A possibly stupid idea: direct water on delid

[Bombastinator]

TLDR: does direct water (delidded cpu in the water flow) actually even work?

So I’m looking at these new CPUs and the issue seems to be that it’s not that they produce incredible amount of heat so much as there seems to be a problem getting the heat out of them in the first place.  I had a thought (which is often not a good thing, but they plague me)

 

a water cooler works basically the same way as an air cooler in that the heat connection is from the die though internal TIM, through the heat spreader, through more TIM, through the block plate, to the water.

 

Would it be possible to delid a CPU, seal the block to the edges of the cpu board much like the heat sprees set had been, and actually run the water directly against the die?  Lots of possible problems which would include shorting.  Is this done?  Has it been done? What happened?

[emosun]

you're asking in the wrong place , people here are surprisingly closed minded to new ideas and rather enjoy shooting down anything outside the norm. happens to me plenty when i come up with water cooling ideas

[Bombastinator]

1 minute ago, emosun said:

you're asking in the wrong place , people here are surprisingly closed minded to new ideas and rather enjoy shooting down anything outside the norm. happens to me plenty when i come up with water cooling ideas

Shooting down the idea is fine with me.  It’s very possibly terrible.  All the water cooling experience I’ve got comes from a couple terrible ones with AIO water coolers (bubbles.  People don’t mention the bubble noise. That annoys me. I dropped $200 on those things and they suck) and 20 years of dealing with hot water radiators in an ancient building.

[Letgomyleghoe]

14 minutes ago, Bombastinator said:

TLDR: does direct water (delidded cpu in the water flow) actually even work?

So I’m looking at these new CPUs and the issue seems to be that it’s not that they produce incredible amount of heat so much as there seems to be a problem getting the heat out of them in the first place.  I had a thought (which is often not a good thing, but they plague me)

 

a water cooler works basically the same way as an air cooler in that the heat connection is from the die though internal TIM, through the heat spreader, through more TIM, through the block plate, to the water.

 

Would it be possible to delid a CPU, seal the block to the edges of the cpu board much like the heat sprees set had been, and actually run the water directly against the die?  Lots of possible problems which would include shorting.  Is this done?  Has it been done? What happened?

so your asking about direct die cooling with water?

it might work if you could find a way to seal the cpu to the watercooler so it dint leak all over your system (Epoxy maybe?)

then you would have to worry about corrosion of the die and capacitors (might be avoided by just using a sealent, but i dont know much about the inards of cpus and if they can be sealed or not...)

but im sure it could work and would make for some awesome temps and possibly move to ln2 at some point.

[Chen G]

33 minutes ago, Bombastinator said:

Would it be possible to delid a CPU, seal the block to the edges of the cpu board much like the heat sprees set had been, and actually run the water directly against the die?  Lots of possible problems which would include shorting.  Is this done?  Has it been done? What happened?

It doesn't work because water isn't a good conductor of heat. if you put water directly over a normal die you only have the die's surface area in contact with the water which won't transfer a lot of heat at all.

This is a solvable problem but not DIY solvable, you could for example etch patterns into the die to increase surface area or increase pressure and flow speed of the water to increase heat transfer, and there are legitimate research in these areas.

[WereCat]

If you mean that the water will be directly touching the die, I don't think it has been done... for a good reason.

 

Issue #1

It would be extremely difficult to seal it in such way that the water doesn't get on the small caps that are usually located right next to the die while at the same time getting good and evenly spread mounting pressure. 

 

And that's not even the main issue, even if you got that to work there is 

 

Issue #2

It would simply be worse than using a water block. 

Why? Because you would be missing the fin stack as there would be no copper plate. 

That fin stack is extremely important for the water flow as it forces the water trough multiple channels and it accelerates the cooling process as the water is able to soak more heat from the fins due to much more surface area. 

 

You actually use the copper plate with fins for cooling, the water is there just to transfer the heat into a radiator. 

When you remove those fins and leave just water, you significantly reduce the surface area and thus will get worse cooling. 

[PopsicleHustler]

13 minutes ago, WereCat said:

If you mean that the water will be directly touching the die, I don't think it has been done... for a good reason.

 

Issue #1

It would be extremely difficult to seal it in such way that the water doesn't get on the small caps that are usually located right next to the die while at the same time getting good and evenly spread mounting pressure. 

 

And that's not even the main issue, even if you got that to work there is 

 

Issue #2

It would simply be worse than using a water block. 

