Direct-die WCing! Cheap and easy.

I know it sucks to put time and effort into something to not get back expected results out of it. But you really do have balls of steel. Do they clang when you run? :ROTF: Good job and thanks for the test.

I third the above... not only did you have the balls to do a cool as hell experiment, but you also had the kohonies to post honest results afterwards despite being negative results :up::up::up:

I would definitely also like to see posted results of direct die with your fuzion if you do that next. At least there you will have a "heat spreader" via copper bottom of the waterblock, plus milled channels with more surface area inside waterblock. Again be interesting to see trade off of solder heat transfer to IHS before paste/waterblock versus directly to paste/waterblock.

i just saw the results...man that was awesome! well the corei7 sure is hot thats for sure....but damn you stripped a xeon proc's ihs ...thats fearless :banana::banana::banana::banana:t :D

Pitty :(

I wonder as a thought, to salvage it, how about artic ceramique gluing some copper BGA sinks to the IHS.

Awesome project, IMO, even if it didn't turn out as well as you hoped.

:up: for this experiment man! :yepp:

Reading the quote of Cathars post I can't get the thought of 4 RD30 hammering water on that die in parallel with an 4 (3/4") to 1 (1/2") manifold. :slobber:

Great experiment. Kudos for having the balls to do this... I know mine aren't strong enough.

Would it have been possible to have mounted something to the die to increase surface area and, potentially, spread the heat?

I know what you're thinking, no, not like an IHS. Something that the water could flow on, under, and all around to improve heat transfer beyond using the IHS.

Quote:

---

Originally Posted by p2501

:up: for this experiment man! :yepp:

Reading the quote of Cathars post I can't get the thought of 4 RD30 hammering water on that die in parallel with an 4 (3/4") to 1 (1/2") manifold. :slobber:

---

Someone needs to do that!! :bravo:

All i can say is "That is wack"...

Hey sure the temps weren't good, but it's still a success since the chip still worked and your not down $270 on a 920!:D Not all experiments are a success, hence why it's called and "experiment" and not a "sure thing"!;)

Quote:

---

Originally Posted by BringerOdeath

hence why it's called and "experiment" and not a "sure thing"!;)

---

Doh....Have to edit my bet's as an experiment. :rofl:

Thanks guys! Attaching something to the die like a RAM sink has it's merits. I'm going to think on this experiment a while and may get back to it one day.

I tried mounting my Fuzion to the bare die but the 1366 socket is so big that the waterblock doesn't supply enough down force to make the CPU get good contact with the socket pins. I don't have a backplate on my board and my top plate is kinda flimsy so I didn't want to force it. I needed my rig back up and running so I just swapped my old chip back in for now.

Wallwind, please note that the otherguy who did this same had waterchller in use which cooled water to 20c :)

I honestly think this would work VERY well if the channel was restricted so that high velocity water would jet across the CPU surface.

Thanks for trying something new!

Quote:

---

Originally Posted by fallwind

Thanks guys! Attaching something to the die like a RAM sink has it's merits. I'm going to think on this experiment a while and may get back to it one day.

I tried mounting my Fuzion to the bare die but the 1366 socket is so big that the waterblock doesn't supply enough down force to make the CPU get good contact with the socket pins. I don't have a backplate on my board and my top plate is kinda flimsy so I didn't want to force it. I needed my rig back up and running so I just swapped my old chip back in for now.

---

I think that is the problem. It seems (and this is totally uneducated guessing on my part) is that the problem is a combination of flow and surface area. Basically you need more water hitting more area to dissipate more heat.

fallwind it was a good try and with a more powerful pump(s) it would have had better results.


btw, I like my crow with a little garlic.

Quote:

---

Originally Posted by ecat

Disappoint ?

In spite of the controversy you stuck to your guns, went through with the test and posted the results. Excellent work :) Be sure to update this thread with your Fuzion on die results :yepp:




No cores were harmed in the making of this thread.

---

QFT! Fallwind, you and we all leanred from this and for that you have my thanks.
Like Edison said, he didn't learn one way to make a light bulb, he learned 2000 ways not to make one.;)

I'm certainly not disappointed. Mad props for seeing it through.

I was hoping you got better temps though. Seems the closed block will work better for you after all. :shrug:

This is the sort of Xtreme LC experiment I come here to read about. Anyone can bolt up some store bought parts, it takes courage and imagination to attempt something extraordinary.

If you're not ready to give up just yet, jamb a tube that sits just short of the die into the inflow barb and retest it.

BTW, what is that oh so shiny pump you're using?

Maybe it's because there's a dead-spot of low flow right over the surface of the die, allowing for excess heat to build up in liquid that's taking a long period of time to exit the block.

http://img291.imageshack.us/img291/9585/deadspot.png

Quote:

---

Originally Posted by YukonTrooper

Maybe it's because there's a dead-spot of low flow right over the surface of the die, allowing for excess heat to build up in liquid that's taking a long period of time to exit the block.
[/IMG]

---

Quote:

---

Originally Posted by eligray

I honestly think this would work VERY well if the channel was restricted so that high velocity water would jet across the CPU surface.

---

:up:

If you can get velocity across block a lot higher (which you definitely can by using a narrow jetlike channel), you will see big improvements :D

Awesome experiment and definitely BALLS OF STEEL. Keep the faith and keep up the great work.

