefully being a single fused mass that covers nb/sb/mosfets and has only one inlet and outlet, tapped G1/4". Whether the CPU will be in this or not is a difficult question, as it would make changing CPU difficult but the watercooling tidier. Oh, and solder absolutely everywhere.

So, much designing to do. I can now plagarise multilayered blocks of my choice however, which may make life easier.

Cheers
 
The bay might do better for acrylic, for copper the best I could manage was 4mm thick pieces of bullion for 8 quid a throw. 49 a meter of 1/4" copper is reasonable I think, especially since they'll give me whatever fraction of a meter I require.

Copper work hardens doesn't it. That makes things much more exciting. 0.7mm thick sounds good, it is indeed plumbing stuff. Ill hazard a guess that I can anneal it then bend it without specialised equipment, so turn a blowtorch on it, bend it while hot, reheat, throw it into something cold. Probably water. I'll look into that with some care, as I'm a materials science guy at heart. 85 a coil is too much for me when a blowtorch should do the same thing.
The engineers offered 10mm with 3 mm bore, which would bend beautifully but otherwise be little help. Its sold by mass, so no benefit to buying that and beating it out into 6mm plate.

The silver solder in question is only 4%, as its for electronics use. Wiki puts the composition of each at the 220 +/- 10 or so, which might be reachable in my oven. My oven is not very high quality however, blowtorch might be a better shout. As for mechanics of doing so, I plan on milling a 1/2mm deep grove around the block, putting a line of solder in this, then bolting it all together and cooking it up. Ill check dimensions, density change between solid and liquid solder etc. It may work out better to line the groove with solder, melt it to remove air, sand flat with 1200 then clamp together.

Tube horizontally between blocks with G1/4 fittings would be better. I think waterproofing it will be a nightmare though unless I solder the barbs in place. Even with this I can see the seal not being great, issue is that I'm using a stack of plates fixed together. So I can't solder then drill holes for tube, or the solder will melt. So Id have to clamp, drill down, separate then try to solder. I can just see it going wrong really easily. You're absolutely right in that it'll limit it to this board. Its an asus P5Q premium, so it would be difficult to do significantly better with P45. 48X is meaningless to me as I don't want multiple graphics cards. I look at it as when I move to i7, and repeat this effort, I'll do a better job of it. This is largely a learning exercise after all, otherwise I'd just buy the blocks.

Off to find a machining engineer. Cheers all
 
You must log in or register to reply here.
Back
Top Bottom