Why? Because you would be missing the fin stack as there would be no copper plate. 

That fin stack is extremely important for the water flow as it forces the water trough multiple channels and it accelerates the cooling process as the water is able to soak more heat from the fins due to much more surface area. 

 

You actually use the copper plate with fins for cooling, the water is the just to transfer the heat into a radiator. 

When you remove those fins and leave just water, you significantly reduce the surface area and thus will get worse cooling. 

I'll leave this here.

https://www.overclock.net/forum/134-cooling-experiments/1599292-direct-die-water-cooling.html

[GenericFanboy]

If you can find a way to attach it directly to the die and have it sealed off properly. It ofcourse will work. But definitly not as good as you'd think. More surface area means faster cooling (correct me if i'm wrong), If you have a small concentration of heat then its gonna be harder to cool with water because there really isnt that much space for the water.

 

But hey, If you have the money and the patience and you still want to try. Go for it

[GenericFanboy]

2 minutes ago, PopsicleHustler said:

I'll leave this here.

https://www.overclock.net/forum/134-cooling-experiments/1599292-direct-die-water-cooling.html

Bassically this.

[PopsicleHustler]

Basically @GenericFanboy explained everything in great detail. To have more heat transfer, you need more surface area. CPU die is incredibly small and cannot provide enough area for water to cool it.

 

You would still need a fin stack attached to a CPU die to get somewhat decent cooling. Think of it as reverse radiator. More area means water has more time to give of the heat. Same for taking up the heat. More area = more time for water to absorb the heat and carry it away. Its not a matter of "how much stuff is in between."

[GenericFanboy]

https://forum.hardware.fr/hfr/OverclockingCoolingModding/Water-Xtreme-Cooling/watercooling-impression-fixation-sujet_280898_1.htm#t2322872

 

Pretty interesting even though its in french

[PopsicleHustler]

4 minutes ago, GenericFanboy said:

https://forum.hardware.fr/hfr/OverclockingCoolingModding/Water-Xtreme-Cooling/watercooling-impression-fixation-sujet_280898_1.htm#t2322872

 

Pretty interesting even though its in french

People did some weird stuff with early 2000s hardware.

[Damascus]

100% what @Chen G and @WereCat have mentioned.  A low surface area, perfectly flat and extremely high power density heat exchanger is the worst possible option.  And those three criteria describe a cpu die to perfection   

 

This is why a high quality block costs so much.  To get the highest surface area possible (and therefore, performance) you need to have incredibly fine fins on par with those found on a skivved or stacked tower cooler, with a gap only slightly larger than the fins themselves.  This drives machining complexity way up, and increases a risk of failure 

 

 

Here's a cryos cuplex NEXT, arguably the most advanced cpu block on the market/ever.  The fins are layered, sectioned and maximize the amount of space possible in the area of a waterblock

 

Heatkiller IV, representing the current high end block design.  It's simple, it works and it has dependable performance.  The fins direct flow towards a drain area, maximizing the use of that finnage

 

Regressing backwards, we have the thermaltake W4.  It uses what appears to be skivved aluminum, wherein aluminum sheets are pressed together, creating a finstack.  This is inefficient, but can get the job done.  It takes a hit in directed flow though, leading that surface area to be largely wasted

 

Finally, we get back to the ancient days.  This is the closest you'll get to a flat sheet of copper, with a minimal amount of surface area and machining.  Even without that, it could still get by on the sheer mass and flat area available for heat transfer.

[GenericFanboy]

1 minute ago, PopsicleHustler said:

People did some weird stuff with early 2000s hardware.

Wish i was old enough to be a pc modder at that time, Looks amazingly fun to just try. And share ideas. Really doesnt happen as often anymore.

[PopsicleHustler]

Just now, GenericFanboy said:

Wish i was old enough to be a pc modder at that time, Looks amazingly fun to just try. And share ideas. Really doesn't happen as often anymore.

Those people were Columbuses of our time.  Today everything is know, standardized and boring. You can get any cooling solution from air to LN2 in a matter of days, install it in a matter of hours and go at it with overclocking.

 

I also wish that there was some concept that would pop up and everyone would get on it to make it work using wacky old hardware and pure DIY.

[Damascus]

3 minutes ago, PopsicleHustler said:

Those people were Columbuses of our time.  Today everything is know, standardized and boring. You can get any cooling solution from air to LN2 in a matter of days, install it in a matter of hours and go at it with overclocking.

