Direct Die WC

[okashira]

I found a decent thread on this - but it was a good four years old ...

Anyways... I want to try this.
-Direct to IHS
-Direct to Core [removed IHS Ivy Bridge CPU]

I have access to professional 3D design software and a high end 3D printer to make a block. Hell I could even print the hose connection barbs into the block. I'll just use an o-ring to seal around the core or the on the IHS surface. I don't see much of an issue keeping enough clamping force on the block to the CPU to maintain an o-ring seal.


-Is it worth it for Ivy, though? I have read that Ivy's not much of an overclocker. My sandy does 4.6 at quite low voltage on air.

-Has anyone done it direct to core on IVY and determined it will NOT work? (i.e., shorted circuitry, etc) I could always lay down some of my own epoxy around the core if needed.

-Anyone have some good #'s for water cooled Ivy's so we can make a comparsion to block cooled specimens?



I currently have a p8z68-v LX motherboard with a i5-2500K @ 4.6ghz on air. This mobo was about $20 after discount so I suppose ill need a better board. :-P

 

[Thor]

The issue is, if your going to bother going with the hassle of de-lidding and water cooling, and trying to fab a block, then getting the damn thing water tight, and only getting a degree or 2 at best with direct die vs de-lidded and a normal water block, then you might as well go phase change. De-lid and water cool. by all means, but you wont see a big return on temps vs time spent (and possibly parts lost if it leaks) when compared to just normal water cooling.

/edit if you want to do it because i can then more power to you

 

[okashira]

The issue is, if your going to bother going with the hassle of de-lidding and water cooling, and trying to fab a block, then getting the damn thing water tight, and only getting a degree or 2 at best with direct die vs de-lidded and a normal water block, then you might as well go phase change. De-lid and water cool. by all means, but you wont see a big return on temps vs time spent (and possibly parts lost if it leaks) when compared to just normal water cooling.

/edit if you want to do it because i can then more power to you

You have seen only 1 or 2 degrees block compared to direct? I expected more then that. I would not compare this to phase change. phase change has much more power consumption, this is free...

 

[NKDietrich]

Interesting idea, but my experience with removing IHS for air cooling has never really shown much more than a couple degrees improvement at max and I imagine you'd get the same results from water if you could pull this design feat off. Only with more risk of damage to equipment.

 

[okashira]

Interesting idea, but my experience with removing IHS for air cooling has never really shown much more than a couple degrees improvement at max and I imagine you'd get the same results from water if you could pull this design feat off. Only with more risk of damage to equipment.

I've been reading some good arguments on XS that there isnt enough surface area on the die to allow efficient direct water cooling.

Soldering a copper block with alot of surface area to the die directly might be the better option. I'd like to try both :-P

 

[Thor]

You have seen only 1 or 2 degrees block compared to direct? I expected more then that. I would not compare this to phase change. phase change has much more power consumption, this is free...

The only reason I compare it to phase change would be on the order of time spent vs return on investment.

I hadn't considered surface area and weather it would even be possible to cool the chip quick enough. You might have to find a way to jet the water so that it is moving very quickly on the chips surface. This requires pressure, and pressure +water is corrosive, this means hard metal surfaces vs plastic, more tension on the block surface, protecting the sub-straight, bigger pumps, which add more heat into the loop.

Looking at it, if you can cool an old amd thunderbird (palmino any way, the t-bred and barton cores had caps on top that made this very hard/impossible) this way (which where lidless) you should be able to do the same here, your working with a bigger die area I'm not sure about heat output totals core vs core, but I wouldn't suspect it to be so different that it wouldn't work.

I would try it lid on first, go full W/C normal blocks, load the system and log your temps. do a direct cool on a lidded processor load the system and log your temps. If they are better. I would then move back to water cooling with real blocks and a de-lidded processor.
Then move to the de-lidded direct cooled block.

Find your self a small linux distro that can boot very quickly so you can check idle temps and partial load temps. You don't want to have your processor spike while trying to load windows before you ever get to see if it is going to work or not. If idle temps are better then add some load (say 20%) and watch your cores, because you are not running a heatsink any increase in temps will be nearly instant, then just keep adding more load, if it never "runs away" (i.e. the range that makes you cringe and pull the plug from the wall rather than trying for a safe shutdown) fine and dandy. next would be 50% spikes 75% spikes and 100% spikes.

It would be wise to try and figure out what your temp sensors maximum polling rate is and monitor it to a graph as quickly as it can be done. If your doing say 10 second 100% bursts but your polling rate is 15 seconds you may never see the processor hit 110°c then your left scratching your head as to why it died.

I'm trying to remember what else you need to know from my days of doing direct die / phase change / cascade and bong coolers. but as you saw, that thread was over 4 years old, and my stuff was even older. If I remember anything else I'll post back

 

[Incoherent]

We were playing around with this several years ago. Looking at the calculations we came to the conclusion that you would need a firehose level of flow rate in order for it to match a decent waterblock. This was borne out in practice.

This was with the Tbird sized dies however (~100mm^2). The modern dies are bigger, but denser.

I would confidently expect that you would be quite successful direct cooling the IHS since you eliminate a TIM joint, but direct die will be inferior, unless you have an exotic pump setup.

Machining cooling fins/pins/holes into the IHS might be an optimal solution. I am quite tempted to try this myself on my new i7 3770k. But I'd solder a cap on and thus make the IHS, (or a facsimile of one) the base of the waterblock.

An old discussion over at the long abandoned Procooling site

 

[okashira]

We were playing around with this several years ago. Looking at the calculations we came to the conclusion that you would need a firehose level of flow rate in order for it to match a decent waterblock. This was borne out in practice.

This was with the Tbird sized dies however (~100mm^2). The modern dies are bigger, but denser.

I would confidently expect that you would be quite successful direct cooling the IHS since you eliminate a TIM joint, but direct die will be inferior, unless you have an exotic pump setup.

Machining cooling fins/pins/holes into the IHS might be an optimal solution. I am quite tempted to try this myself on my new i7 3770k. But I'd solder a cap on and thus make the IHS, (or a facsimile of one) the base of the waterblock.

An old discussion over at the long abandoned Procooling site

Thanks for the reply. I have not done the hand calc, but it makes alot of sense. Thinks would be even worse with a super dense ivy bridge chip.

My plan is to soldier a high end waterblock directly to the die.

I just need to decide on the CPU that I want to do it to, purchase the mobo, and research the soldier and do some testing. I think soldering should be reasonably simple. I could place a small sheet of it on the die itself, and heat the copper block exernally, and press on the solider into the core until I get fusion.