Quote:

---

Originally Posted by YukonTrooper

Maybe it's because there's a dead-spot of low flow right over the surface of the die, allowing for excess heat to build up in liquid that's taking a long period of time to exit the block.

http://img291.imageshack.us/img291/9585/deadspot.png

---

He said there was a turned-around barb in there jetting water onto the die about 1/8-1/16" away from it. Back on page 2 or 3 I think

Quote:

---

Originally Posted by paul_

He said there was a turned-around barb in there jetting water onto the die about 1/8-1/16" away from it. Back on page 2 or 3 I think

---

:rotf: :doh:

Actually, it was 1/4" away.

http://www.xtremesystems.org/forums/...7&postcount=56

Quote:

---

Originally Posted by YukonTrooper

http://img291.imageshack.us/img291/9585/deadspot.png

---

Yep, add a hose barb running from the inlet to the die on that diagram and that's exactly how it was. The core was getting hit hard by the water, but there's just not enough surface area for the water to absorb the heat.

I think soldering a small ramsink to the core would do wonders. Something with alot of surface area, like this:

http://www.ixbt.com/short/2k3-08/ramsink_ocz.jpg

can't get picture to load

okay, that was wierd, I copied the picture link into the address bar, it found it, then after that the picture loaded. must be to prevent deep linking

Quote:

---

Originally Posted by fallwind

Yep, add a hose barb running from the inlet to the die on that diagram and that's exactly how it was. The core was getting hit hard by the water, but there's just not enough surface area for the water to absorb the heat.

I think soldering a small ramsink to the core would do wonders. Something with alot of surface area, like this:

http://www.ixbt.com/short/2k3-08/ramsink_ocz.jpg

---

Yep, that would work better and be interesting to see. Or use solder tim to just attach fuzion waterblock to die, that way you get best of all worlds (except remounts). A waterblock that wont leak and already has good jet and surface area on inside, copper bottom of waterblock makes great IHS/heat spreader, and instead of some that have done the waterblock to die with 2-4 W/M*k paste which to me is a crap shoot versus having an IHS (mine and others gave slightly worse temps others have reported better), no doubt using solder if could get ~80 W/m*k, without IHS and without any crappy paste tim, that would give best temps...though remounts may take a little more time and effort.

RAM sink FTW

How did that german guy get 30c load @4.2ghz (i7) ?

Even chilled, that's much lower than this guy's, the only difference was the nozzle instead of using the IHS to flow water across the core.

And this my friends is what, why and how XS is awesome.:up:

you hit the nail right on the head phelan1777 :up: :yepp: :clap:

@ fallwind Great job :up: now tweak it a little and you mabey on to something here.

Quote:

---

Originally Posted by Bradan

How did that german guy get 30c load @4.2ghz (i7) ?

---

German engineering in da house?

I don't think you should bond a ram sink to the cpu. That will no longer be direct die, just a ghetto water block. I suspect your fusion is better than a ram sink.

Nice work, fallwind. I didn't expect, that someone would really build another direct-Die-cooler.

@YukonTrooper
:yepp:Youre right! I'm working on the 2nd revision now.

@rge
Yes, the surface area is vitally important, when you use a material with "bad" heat conductivity. Some facts: The Thermal conductivity of Arctic Silver 5=7,5W/mK, Copper(99,9%)=390W/mK. Before you die trying: CNT=6000W/mK (available via Bayer AG, Laufenburg, Germany). There was a trick to get better thermal conductivity: an admixture of ... ;)

For everybody, who knows what to do now: You haven`t seen this post. This solution is currently being patented/not disclosed until jet.

No real supprise at the results.

I remember trying this with socket A processors, but the earlyest instance I can recall is this - http://www.spodesabode.com/archive/c...cle/directdie1 on a pentium 3.
The results were still poor :)

The problem with direct die is that you need insane flowrates and pressures to get anything like decent results. As soon as you ramp up the pressure you will spring a leak!

Quote:

---

Originally Posted by Holst

No real supprise at the results.

I remember trying this with socket A processors, but the earlyest instance I can recall is this - http://www.spodesabode.com/archive/c...cle/directdie1 on a pentium 3.
The results were still poor :)

The problem with direct die is that you need insane flowrates and pressures to get anything like decent results. As soon as you ramp up the pressure you will spring a leak!

---

yeah, I found several that had tried it and the few that posted screenshots of actual results had worse temps, and that was with cpus with much lower power density and watts than a core i7. Quite a few that tried never even posted results, at least fallwind posted honest results with a screenshot.

IBM has research for years and has access to nanofluids (copper or aluminum oxide or carbon nanotubules that increase 20-40% water conductivity, but that would still only be around 1 W/M*K), etc, yet they are currently milling 10,000 channels to increase surface area in cpu for their attempt at direct die cooling, which pretty much tells me it is not possible without drastically increasing surface area.

An admixture of 0.6 Vol.% of multi-walled CNTs in destilled water resulted in an 34% increase of thermal conductivity. Suspensions with more Vol.% were not stable, wich was the problem in the past (CNTs were difficultly soluble).
We've increased the amount of CNT up to 30Vol.%:up: