Direct contact watercooling!

This was a surprise! I never thought of cooling the chip directly, just imagine if you got a bad mount though and coolant goes shooting out the sides... :cool:

Quote:

---

The new system – which further increases efficiency by using a system of jet impingement cooling where the water actually makes direct contact with the surface of the chip – will use a sealed-loop system containing around ten [liters] of water, which will be pumped through the system three times every minute.

---

http://www.bit-tech.net/news/hardwar...upercomputer/1

There was a fairly recent thread around here from someone direct-to-die cooling a i7, I think. I'll have to see if I can dig it up.

Yea, there have been a few trys done on this. I think there are a few threads on here of people trying it.

hmmm :confused:

direct die has been tried before a few time's IIRC maybe this time they will get it right?

Quote:

---

Originally Posted by madmaxx

hmmm :confused:

direct die has been tried before a few time's IIRC maybe this time they will get it right?

---

No. They use 60'c water at 30LPM. They are not looking for ideal temps, just stable ones (under 85c). To the people over here, they are doin' it wrong. :)

Not sure where Bit-tech got that they were using direct die cooling. IBM simply says they have a watercooling block for each chip and are doing "chip-level" water cooling. Read the actual press release and look at the picture of the blade and it doesn't look like anything special. Doesn't look like direct die cooling to me.

http://www-03.ibm.com/press/us/en/pr...ease/27816.wss
http://www.engadget.com/2009/06/25/w...ts-dorm-rooms/

I thought the consensus was that direct-die wasn't as good because water did not absorb heat as quick as copper, so you need to make the water go insanely fast to make up for it.

Bringing the heat from the die to copper is more efficient than die to water, and allows you to add more surface area to the copper. Once you have more surface area, the water will pick up more heat from it at slower speeds.

Quote:

---

Originally Posted by 3Z3VH

I thought the consensus was that direct-die wasn't as good because water did not absorb heat as quick as copper, so you need to make the water go insanely fast to make up for it.

Bringing the heat from the die to copper is more efficient than die to water, and allows you to add more surface area to the copper. Once you have more surface area, the water will pick up more heat from it at slower speeds.

---

Exactly. This was confirmed in the i7 tests dones here recently.

Quote:

---

Originally Posted by 3Z3VH

I thought the consensus was that direct-die wasn't as good because water did not absorb heat as quick as copper, so you need to make the water go insanely fast to make up for it.

Bringing the heat from the die to copper is more efficient than die to water, and allows you to add more surface area to the copper. Once you have more surface area, the water will pick up more heat from it at slower speeds.

---

so im lost in this statement.

if water cant pick up the heat from the silicon better then copper, which u are right, i think its because silcon might be paritally hydrophobic... How would increasing flow compensate since water can only pick up a set number of molecules anyhow each pass?

Quote:

---

Originally Posted by NaeKuh

so im lost in this statement.

if water cant pick up the heat from the silicon better then copper, which u are right, i think its because silcon might be paritally hydrophobic... How would increasing flow compensate since water can only pick up a set number of molecules anyhow each pass?

---

wut?

Quote:

---

Originally Posted by SoulsCollective

wut?

---

1 water molecule can only pick up a limited amount of heat.

The more molecules which have picked up heat collectively represents how hot the water is.

So how is it when your limited on the pick up side, more flow will help you?

do u understand the bottleneck i am pointing out?

Ah, you meant heat. You originally said water molecules pick up other molecules, which made me go :confused:

Assuming each molecule can pick up a certain amount of heat, taking into account inefficiencies, the more molecules you have impacting the die surface in a given period, the more heat can be removed - thermal transfer efficiency for a set silicon/water molecule pair decreases exponentially over time. Even if silicon wafers are hydrophilic, there isn't some set number of molecules that will be "allowed" to contact the surface - you've just got to overcome additional inefficiencies caused by the repulsion of X water molecules.

Although really, any hydrophilic effect will be miniscule in comparison to the surface-area issue.

Souls explained it well. Water is flowing over the die, so effectively, your surface area is calculated by how much water passes over the surface of the die in a given amount of time... where as with a passive cooled copper block, the surface area is static, because your copper isn't flowing over the die.

So the water will get a much larger surface area, but absorbs water MUCH slower than copper... so if you use copper to get the heat away from the die, then flow water over the copper, you can make the surface of the copper as large as you want, giving the water exponentially more surface area to absorb heat from.

In any given amount of time, to get the same volume of water to transfer over the hot surface area, you will have to have the water moving VERY fast over the die to make up for the raw surface area of water flowing over copper.

Quote:

---

Originally Posted by Bradan

No.

---

"no" what? there have quite a few attempt's both by regular people and companies to achieve a feasible direct die cooling solution, so far no dice :(

higher flow + turbulence can improve the cooling capabilitie's IIRC

Wouldn't you be able to achieve better results creating a larger chamber over the die and spraying the coolant (thus surrounding the die in "mist") rather than letting it flow over it?

