Direct-Die water cooling

[Gizmo]

So, I'm a noob when it comes to OC and WC, but I'm bouncing around the net one day and I come across Spode's, where he's talking about the experiments he and SurlyJoe did with direct-die a while back. I'm looking at the numbers and I'm thinking, 'this sounds like a Good Thing'. Being of questionable sanity and a geek to the extreme I figure I'll jump right in.

So, I build my own direct-die water block. The initial results were, to say the least, disappointing. I had better luck with my stock HSF. After much work and refinement, I finally come up with a water block that will let me OC my 3200+ Barton to about 2.5 Ghz (202 FSB and 12.5 multiplier) with the VCore at 2.0v. This is pushing my system to the ragged edge of stability. Here're my observations and questions:

Mobo temp when running CPU burn hovers around 3C above ambient (measured this with my RadioShack thermal meter; dunno how accurate it is, but seems pretty close). CPU runs about 30C above water temp (CPU temp measured by socket thermister and read by MBM, water temp measured by CrapShack meter). I think I can get more performance out of the chip, but I can't keep it cool enough, and experience shows that when MBM gets a reading of 60C or better on the CPU, the system is getting ready to lock up.

Doing the math on power consumption vs clock frequency vs temp rise, I'm supposedly getting about .25C/W out of this setup. Does this seem reasonable? The numbers I've seen on other sites would seem to indicate that a good water block can get into the .15C/W - .2C/W, without the associated headaches of direct-die. However, we all know about benchmarks and how you have to be carefull in comparing the measurements and drawing conclusions from them. Is it reasonable to have to push the 3200+ to 2.0V to get 2.5 Ghz out of it? 120W is a LOT of power, and does the temp rise I'm measuring seem reasonable?

I'm also measuring a drop of about 1C across my radiator (Ford Escort Heater Core). This doesn't seem like enough. Is it?

[Gizmo]

<BUMP>

Any thoughts on this? Anyone else out there experimenting with direct-die water cooling?

[purplemohawkman]

my insanity hasnt kicked in yet. but what kind of pump are you using?? im thinking the more flow the better. and im supprised your only getting 2.5ghz out of a xp3200 my 3000 does 2.62 @1.95 i havent really seen any stability inprovement @ a 2.0vcore or higher just a bunch of extra heat. and maybe its just my chip but i can ususally get better OCs with lower multi's untill i reach my ram's limitations

[Gizmo]

Thanks for the response.

I'm using a Danner MagDrive Pondmaster 700. 700 GPH at 1' head and a max head of 12'. My water faucet doesn't run this fast. :-)

My experience with increased VCore is similar, I don't get any benefit in performance, but I manage to generate a lot more heat.

I'm quite certain that I could get into the 2.7 range with this chip if I had better cooling on it, even without going to refrigeration. My problem is that I've got a 30C core temp rise over water temp at full load, and that is limiting what I can do. I'm thinking about buying a Danger Den RBX and seeing how that performs, unless I can sus out what I'm doing wrong with my current set up.

[surlyjoe]

Hi Gizmo, I would love to see your setup! I doubt your doing anything wrong with the D2D setup, in the end I was resorting to high pressure to keep up with the new chips, and the water blocks were getting good enough that there wasnt any real benefit in D2D anymore. and now with the covered die packaging...

I think your on the right track with the RBX...

[Gizmo]

Bit more info:

I've got a 2" cube I built out of plexiglass. In the top of the cube, I put 1 hole for mounting a 1/2" hose barb dead center. In the side of the cube on one face I put 1 hole for mounting another 1/2" hose barb dead center. In the bottom of the cube I cut a hole approx 1/2" x 3/4" (just big enough for my XP core to fit in).

I got a piece of 1/2" copper tubing about 1 1/2" long. I also got a thermocouple for a central air unit ($2 at Lowe's), and cut it into 3/4" sections, so that I had a bunch of tubing roughly 1/16" ID and about 3/16" OD. I took 8 of these and arranged them into a 2x4 grid. I took the 1/2" copper tubing, smushed one end until it was just large enough for the thermocouple tubing to fit in, and then brazed the whole mess together into an assembly just a little over 2" long. I then took the tubing assembly and attached it to a 1/2" plastic hose barb (basically, I heated the tubing and melted it into the end of the hose barb). I then dry fit the assembly into the cube, aligned the end of the thermocouple tubing so that they came out where the core will sit. After that, I took it back apart, cut the tubing flush with the inside edge of the bottom of the cube, mounted the assembly, and attached the side hose barb. The result is an assembly the squirts water directly onto the cpu core from a distance of about 1/8" (about 3 mm) using 8 1/16" (about 1 1/2 mm) jets, and then the water is evacuated out the side hose barb.

I've tried a number of different approaches to this problem, and this has worked the best of all, but I'm still not getting the kind of results that I had hoped for. A good standard water block will out-perform this setup, but I've seen articles where people have gotten MUCH better performance than this from similar CPUs.

[Gizmo]

Originally Posted by surlyjoe

Hi Gizmo, I would love to see your setup! I doubt your doing anything wrong with the D2D setup, in the end I was resorting to high pressure to keep up with the new chips, and the water blocks were getting good enough that there wasnt any real benefit in D2D anymore. and now with the covered die packaging...

I think your on the right track with the RBX...

I'll try to post some pics later, right now I'm packing to get ready for a trip to Atlanta. I may see if I can pick up an RBX while I'm there. I've got some pics right now, but they are all blury and out of focus.

