Overclock.net banner

1 - 20 of 24 Posts

·
Registered
Joined
·
1,959 Posts
Discussion Starter #1
I was wondering one day what would happen if you tried cooling a CPU with directly with water.
By completely removing the IHS cutting a hole in the bottom of a water block then gluing onto the PCB, allowing water to come in direct contact with the die.
Easier to explain with pictures.





Well the first things first the CPU survived the modifications which was my biggest fear after such butchery.

As for results I wont lie they were dreadful even with extreme pumping power with the water flowing over the die at an estimated 8m/s and modifying the water thermal conductivity with graphite Nano particles.

at best the coolest core was 65°C-85°C-95°C-105°C to the hottest.

My conclusion is the laminar flow of the water is why the last core topped out at 105°C
while the poor thermal conductivity of water is why even the first core was rather warm as I was working in 12°C ambient garage temperature.
 
  • Rep+
Reactions: mfdoom7

·
Registered
Joined
·
1,205 Posts
##Doing this will probably ruin this CPU, but what a cool experiment!##
How thick is the outside of the die? Totally score it with an x-acto knife to make microchannels in the chip itself.

Also, mad brave trying this!
 

·
Registered
Joined
·
1,959 Posts
Discussion Starter #3
Quote:
Originally Posted by McMogg View Post

##Doing this will probably ruin this CPU, but what a cool experiment!##
How thick is the outside of the die? Totally score it with an x-acto knife to make microchannels in the chip itself.

Also, mad brave trying this!
I thought about sand blasting the die to create turbulence but still even if I managed that... I still need to over come the poor thermal conductivity of water 0.58W(m.K) compared to copper 401W(m.K)
Now I know why there is a IHS
to be honest with pure water and a D5 pump the temps were 95°C- 105°C-105°C-105°C

I am wondering how to remove the copper base plate from a perfectly good 3770k, I used a used a NASA approved epoxy glue for its outstanding characteristics, I.e. resistance to de-laminating when submerged in solvents.
it is good for 4.8Ghx @ 1.35v
 

·
Registered
Joined
·
151 Posts
I think the best option for this sorta stuff is to solder copper fins onto the die, almost like taking a soldered CPU and CNC milling fins onto the IHS.
Direct die cooling isn't a great idea unless you pump water through the die itself, but then you're going to struggle with flowrates and whatnot.
 

·
Registered
Joined
·
3 Posts
Hi,

From what I see the water is just passing over the die : that's really not gold for the thermal transfert.
That's why your cores are overheating and reaching 105 degrees at which point they throtte to avoid heating even more.

What you need to do is to force the water to impact the die, this method is called jet impingment cooling.
 

·
Registered
Joined
·
2,410 Posts
Quote:
Originally Posted by Bap2703 View Post

Hi,

From what I see the water is just passing over the die : that's really not gold for the thermal transfert.
That's why your cores are overheating and reaching 105 degrees at which point they throtte to avoid heating even more.

What you need to do is to force the water to impact the die, this method is called jet impingment cooling.
+1
Several years ago Ive seen a thread where someone delidded a 920 and made a custom waterblock where the incoming water would directly impact the die. Needles to say I dont remember the specifics but the temp drops were not substantial over a lapped IHS setup.
 

·
Registered
Joined
·
3 Posts
Interesting, I've never see that one, I'll go search for it.
I am maintaining a list of all attempts made at direct die water cooling, in french but with pictures, here
 

·
Registered
Joined
·
1,959 Posts
Discussion Starter #8
interesting I found this article on jet impingement written in 2007! 50,000 nozzles
bigeyedsmiley.png
bound to cause issues with foreign particles
I am wondering why it never really took off as we are entering an age where liquid cooling is almost a necessity with modern CPUs

Quote:


Looking beyond the limits of air-cooling systems, IBM researchers are taking their concept of branched channel design even further and are developing a novel and promising approach for water-cooling. Called direct jet impingement, it squirts water onto the back of the chip and sucks it off again in a perfectly closed system using an array of up to 50 000 tiny nozzles and a complicated tree-like branched return architecture.

By developing a perfectly closed system, there is also no fear of coolant getting into the electronics on the chips. IBM said its team was able to enhance the cooling capabilities of the system by devising ways to apply it directly to the back of the chip and thereby avoiding the resistive thermal interfaces in between the cooling system and the silicon.