I think, then, you have the issue of "how do you now take the warm mist away in an efficient manner to allow more to be sprayed on ?"

Water is nice because it doesn't expand/contract nearly as much as air when it heats up, so you can run a closed loop without worry of high pressures causing leaks, and because its density gives it a higher capacity for carrying heat away from the block than air.

Quote:

---

Originally Posted by 3Z3VH

I think, then, you have the issue of "how do you now take the warm mist away in an efficient manner to allow more to be sprayed on ?"

Water is nice because it doesn't expand/contract nearly as much as air when it heats up, so you can run a closed loop without worry of high pressures causing leaks, and because its density gives it a higher capacity for carrying heat away from the block than air.

---

I see your point. Perhaps sprays would be better used in combination with chilled and more dense coolants (like fluorinert cooled with a waterchiller for instance) which is, of course, a completely different ballgame...

Phase Change cooling still scares me. :D

Too much noise, and power hungry... I just can't bring myself to do it, even if it is great cooling !

Quote:

---

Originally Posted by 3Z3VH

Phase Change cooling still scares me. :D

Too much noise, and power hungry... I just can't bring myself to do it, even if it is great cooling !

---

Same here, but I really want to try and piece my own system together just for the heck of it.

Quote:

---

Originally Posted by 3Z3VH

Phase Change cooling still scares me. :D

Too much noise, and power hungry... I just can't bring myself to do it, even if it is great cooling !

---

Without going to phase change, I had a good "dwell" on the chilled cooling section.

In the end of the day it does offer advantages over liquid cooling, and with the wonga I've spent on the stuff, I could probably have bought half decent kit...

Then I saw the power consumption of the water chillers and... ehm... :eek:

Not only because of the consumption of the chiller itself, which is grotesque, also because I like to have good 15-20 mins of UPS run... To keep the rig AND the chiller running for 20 mins, the UPS was worth the price of a small flat :p: :p: :p:

Quote:

---

Originally Posted by SoulsCollective

Ah, you meant heat. You originally said water molecules pick up other molecules, which made me go :confused:

Assuming each molecule can pick up a certain amount of heat, taking into account inefficiencies, the more molecules you have impacting the die surface in a given period, the more heat can be removed - thermal transfer efficiency for a set silicon/water molecule pair decreases exponentially over time. Even if silicon wafers are hydrophilic, there isn't some set number of molecules that will be "allowed" to contact the surface - you've just got to overcome additional inefficiencies caused by the repulsion of X water molecules.

Although really, any hydrophilic effect will be miniscule in comparison to the surface-area issue.

---

okey your looking at it from the surface impact point aspect of having more flow.

Assume this, your block is a mini reservoir, which is the case of most direct die/ihs blocks.

You wouldnt have a shortage of water molecules picking up heat. Infact you would have too much water that the flow would dump the water b4 it was even 5% utilized.

In short, i think you wouldnt see any scaling of flow after u hit that critcal point like in d-tek blocks with no nozzles.

If we used your example, then all blocks would scale, however this is not the point we see. And martin also showed us that we get hardly no benifit after we pass the 2gpm mark. :P

Quote:

---

Originally Posted by NaeKuh

You wouldnt have a shortage of water molecules picking up heat. Infact you would have too much water that the flow would dump the water b4 it was even 5% utilized.

In short, i think you wouldnt see any scaling of flow after u hit that critcal point like in d-tek blocks with no nozzles.

---

Exactly - you're proving the point :) Talking about 'utilizing' the heat-carrying capacity of water molecules is starting to sound a lot like the old low-flow arguments - shouldn't the water spend more time in the CPU block/rad so it can pick up more heat? Fact is, as long as that water in the block is actually hitting the die (so we're assuming injector design here, not CPU-block-as-mini-reservoir as you put it), while each molecule may pick up less total heat because it isn't in contact with the die for as long, because thermal efficiency decreases exponentially with time, each molecule has a higher heat-pickup-rate-per-time-unit, because we aren't progressing as far down the heat/time efficiency curve. Given that, and that we effectively have an unlimited supply of water molecules, more flow will give you better temps, as you have a higher total rate of heat pickup from the die surface.

Quote:

---

If we used your example, then all blocks would scale, however this is not the point we see. And martin also showed us that we get hardly no benifit after we pass the 2gpm mark. :P

---

They do, and that's only partially true. I'm glad you said "hardly no benefit", because that's completely true - diminishing returns of course apply. But the principle is sound and continues to apply, it's just that our ability to measure the benefits now starts to come into play. You're still getting better temps with more flow (ignoring for the sake of argument pump heat-dump), you just can't tell becaue your measuring instruments aren't sufficiently precise (or accurate, for that matter).

