How-To build a GPU waterblock — Ryan Norton
And, for my next trick! Got bored, decided to build.
The main goal of this project is to build a functioning water cooling system for the GPU on my MSI Starforce 815 Geforce2 GTS vid card. The cooler should be small as possible so as not to obstruct too many PCI slots, it must use the original cooler mounting holes on the card, and it must last for a good long time before needing replacement.
As for performance – it needs to be at least as good as the original, and hopefully a lot better. I have a neat idea to try out as well, a big twist on a block I’ve seen before. Read on.
For my victim, I chose to use the stock cooler off the GTS as the heart of the operation.
The stock HSF was a Blue-Orb-style unit, consisting of a (very) small fan mounted in an aluminum extrusion base. The unit is supposed to take advantage of the nature of radial fans – airflow is greatest at the edges of the fan blades, coming off at a 45° angle relative to the hub. Directly under the hub is a "dead zone" of almost no airflow.
The Orb-style coolers attempt to capitalize on this effect, and fail miserably for the most part due to extremely bad build quality. My unit was no exception – it lasted for less than two weeks before starting to fail and was amazingly ‘buzzy’ and loud the whole time. It was also extremely lightweight, made almost no contact with the GPU, and limited my core overclock to only 220 MHz (from the stock speed of 200).
It was quickly replaced with the stock HSF from an Intel Celeron 566, which performed admirably well, but had the nasty downside of eating two PCI slots as well as being a major dust collector. Noise was, thankfully, not a concern – the stock HSF fan spins at a leisurely 4900 rpm and is too small to make much noise.
The dust, however, was – it piled up directly on the PCB containing my card’s TV-Out features. Just how conductive IS an inch of cigarette and pot smoke dust? I don’t know, but I’m not waiting around to find out. As if that wasn’t bad enough, the card’s weight with the Intel ‘sink must exceed maximum recommended weight for an AGP card by quite a lot – the card was wildly bent and would constantly be pulling itself partway out of it’s slot.
Add to that a handful of HSF mounting problems – the Intel ‘sink required a ghetto style mount, and may or may not be making decent contact with the core, I could never be really sure… Enough of this Mickey-Mouse stuff, on to water!
!SUPER BIG WARNING!
You MUST take care to clean all surfaces you intend to epoxy!! Oils from your hand are enough to keep the epoxy from bonding to metal. I used Goof-Off and had no problems. Once you have cleaned a part, do not touch it again! A few times partway thru my first block the epoxy just fell apart in my hands because I did not clean it. This sucks – do not let it happen to you.
Step One: Remove the fan assembly from the stock cooler via screwdriver and throw it far, far away. Be sure to laugh as it goes and perhaps use bad language. This leaves you with a bare heatsink with a large round flat are in it’s center where the fan used to go.
Lap the GPU contact area if needed (probably is – mine had a thick layer of black anodization that had to go). Be SURE to save the clips that held the cooler onto the card or this whole project is for nothing.
Step Two: Get supplies: A 3/4 inch brass pipe cap or the biggest that will fit, two 3/8 inch OD nylon 90° nipples, and some good epoxy if you don’t already have it – I’m down with JB Weld. Also, most important: A brass or preferably copper screw.
Step Three: Bust out the drill – you need two holes in the top of the pipe cap for the nipples and one hole in the very center of the heatsink for something special. The hole in the HS base needs to be as exactly close to the diameter of your copper screw as possible – this is important!
Be very careful to clean all edges on the pipe cap till they are more or less smooth. To the right, see a pic of the cap with holes drilled and nipples seated. Below: Drilled cap.
{mospagebreak}
Step Four:
Thread the copper screw through the base of the ‘sink till it looks like the picture at left. You’ll want to get it as flush as you can with the bottom (contact area) of the sink, but don’t – it’s imperative that it sticks out slightly from the base about a millimeter.
Be careful – the pipe cap is going over this screw and will be epoxied to the base of the HS; if the screw is too long, the cap will not fit and you’ll look like a dumbass.
Once you’ve got it in there, secure it with a dab of epoxy. DO NOT get epoxy on the threads of the screw if you can help it. Work a bit of epoxy around the screw/heatsink joint – just rub it in with your fingers but try not to overdo it – the idea is just to secure the screw.
You should have drilled the hole close enough to perfect so that there is little, if any, gap between the sides of the hole and the screw, so should not be much need for a bunch of epoxy in the hole itself. Too much epoxy will make it harder to re-lap the base when we are done.
Now is also a good time to rough up the surface around the screw – this will increase surface area and thus cooling ability by a little bit, and give the epoxy something to hold onto when you attach the pipe cap.
Step Five: Let the epoxy dry – I always give it at least 12 hours no matter what it says on the package. In the mean time, work on the nylon nipples till they fit well into the two 3/8" holes in the top of the pipe cap. I C-clamped my Dremel to a bench, and used a sanding drum tip as a grinder. Perfect!
Epoxy those bad boys in there – be careful not to obstruct water flow with too much glue. Again – 12 hour rule. Also, be careful to leave room for the hoses and pipe clamps – offset the nipples a few degrees from each other.
Tip: If you toss the nipples in the freezer for a while, they will contract – maybe by enough to fit snugly into the holes you hand drilled so poorly. If you use a drill press, ignore this tip ‘cuz you’re a lucky bastard and probably drilled them to fit perfectly in the first place.
Step Six: Final test! Assemble and make sure all the parts fit right. When you’re sure they do, goop epoxy onto the base of the pipe cap and attach it to the heatsink so it goes over the screw. C-clamp it and get ready to wait.