IBM has demonstrated cooling power densities of up to 370W per square centimeter with water as coolant. This is more than six times beyond the current limits of air-cooling techniques at about 75W per square centimeter, according to the researchers. Yet, the system uses much less energy for pumping than other cooling systems do.
Source

another article 2005

http://www.electronics-cooling.com/2005/08/high-powered-chip-cooling-air-and-beyond/
 

·
Registered
Joined
·
1,959 Posts
Discussion Starter #9
Quote:
Originally Posted by Bap2703 View Post

Interesting, I've never see that one, I'll go search for it.
I am maintaining a list of all attempts made at direct die water cooling, in french but with pictures, here
I am just reading through it now using google chrome with translated to English
 

·
Registered
Joined
·
2,562 Posts
The surface area is too small and it gets too hot. Resulting that it cannot transfer all it's heat quicly to the water I guess...

I also would add some cooling fins to it (create bigger surface area) and let the water flow through that.
 

·
Registered
Joined
·
1,959 Posts
Discussion Starter #11
Quote:
Originally Posted by ASUSfreak View Post

The surface area is too small and it gets too hot. Resulting that it cannot transfer all it's heat quicly to the water I guess...

I also would add some cooling fins to it (create bigger surface area) and let the water flow through that.
It would probably work but the idea already becoming too complicated/expensive to be able to sell a kit, most people think de-lidding a CPU a complicated process.
@Bap2703 [ Thread here has intensive experiments with direct die water cooling

major problems I foresee with jet impingment ;
Pumping power I used a 80w unit probably still not enough for jet impingment and what I can understand you will need to generate 60+psi
Risk of failure during installation and during extended operation
Gains at best seem to be marginal over traditional water blocks with simpler 50 jet units
the Number of jets - IBM used 50,000 jets for their experimental work to fully reap the benefits of jet impingment
 

·
Registered
Joined
·
3 Posts
You are right, with direct die cooling to get performance on par with the top "standard" waterblocks you need either
- a high pressure pump : would be noisy since it would probably not be a centrifugal pump
- a microfabricated block : to get many small nozzles, but then in our amateur loops nozzle clogging would be a real issue

One niche advantage is that you can build cheap moderate performance blocks : full plastic with millimetre sized nozzles.
And of course if that matters for you, the aesthetics will probably be unique and it's fun to do.
 

·
Registered
Joined
·
1,959 Posts
Discussion Starter #13
yes it would be a cool talking point with direct die water cooling block
sadly I gave up on my chill boxtoo, due to cost and complicity, it can be done but gains too are still too small for me.
besides with the 1080s coming out it seems we have the power without the heat.
 

·
Registered
Joined
·
2,562 Posts
Well I meant something like this. Paste a copper fin on it and build a plexi box around it with an IN/OUT tube

I think the lower the red box, the faster the water passes, resulting in taking more heat away faster. Then again if you see some real water cooling, it has lot's of ducts/canals in the block for creating a whirlwind inside...

WARNING: I've never did or see a delid or watercooling piece in real life
tongue.gif
so all my "experiance" is based on my drunken mind atm
thumb.gif


Be amazed by my MAGICAL paint art stuff thingy...

 
  • Rep+
Reactions: dmfree88

·
Registered
Joined
·
1,045 Posts
it's been done before and proved to be less effective due to the reduced surface area, the die is quite small and very smooth, which results in minimal heat transfer from small contact patch and no turbulence on the surface of the die. A naked mount water block without the IHS is the best solution for better heat transfer but has it's own inherent problems, ie bad socket contact if mounted incorrectly resulting in ram issues or other stability issues.
 

·
Registered
Joined
·
1,959 Posts
Discussion Starter #16
Could be done
Quote:
Originally Posted by ASUSfreak View Post

Well I meant something like this. Paste a copper fin on it and build a plexi box around it with an IN/OUT tube

I think the lower the red box, the faster the water passes, resulting in taking more heat away faster. Then again if you see some real water cooling, it has lot's of ducts/canals in the block for creating a whirlwind inside...

WARNING: I've never did or see a delid or watercooling piece in real life
tongue.gif
so all my "experiance" is based on my drunken mind atm
thumb.gif


Be amazed by my MAGICAL paint art stuff thingy...

It would look cool Just I would be nervous about the glue delaminating or the varnish/PCB absorbing fluid over time That's why I covered the whole PCB with the copper plate.

Quote:
Originally Posted by LiamG6 View Post

it's been done before and proved to be less effective due to the reduced surface area, the die is quite small and very smooth, which results in minimal heat transfer from small contact patch and no turbulence on the surface of the die. A naked mount water block without the IHS is the best solution for better heat transfer but has it's own inherent problems, ie bad socket contact if mounted incorrectly resulting in ram issues or other stability issues.
I honestly thought you could over come the poor heat transfer by simply increasing the velocity of water across the die surface which is true to a certain extent, but I highly underestimated how fast the fluid would need to go at some stage you are going to start creating heat from the friction.
Even then the 80w pump I used would heat water significantly over a short 1 hour period of pump testing I had 20ltrs of cold water heat up to a luke warm temperature, no actual measured data but it was very noticeable.
The point of being practical was easily exceeded from my testing
 

·
Registered
Joined
·
1,045 Posts
The guy who did it did a pretty extensive test with it. He seemed to be unable to get close to traditional water block performance, and this was about 5-10 years ago, so the water blocks were nowhere near as good as they are now, and I don't believe he tested a naked water block mount without the IHS.