Quote:

---

Originally Posted by 3Z3VH

Souls explained it well. Water is flowing over the die, so effectively, your surface area is calculated by how much water passes over the surface of the die in a given amount of time... where as with a passive cooled copper block, the surface area is static, because your copper isn't flowing over the die.

So the water will get a much larger surface area, but absorbs water MUCH slower than copper... so if you use copper to get the heat away from the die, then flow water over the copper, you can make the surface of the copper as large as you want, giving the water exponentially more surface area to absorb heat from.

In any given amount of time, to get the same volume of water to transfer over the hot surface area, you will have to have the water moving VERY fast over the die to make up for the raw surface area of water flowing over copper.

---

+1...which is why passive direct die cooling wont work on i7, I drew a pic back when arguing with someone else ...there is no getting around the very small surface area of die, and .58 W/M*k heat conductance of water is not going to dissipate 180W. I could see IHS/waterblock combo without thermal paste, ie soldered as doing better, but not direct die.

http://img188.imageshack.us/img188/1...rectdiepst.jpg

Quote:

---

Originally Posted by SoulsCollective

They do, and that's only partially true. I'm glad you said "hardly no benefit", because that's completely true - diminishing returns of course apply. But the principle is sound and continues to apply, it's just that our ability to measure the benefits now starts to come into play. You're still getting better temps with more flow (ignoring for the sake of argument pump heat-dump), you just can't tell becaue your measuring instruments aren't sufficiently precise (or accurate, for that matter).

---

actually ud be suprised...

lets see if i can get vapor to shed some light.

^
^
^
the pic give clear understanding

http://www.xtremesystems.org/forums/...d.php?t=222040 -- this guy seems to get good direct-die results with slightly chilled water.

I've been looking for the lowest temps with the least power consumption. Although it is true that lower temps will allow the same clocks at lower voltages = lower overall CPU power consumption (which I'd be willing to allocate to the cooling)

There seems to be some conflicting evidence for direct-die cooling here. I though I "knew" that it wasn't as effective as using copper to expand the surface area, but uh... I'll look into it again later.

Phase-change is tempting, but I just can't justify the power/noise.

When you add all those small copper pins to a water block, you're increasing the surface area by over 100x. For direct die cooling to be as efficient, you'd have to use a medium that is 100x as effective as water.

Quote:

---

Originally Posted by Vampiyer

http://www.xtremesystems.org/forums/...d.php?t=222040 -- this guy seems to get good direct-die results with slightly chilled water.

I've been looking for the lowest temps with the least power consumption. Although it is true that lower temps will allow the same clocks at lower voltages = lower overall CPU power consumption (which I'd be willing to allocate to the cooling)

There seems to be some conflicting evidence for direct-die cooling here. I though I "knew" that it wasn't as effective as using copper to expand the surface area, but uh... I'll look into it again later.

Phase-change is tempting, but I just can't justify the power/noise.

---

In post below this person did direct die and reported actual results with pics. At 1.13 vcore and stock mhz, load temps were 85C immediately on prime, or at least 55C above ambient and probably water temp.
http://www.xtremesystems.org/forums/...84&postcount=2

Below link is pic of temperature gradient (courtesy of stanford paper) through die at 78W load ...now imagine an i7 at 150+W OCed gradient just though the die. Removing the IHS or a heatsink increases the gradient through die according to intel, unless you replace with liquid that has a thermal conductance similar to copper. Fallwind's experiment in link above agrees with intel docs and experiments. The link you quoted does not.
http://www.xtremesystems.org/forums/...postcount=3189

http://vapor.skinneelabs.com/i7/Roun...Dependence.png

How many of you guys can interpret that graph?

Because u'll see what i mean about my statement.

to elaborate, the ihs on a cpu block would be way b4 the d-tek's which would mean, flow increases wouldnt do jack squat to help it.

"The company said commercial availability of the technology is "very near". In a first step, standard server blades with readily available third-party CPU chips will be subject to the liquid cooling approach. However, IBM researchers are already working on 3D chip technologies where micro cooling conducts, much like vascular systems, are used to pump the water through the chips instead through a copper heat sink. This will further reduce the heat resistance between chip and cooling system and thus increase the effectiveness; IBM claims a cooling effect of up to 350 W/cm2. The company however did not specify when this chip-scale liquid cooling will be commercially available."
http://eetimes.eu/germany/218000027

You guys have supported your case well enough.

Of course it's obvious that copper is more thermally conductive than water, and that the water will make more contact with a copper block as well.

I couldn't reconcile the results from the thread I linked with what we've come to know as common knowledge, but I guess I have enough information now. I still can't explain those results though :confused: -- maybe I should read through that thread in-depth. Perhaps I missed something...

