I had to insulate some components on the radeon's GPU (used my trusty epoxy) I see no problems with doing the same thing with an AMD XP.

In the new year I'll grab a throughbred and direct die cool it, the small hot core will be ultimate test of my jet design.

If you were to do an XP, I would suggest cover every spec of substrate with something, JFettig over at ocforums lost his 1600+ to an unknown cause, but thinks the substrate absorbed water, maybe it was the core

epoxy and water dont mix.
epoxy is sooner or later going to be dissolved by the water if no insulation is used. I think a bit of silicon might be in order. or just use loctite glue.

My XP is running pretty well now with direct die cooling, and has not shown any problems in the month or so I've been running it.

It does not have very much of the substrate exposed to water, though.

I'll give temp readings as soon as I get around to soldering my AXP internal probe reader soldered back on.


Edit:

Volenti: what do you use to bond the pieces of acrylic together? I've been using superglue (cyanoacrylate), but I suspect there's something better... :D

Oh, also, how do you bond the acrylic and copper?

Epoxy was recomended to me for Cu>poly~carb bonding. Any chance of some pics of your XP setup?, is it a palimino or a t~bred A or B?...

What about combining this project with your Pumpblock Project? I think high flowrates above the die would bring a great performance improvement.
But it might be quite hard to fix such an constructions on the little P4...

I wonder if this would work better than a good Watercooler when usig a T-Bred. the P4 die looks much bigger than an T-Bred. When both use the same energy you have to carry away much more heat/mm². Would also be interestig if the direct cooling is still better when using really extreme CPU Voltages. Might be possible that an ordinary Watercooler is better in that case.

The pump block can't come close to competing with the water volecities achieved with the direct die style jets, though it may have some limited applications in "koolance" style all-in-one solutions,(using a small 12v pump say) since it can still give good performance with very low system flow rate.

With the understanding of jet design that's been developed here I recon we could come up with a jet setup that would give the best blocks a good scare, but I'll not going to wreck another cpu prooving it;)

It's a Tbred 1700+. I forgot to look up the serial, but it seems to get flaky around 1800MHz. I don't remember off the top of my head, but I think I'm running it at 1720 right now.

I would have pictures, but I don't have a digital camera, and my friend's just got stolen from his car.


Epoxy makes sense. It sounds like a good material to use for the layers of plastic as well, but I dislike the long cure times, and I wonder how to keep bubbles out--they're especially noticeable when bonding two pieces of clear plastic. :D

Volenti: how do you keep bubbles from forming when you bond parts, and what have you been using to polish your blocks? Mine always melt when I try the buffing wheel on my dremel, and I don't quite have the patience to do it by hand. :p

I didn't really have any trouble with bubbles in the epoxy, but I usually mix the epoxy slowly to avoid "whipping" air into the mix.

I use a bench grinder with stitched and loose leaf calico buff wheels, as well as the correct buffing compound for each. It's the buffing compound that does most of the work.

Y not run 2 or more pumps parrallel? That would provide alot more pressure. Put a Y connect b4 the block.

Parallel pumps give higher flowrate, not higher pressure, you need series pumps for higher pressure.

One question... why didnt u use an o-ring for sealing? It worked great for me while playing with direct-die.

Edit: Thugh the waterblock sucked, i got temps similar to my previous WB but i could overclock 150MHz extra with the old WB :D I werent brave enough to keep experimenting with nozzels and so on, it was quite creepy to use the block, when i turned to pump off the water inside the WB started to boil within 2-3 seconds.
http://w1.858.telia.com/~u85821090/direct-die2.jpg
http://w1.858.telia.com/~u85821090/direkt-die1.jpg

I like the idea of slotting the holes on the retention plate--I always have problems lining up the bolts that go through the MB.

Those are 3/8 barbs, right?

How deep is the 'pocket' in that water jacket? I didn't have problems with the boiling so quickly (or at all) when I turned my pump off for a few moments (my curiosity will kill me one day... :D )

The only time I ever saw the slightest indication of boiling was when I was first experimenting with my K6-2 350@500/3.5V, the top 1.5" of a pop bottle, and 12ml of static water.

Slotting the holes was necessary in order to not to have the o-ring rest on electrical components on top of the cpu.

The babs were handmade and are for 3/8 barbs yes.

The pocket is about 7-8mm.

Well i ran a tbird 1400 @ around 1500MHz with this waterblock, vcore around 1,85-2,1V so boiling did occure rather fast ;)

I found the info on my Tbred:
Batch RIWGA (Tbred A), week 35 of 2002

AXDA1700DLT3C
RIWGA0235MPMW
9345097270086

At the time I was playing with direct die I didn't have the ability to mill an o-ring channel, besides the blu-tack gasket worked well enough for me, I never had it leak.

I like the idea of a wider more redundant seal, O'rings work great but they have very fine precise tolerances, a bit of dirt and they're buggered. I'd prefer something with more of a 'feel safer' factor than BlueTack but it appears to do the job :D ...

There's not many other re-useable gasket like materials that you can use, not only does blu-tack fullfill the requirements of a low pressure gasket (under 10psi) it's also adhesive, seals very well with low mounting pressure, and easily removeable for adjustments, changes ect. I looked, and couldn't find a better alternative (other than something permanent like silicone)

I like your jet/spray idea for the WB. Perhaps it could be used in other designs to evenly distribute, channel, and remove water from a WB. I'm sure your approach with the jet/spray will become popular. :drool: It could also be used to direct water to an exact location.

How can you take a direct die block off your cpu without killing everything? I think im just not thinking but unless its epoxied it would be diffucult to take the block off without getting your mobo wet.

in my setup you could remove the cpu from the socket with the block still attached (and sealed), you can then remove the block from the cpu over a towel or something, the block only holds a couple of cc of water.

All looks very interesting indeed :) kudos for trying it Mr Volenti.

Couple of questions :

1. Does the removal of all the thermal barriers ( copper, AS3 etc ) between the water and the core, make up for the loss of surface area possible with a milled copper block design?

2. How safe is this design? I looked into Direct Die a while ago, and spoke at length to Spode about it, as he had had some success trying it. He found that the main problem was time, a DirectDie block that works for a week may not work several months later.

How do you intend to get over the problem that over time, a chip may absorb water?

3. How does the surface of the P4 differ from the surface of a AMD chip?

The original grey Athlons seemed to be reasonably waterproof ( Spode reckoned ) but the newer red, green and brown ones, seemed to fail faster, again, given the timescale mentioned above.

4. Do you reckon this is the next step for watercooling?

As surface area to power output ratio decreases, one has to find new ways to spread the heat. That is why direct die cooling can't be as effective as (for example) a WW block.

It is interesting to experiment with though. Keep it up Volenti. :)

1. I believe so yes, given enough water volecity and the right jet design. I have made a few water blocks now that have an active heat removal area that isn't much bigger than the core anyway.

2. it's not safe long term untill the manufacturing of the cpu's is designed around direct die as a cooling option, ie using water proof materials.

3. the p4 die and paly amd are similar in area, the T-bred is signifigantly smaller in area.

4. several companies are investigating a similar approach using controlled jet's of nonconductive fluid to directly cool hot components, but that's more phase change than simple convective water cooling.

Do you think that if you where to angle the jet of water say at 45 deg, it would make a diffrence?

just thinking that you could direct flow better and make the water chambers a bit smaller so no hot water gets traped anywhere?

Hmmm, just a tought...

-P-