 

I also wish that there was some concept that would pop up and everyone would get on it to make it work using wacky old hardware and pure DIY.

Just do what I did and drag 2005 into the present, kicking and screaming   It won't be on that level of 1999 jank until I start slicing up some vrm heatsinks, but still fun

[Bombastinator]

21 minutes ago, VegetableStu said:

i mean you could try with an old CPU model that has an extensive history in the overclocking circles, but aside from getting a robust watertight seal (especially under heat) without cracking the die in the process, that might be pretty insane for cooling, considering you're removing the IHS and the coldplate out of the equation

 

could try with the open-cup-and-top-up or waterfall method, but most definitely not in a pressurised loop like if you cut a hole in the coldplate of a watercooling loop (because it'll be the most likely spot to leak first). maybe circulate an open-cup through a radiator, but keep tabs on the amount of water in the open-air loop and the amount of water in the "cup"

Watched the video.  I wanted to yell out to the guys “ there’s a 3/8 copper fitting under the sink that has EXACTLY the same threading as your fittings!  Why are you trying to fit the faucet into this?!”

 

This did cracking thing is concerning.  There wouldn’t be any pressure to speak of on the die at all.  There wouldn’t be anything touching it but water.  You might be including the PCB the die sits on through and there would be pressure on that.  If it flexed I don’t know what would happen.  I was thinking a silicon O ring to seal the pcb to the block.  The things are designed to take a certain amount of pressure from the standard attachment system.  Should be enough to make a single O-ring seal.  There’s only maybe room for double o-rings.  They’d have to be pretty thin ones.
 

possible problems:

1. The pcb is fiberglass and epoxy.  Should be somewhat waterproof, but there’s heat and there’s time. if there’s partial water infiltration over time, swelling could become an issue.  

2. I also don’t know how well it holds against moving water.  Moving water abraded things.  It generally does it very slowly, but the Grand Canyon was made with moving water.

 

3. electrical conductivity.  Pure h2O doesn’t conduct electricity, but it also doesn’t stay pure.  One would have to assume the water is at least an only moderately poor conductor because the purity of the water can’t really be controlled.  It’s going to pick up things.  I don’t know what the effect of a conductive fluid with good infiltration might have.  Could it get under the die?  One option is to use a silicone sealant, but silicone wears even  more easily than the other parts and it would be some really thin silicone.  Having a brief thought about carbon-carbon coatings but they’re better left repressed.

 

Things I know I don’t know:

1: How is the die connected to the PCB?

2: what is the ability of the PCB to resist water?

3: is there anything on that side of the pcb besides the die and the PCB surface?  Laquer? Some glue for the die/PCB bond?

[Bombastinator]

37 minutes ago, PopsicleHustler said:

I'll leave this here.

https://www.overclock.net/forum/134-cooling-experiments/1599292-direct-die-water-cooling.html

Well that solves that one.  Metal contact conduction is actually so much better than water contact, even moving water that it’s better to spread it first even through several layers of TIM.  thankyou

[MaxPower4U]

On 1/10/2020 at 4:39 PM, Chen G said:

It doesn't work because water isn't a good conductor of heat. if you put water directly over a normal die you only have the die's surface area in contact with the water which won't transfer a lot of heat at all.

 

Water is an excellent conductor of heat, water directly on the die would be the most efficient way to cool a CPU.

 

[Bombastinator]

2 hours ago, MaxPower4U said:

Water is an excellent conductor of heat, water directly on the die would be the most efficient way to cool a CPU.

 

This is what I thought too, Except it seems to apparently be be it’s not. Apparently the heat transfer of metal to metal conduction is still better even with all the TIM.  Should have realized it in retrospect.  The numbers for metal heat transfer are double that of water.  I thought movement of water would make the difference.  Seems it doesn’t.

 If one could get the same kind of fin stacks onto the die that are apparently put onto water blocks it would be true.  Might work with even with less fin stack.  Apparently there needs to be at least something though.  There just doesn’t seem to be enough surface area on just a bare die.  Looking at the fin stacks shown they easily quintuple surface area or more to get enough heat transfer.  If a die was built at the factory to do this it might be possible. Fins cut into the silicon for example.  I don’t know how possible that is. If it is it would require a manufacturer to make a cpu “just for water cooling” at additional expense.  Water cooling is apparently a non manufacturer hack that’s grown in to a manufactured thing.  It’s still a hack though.