I was wondering if the fundamental problem didn't boil down to surface area and watts. Water doesn't conduct heat all that well and, oddly enough, stuff that conducts heat better is either highly corrosive to our beloved electronics or doesn't pump very well.

Strangely enough, going to an RBX (or other similar waterblock) will probably make my ultimate goal simpler. Once I get the WC aspects ironed out, I have the parts to build a phase change water chiller, which is where I intend to end up with this. I just wanted to maximize the performance of my existing system before going to chillin'.

BTW, overall, I'm not disappointed with this project (which is all YOUR fault, BTW, because of that article at Spode's ). I've learned a lot and had a BLAST!

[Daniel ~]

You made front page gizmo! {Now how about those pic's!":O}

http://www.aoaforums.com/

[Gizmo]

Originally Posted by Daniel ~

You made front page gizmo! {Now how about those pic's!":O}

http://www.aoaforums.com/

Oh wow! Front page! Must be my 15 minutes of fame! LOL

Pics to follow.

[Gizmo]

First, the system itself:

All this gear was purchased back in May. Some prices have changed since then.
Athlon XP 3200+ (barton core) (EBay - $350)
1 stick Corsair Value Ram - 256 MB ($52 - Googlegear)
PNY Verto GeForce MX 440 w/64 MB RAM (Best Buy, $79)
AMI Megaraid Enterprise 1500 4-channel Ultra-2 SCSI controller w/64 MB cache (EBay - $80)
10 Seagate 118202LC 18.2 GB Ultra-2 SCSI drives (EBay - $180)
Yeong Yang YY0221 case (EBay - $130)
520W PS - brand unknown, purchased with case $80
Abit NF7-S Mobo - purchased with case - $120
HP CD-Writer 9100 Plus (Used from friend - $25)
8x Sony DVD (local store - $65)
Keyboard, mouse, floppy (local store - $30)

Ford Escort Heater Core (AutoZone - $19)
Danner Magdrive Pondmaster 700 pump (PetWeb - $65)
2 120 mm fans for radiator, 3 80 mm fans for case, laying around.

I've had the CPU up to about 2.7 Ghz, but I can't keep it cool for more than about 10 minutes at that speed. I've had the ram up to 224 Mhz FSB. It stops there.

[Gizmo]

Now some pics.

First is the system from the front. Then we have the fans, rad, and water pump. Third pic is the rad, pump and fans from the side, showing off the ghetto cardboard fan duct. Took me all of 2 minutes to build that, and it shows.

Mobo with waterblock. Left side of Yeong Yang case.

Right side of YY case with 8 hard drives, the 2 cd/dvd drives, floppy and PS. Took me three tries to get everything mounted in there so the cabling would work.

[Gizmo]

Back, showing 2 fans, plus PS (the PS has 4 fans in it), along with mobo stuff. You can see part of rad and water pump at top.

Now to the meat. This is the pump I started with, a Little Giant PSM 006. 54" head and 120 GPH at 0 head. Not much of a pump, but it was only $19 at Lowe's.

My various attempts at direct-die water blocks. The one on the left was my first attempt. A piece of 1/2" tubing came down the center through the hole in the top and sat right above the die. The water went back up around the tubing through the 3/4" PVC, and exited the side. It had worse performance than my stock Antec HSF, which was pretty crappy itself (just barely adequate for my CPU). The one on the right was my second attempt, which refined the first attempt by 'necking' the 1/2" pipe down to a 3/8" adapter. Somewhat better performance, but still pretty lame. I was able to achieve my first CPU overclock with it, though (2.3 Ghz, 100 Mhz OC, woohoo). In the middle is my third design, which has itself been through about 3 revisions. You can just see the tubing coming down the center. I got to 2.4 Ghz with it, using the Little Giant pump. Thinking the pump was holding me back, I bought the Danner, and got 2.5 Ghz.

Bottom view. Sorry for the poor focus. You can just make out the 2x4 grid of 3/16" tubing. I expermented with a 3x3 array, then went to the 2x4 array, then tried pinching the ends of the tubing down so that I got more of a 'sheet' of water from each jet. Going from 3x3 to 2x4 bought me some extra cooling, because the pattern more evenly covered the die. Using a 'sheet' of water instead of the round jet that I'm using now didn't seem to make any difference. The black smudge is the gasket that seals the water around the core. I used closed cell foam insulating tape (you can get a 50' roll of the stuff in the heating and cooling section at Lowe's for like $4). Pasted it on the bottom, cut out the opening, placed it on the cpu to get an imprint of the parts on top, and then cut out just enough to seal everything.

Here you can see me mounting the block. Those 4 bolts hold the block down with two pieces of 1/4" iron. The bolts are insulated from the mobo by rubber grommets on each side, which make dandy flat washers when you cut them in half. Why did I use the grommets instead of rubber flat washers? 'Cause I didn't have any rubber flat washers handy, silly! You can also just make out the core in the center of the bottom face. I had to bevel the top edge of the bottom face as seen here in order to clear the edge of the CPU socket. Oh yeah, those bolts make opening and closing the CPU socket lever a real PITA.

Welp, that's it for now! I'm off to Atlanta for a week.

[surlyjoe]

nice job dood! funny, that looks like the same server cube I'm runnin. where'd ya get it? I had a hell of a time figuring out who made them.