Don’t get epoxy all over the area around the screw – this area is your secondary heat transfer path. JB Weld isn’t as good a thermal conductor as bare aluminum – I think.
Assembled Block
Step Seven: Attach hoses and let slip the dogs of war. Hopefully you’ve got no leaks – pump water thru it for 24 hours in your sink just to be sure. Water + Operating PC = Bad bad things.
FYI, I’ve never built anything that ended up leaking – I don’t consider it luck, I think it’s just the fact that I took my time and did things as well as I could. No Shortcuts is the rule. Once you’re sure the thing is 100% solid and leakproof, go to
Step Eight: The Lap of Luxury. Break out some 220 grit sandpaper, a flat work surface and start slowly smoothing the base of your new waterblock. Gently sand away the epoxy you rubbed into the bottom, as well as the screw that you left not-quite-flush. When you are done, the bottom of the sink should look like this:
Just a pinkish-gold colored circle in the base of your heatsink. Continue lapping on up to 600 grit or so to remove the scratches the 220 put in. Go all out up to 2500 grit if you want, it’s your waterblock. I stopped at 600 wet (note: I took this pic before I finished lapping, which explains the ugliness of the base).
{mospagebreak}
Final Step: Install your new waterblock!
Since you kept the original mounting hardware (right?!?!???), it’s literally snap to attach and remove this block. Tap the hell out of it to remove any air bubbles – since the internal volume of this block is rather small, any trapped air will really work you over. Be sure to work thermal grease between the screw and the hole it’s seated in, if you need to. Mine had no gaps at all.
The overall plan with this block, if you haven’t already guessed it, is to use the copper screw as a direct path from the GPU to the cooling medium (water) with the pipe cap and original ‘sink as a secondary heatpath. This removes the base of the heatsink as a transfer path to the screw – I’ve seen a few similar waterblocks that just glue or weld the screw to the inside base of the block, thereby adding an unnecessary junction.
The threads of the screw serves to more than triple its surface area, as do the threads inside the pipe cap. If you want an even better block, try adding 3 copper screws set 120° apart from each other instead of just the one. Me, I could only find one copper screw the day I decided to build this. Hopefully this project worked well for you – let me know how it turned out for you!
Project Notes:
- Using all copper parts would let me weld or solder the thing together instead of relying on epoxy. Too bad stock coolers are never made of Cu.
- For a CPU cooler, use multiple strands of 18 gauge copper wire instead of a single screw. Flare the wire into a mesh to cover as much of the internal block volume as possible.
- Get a drill press, or that cool drill press thing for my Dremel.
- This was originally an air-cooled Orb sink – the HSF fins are still there and will serve to cool the GPU themselves if the pump fails. Also, perhaps provide a slight bit of additional passive cooling.
First Test: Since there’s no thermal monitoring on my vid card, I strapped this to my trusty old 300A. The 300A is one of the best chips ever, talk about tough – this thing has been fried, soaked, dropped, and everything else in between – and it continues to work as perfectly as the day I got it.
Plus it has a large die and a high heat output, perfect for testing – consider it a testing device, not a CPU. The 4 mounting holes built into the CPU itself were a godsend way before AMD came around. I still think the 300A is one of the most valuable assets an overclocker can have! Too bad mine only goes to 550 MHz, or I’d still be using it.
For the test, the CPU was run at 450 MHz 2.15 vcore. The coolant water and the ambient temp were both kept at room temperature – 70°F or 21.11°C. Full load was achieved by 30 minutes of Folding and Dnetc running simultaneously.
Idle Temp °C | 100% Load Temp °C | |
Stock Intel HSF | 33 | 48 |
Waterpole | 24 | 35 |
Waterpole, 2.20 vcore | 25 | 37 |
Well, lookee here! Not bad at all! Not great, either – the previous block I made performed MUCH better. The Waterpole can’t run the Celeron faster than 450 for more than a few minutes, while the JB-block kept it happy at 504 MHz all day long. The ‘pole did allow me to drop the voltage down to 2.1 instead of 2.15 and still be stable, a nice bonus.
Right now, many of you are thinking, " Big deal, it cools a 300A. But how would it do on a REAL CPU like an XP2000+?"
The answer is: Probably not too well. Then again, this is intended to be a video card waterblock! If it can handle the 36 watts of this 300A, you can be damn sure it will handle 10 watts from a vid card without even flinching.
How does it do on the video card? I have no idea about actual temps, but the back of the card isn’t hot to the touch at all anymore – it’s barely even warm. Rather than try to get a thermistor positioned just right so that it just might be getting a good reading, I rate how my creations go by overclocking – if the cooling is effective, it will overclock better.
So what happened? I can hit core speeds of 275 MHz before things get unstable. Not that a high core speed really makes a difference on a GF2 card that’s severely limited by memory bandwidth, but hey, bragging rights!! I have to admit – I did use ice water to get that high. With room temp water it topped at 255 MHz – still pretty good.
Stock Orb – Clone HSF | 220 MHz |
Intel S370 HSF | 242 MHz |
Waterpole – Ice Water | 275 MHz |
Waterpole – Room Temp Water | 255 MHz |
I’m calling this one a total success! It’s a bit quieter then the first two ‘sinks I used, it got me back a PCI slot, and it let me overclock higher than before. Plus it only cost eight bucks for parts and a few hours of work. Mission complete.
If the mounting holes line up, I’ll see what other cards I can try this on. I’d like to see how it tackles a Ti500!
Be the first to comment