I’ve been around computers for some time and have recently delved into the murky depths of overclocking.
After a lot of experimenting and reading a lot of how-to pages on the net, the most helpful have been from Overclockers.com. I’ve built my own water cooling system.
During this time, I’ve noticed a lot of conjecture and “an engineer friend of mine told me” but not much empirical information.
I like to see the numbers not “expert opinions”. With my background as an engineer (SNIPE) in the Navy and now in the public sector, I’ve been able to apply my experience with heat-exchange systems to this problem, particularly to dealing with the 3 main factors involved; flow rates, surface area and thermal differential.
I realize that the best parts aren’t always available, and the general misconception that “if some is good more must be better” has even bit me in the ass a few times.
But just like in so many other areas, if you push one side of the equation without compensating for the others, you run into a bottleneck and start wasting a lot of effort.
Though heat exchange with water is similar to that of air, to get the most out of it, you need to change your approach a bit.
Since air is pretty inefficient in moving heat, the brute force “more is better” method yields improvement for quite a while. You don’t get diminishing returns until you come close to simulating a hurricane (air speeds of around 60 MPH+).
Water, though, is very efficient; it can carry a lot more energy and will work more efficiently “if” you give it a change. Because watercooling is much better than aircooling, though, you start getting diminishing returns much sooner than with air. Simulating Niagara Falls isn’t going to help you much at the low temperature differences we’re talking about for overclocking.
The oddest thing I’ve seen is the use of multipass tubing type radiators with fins. These look COOL, but are more appropriate if you’re working with high pressure or a high water/air temp difference.
That’s not what we use for overclocking. We’re only talking a few PSI and 10 or 20C at most.
The most efficient choice is more surface area at a slow flow rate.
That’s right, SLOW. You have to give the water enough time to dump all that heat and a lot of cool surface with which to do it.
While they take up more room, a cylinder has less surface area than a polygon shaped tube with the same cross-section. More surface area = better.
A good choice for that is a heater core from a car parts store or salvage yard. They even come in a lot of sizes from dinky compact to monster truck for you overkill types.
If you can find a self-serve wrecking yard, you can save a few bucks and get to rip that sucker out yourself. They may only be a 2-pass design but they have multiple flat channels with lots of surface area and fins to boot.
(I’ve been thinking about modifying mine to make it a 4 pass just for grins:))
Water block design is also important; the single swirl chamber or bore threw with plugs, while quick to make, are not very efficient.
Adding fins in the one chamber design is a step in the right direction but water, like most things, takes the path of least resistance, so some water exits the block without picking up much heat.
You need to remember that the more heat the water pulls out of the block, the bigger the difference between the temperature of the water and that of the air cooling it in the radiator, and the better it will work.
You can pump more water through the block to compensate for the less efficient design, but you won’t get much return for your effort.
If you are using a Peltier, the higher temperature at the block will mean you probably wont hit that ceiling as soon. However, if you don’t use a correspondingly larger radiator to get rid of that extra heat, you’ll just be pumping warm water back to the block.
Get the water in contact with as much surface as possible and get even water flow through the block. That’s what’s important.
My best results have come from drilling a series of holes about 11/16″ deep in a block of 3/4 inch thick C-110 copper in a serpentine pattern, then using a drumel to connect the holes to about 23ds depth.
This forces the water to travel evenly through the block and prevents laminar flow.
After soldiering a plate on top, you have a very efficient water block. My roommates thought I was rather strange when they found me using the electric stove and a butane torch to do this. It may not have that nice neat and even machine shop look but it has lots of surface area and that’s what counts.
I’m sure if I had access to a machine shop or even a drill press I’d be able to make an even better one.
By the way, the type of copper tubing you use can make a big difference in your results. Phosphorus added to copper tubing to increase tensile strength and help prevent crimping when bending. But phosphorus also reduces thermal and electrical conductivity. Just as little as 0.4% phosphorus can reduces conductivity by 50%. Even high grade copper tubing C-122 is only 99.5% pure with as much as 0.02% phosphorus. That’s why I use C-110 copper tubing. C-110 is 99.9% copper cold rolled for higher density. It’s often used in high voltage equipment and can be bought at metals supply warehouses.
So remember, “not all copper is equal”. Don’t even get me started on all the different aluminum alloys out their and their conductive qualities.:)
Back to the water block.
The above mentioned block along with a 6″x 8″x 2″ deep heater core and a 40 GpH pump connected with 1/4″ ID tubing keeps my K-6-III 450 fat and stable at 525 Hz 2.7v (.3v above spec) and only 16C above ambient under extreme load.
I use a program called K6burn to torture the CPU. It runs it at 100% for as long as you like and gets that sucker hotter than any other game or test I’ve found.
On a cool day I can even boot into Windows a 500Hz and 2.4v (spec.) though it will crash after about 20 min of Homeworld.
I’m presently experimenting with different flow rates, block design and air volume threw the radiator to see where I can shave off a few more C.
Finally, another problem that doesn’t get much attention is bi-metallic corrosion. I’ve heard suggestions about using distilled water since it is non-conductive.
I’m sorry to say that won’t work. The electrical conductivity of the electrolyte is not a big factor. The problem is free hydrogen ions (H+) and hydroxyl (OH-). So unless you can get hold of some de-ionized water, and have a way of keeping it that way, you’re out of luck.
The only real option is using an additive, like Redline Water Wetter, or if nothing like that is handy a dash of (non-toxic type) antifreeze will help that aluminum block from rotting on you.
Remember, aluminum has a -1.67v Standard Electrical Potential, and copper is rated at +.345v. Put them together with an electrolyte, and you got a darn good battery.