It's been quite a few years since I read the thread but it made sense. ie the heat from the die needs to be spread out over a larger surface area to lower the heat density, and that surface area needs to create turbulence to further improve the heat transfer, which is why we see these pins/slots in modern water blocks, maximum surface area and turbulent water flow which breaks up the laminar flow causing more water to come in contact with the copper water block.

It's pretty hard to beat the water blocks these days, lapping the IHS and water block base still gets you a few degrees, using liquid metal TIM gets you a few degrees, removing the IHS and mounting the water block directly on the die works very well but it just complicates things if you have bad pin contact.

But yes, this idea can work, just not as optimal as a normal water block. If you were going to go to the trouble of testing it, it might be better to investigate how to solder a water block to the die without killing the CPU, as that is a known point of weakness, ie eliminating TIM.
 

·
Registered
Joined
·
1,959 Posts
Discussion Starter #18
Quote:
Originally Posted by LiamG6 View Post

The guy who did it did a pretty extensive test with it. He seemed to be unable to get close to traditional water block performance, and this was about 5-10 years ago, so the water blocks were nowhere near as good as they are now, and I don't believe he tested a naked water block mount without the IHS.

It's been quite a few years since I read the thread but it made sense. ie the heat from the die needs to be spread out over a larger surface area to lower the heat density, and that surface area needs to create turbulence to further improve the heat transfer, which is why we see these pins/slots in modern water blocks, maximum surface area and turbulent water flow which breaks up the laminar flow causing more water to come in contact with the copper water block.

It's pretty hard to beat the water blocks these days, lapping the IHS and water block base still gets you a few degrees, using liquid metal TIM gets you a few degrees, removing the IHS and mounting the water block directly on the die works very well but it just complicates things if you have bad pin contact.

But yes, this idea can work, just not as optimal as a normal water block. If you were going to go to the trouble of testing it, it might be better to investigate how to solder a water block to the die without killing the CPU, as that is a known point of weakness, ie eliminating TIM.
Googling 'jet impingement' brought up a ton of information a term I was unfamiliar with. I believe that the idea was conceive more than 25years ago, I am a bit late to the party
tongue.gif

The idea works and is very effective it just takes 1kw odd pumping power to remove 100w of heat that is the reason it hasn't gone past a conceptual idea.
You could get a 700w refrigeration unit and go sub zero, even then gains aren't that substantial in my opinion.

As for soldering I will just believe intel for their infinite wisdom, the main reason why they don't solder the IHS to the die is; there has been multiple dies crack during the testing phase.
I have no idea why the hexacore and octacores chips are still soldered as they don't suffer from die cracking problems.
 

·
Registered
Joined
·
1,045 Posts
no solder on mainstream CPU's IHS is purely a financial decision, it costs less than solder.

It's possible to do almost anything, but keeping it within reason is the key.

Maybe you should research thermoelectric cooling (TEC/Peltier), it is the logical next step to enhance a custom water cooling loop.

Here is my custom TEC waterblock, made by Foxrena from this forum. The water cools the TEC, the TEC cools the CPU, possible to go sub ambient and even sub zero with this setup. Only change over standard water cooling is higher power usage from the TEC, in this case about 200w of extra power consumption.

 

·
Registered
Joined
·
277 Posts
Sorry to say it but you wont get a better doing that, what make a cooler efficient is to have a system to transfer the heat fast from one point to another one, for example the heat pipes that use gas to move it from the core to the fins. Water is an excellent heat conductor but it takes some time...

Use this analogy: Think of a frying pan, if you put it on oven and let it get hot, if you drop a few drops of water they will evaporate almost at the same moment. In other hand if you throw a cup of water it will take more time to get to the boiling point.

The fins and jet flows use that thinking, what it does is to lower the amount of water that goes through WB at each time and with the fins they increase the surface of the contact of the water and the metal.

What are you doing in this case is making the surface of contact worst than before (just one side of the water flow), and making more water to flow to it but most of that water won't be heated correctly.
Quote:
Originally Posted by feznz View Post

[...] As for soldering I will just believe intel for their infinite wisdom, the main reason why they don't solder the IHS to the die is; there has been multiple dies crack during the testing phase.
I have no idea why the hexacore and octacores chips are still soldered as they don't suffer from die cracking problems.
The difference is the die size, i think the problem of the die cracking was during testing of vibration and thermal expansion, with a bigger die they don't have that problem since it have more surface of contact.
 
1 - 20 of 24 Posts
Top