Quote:

---

Originally Posted by Chumbucket843

"The company said commercial availability of the technology is "very near". In a first step, standard server blades with readily available third-party CPU chips will be subject to the liquid cooling approach. However, IBM researchers are already working on 3D chip technologies where micro cooling conducts, much like vascular systems, are used to pump the water through the chips instead through a copper heat sink. This will further reduce the heat resistance between chip and cooling system and thus increase the effectiveness; IBM claims a cooling effect of up to 350 W/cm2. The company however did not specify when this chip-scale liquid cooling will be commercially available."
http://eetimes.eu/germany/218000027

---

If you are going to cool the die with water, you have to drastically increase surface area first, not just squirt water on back die, a vascual channel with 10,000 pillars would be like creating a water block's surface area on a micro scale in the chip.... "to solve the problem, IBM etched a liquid aqueduct into the silicon oxide on the back each die. That creates a water-filled separating cavity with 10,000 pillars, each housing a copper via surrounded by silicon oxide." http://www.eetindia.co.in/ART_880052...T_855c455e.HTM

Still interesting even though 5-10 years away according to IBM. Basically designing a high surface area water cooling solution in the die on micro scale...maybe one day intel will sell water cooled chips that come ready with a larger IHS and barbs.

I think the biggest obstacle to the sort of capillary direct die cooling being proposed here by IBM will be clogging of the cooling channels. This will definitely not be a user serviceable cooling system.

I've read about various direct die cooling attempts over the years and the reduced core size seems to be making this approach less viable. I believe the optimal technique for current processor designs would be IHS removal for direct Water block contact or "direct IHS cooling".

If one were to drill a circle of small holes into(but not through) the top of the IHS and secure a block top with nozzle directly above(perhaps an O-ring seal tensioned with 4 bolt mounting through mobo?) the soldered on IHS(on I7 anyway) would probably transfer heat better than any thermal paste ever could.

I don't know if anyone has tried this yet but I'd love to see it.

Quote:

---

Originally Posted by rge

...maybe one day intel will sell water cooled chips that come ready with a larger IHS and barbs.

---

That's basically what I was thinking. We can dream of a larger copper (or silver!) soldered IHS-waterblock, or we can MAKE one...

If we are able to remove the IHS, then perhaps it's possible to re-solder an IHS-block on eh?

Oh well... that's far too much effort for me at this juncture. I'm not Xtreme enough (yet). Apparently "rge" is according to the claim beneath his name.

here is a thread about removing ihs on the core i7. he isnt quite done yet with testing but he said:

Quote:

---

Here's a teaser, but don't get too excited. The results aren't that good and some may be surprised.

---

Quote:

---

Originally Posted by rge

maybe one day intel will sell water cooled chips that come ready with a larger IHS and barbs.

---

and just how are you expected to get that clamp over to secure your socket?

Intel will never sell a h2o computer until it was completely built by themselves.

Its too much of a liability on there end, and also they would most likely charge and arm and leg for it.

At that point i think they would use carbon nanotubes on an air sink.

Quote:

---

Originally Posted by NaeKuh

and just how are you expected to get that clamp over to secure your socket?

Intel will never sell a h2o computer until it was completely built by themselves.

Its too much of a liability on there end, and also they would most likely charge and arm and leg for it.

At that point i think they would use carbon nanotubes on an air sink.

---

I would not have expected anyone to take a half baked comment seriously...nor would I waste my time seriously trying to guess what intel or any other company would do years in the future.

Quote:

---

Originally Posted by rge

+1...which is why passive direct die cooling wont work on i7, I drew a pic back when arguing with someone else ...there is no getting around the very small surface area of die, and .58 W/M*k heat conductance of water is not going to dissipate 180W. I could see IHS/waterblock combo without thermal paste, ie soldered as doing better, but not direct die.

http://img188.imageshack.us/img188/1...rectdiepst.jpg

---

i think you can't argument with the heat conductance of water / copper...
You have to mind that the water comes with high speed and only stays for a very short time directly on that cpu. So take a stick of copper and heat one side of it... Don't you think the waterflow will be faster than the heat conducting through the metal?

The fakt is, that direct die watercooling offers continuously new cold material (water) to dispate the heat. Copper heats on and on, there is no new material coming... heat have to to through the material.

I think direct die cooling only works if the water comes with a quite fast an continuous stream over the die. :)

Example:

Try to brun a hole in a plastic bottle filled with water. Think you can't...
Try to burn a hole in a bottle filled with copper :D - or a brick of copper packed in plastic... i think you will know what i'm talking about...

I answered the same question in post here
http://www.xtremesystems.org/forums/...1&postcount=60

I think you guys are missing the most interesting lesson to be learned from that IBM story:

But first here are my responses to this thread so far:

(1) First of all it's a no-brainer that using a copper heat-spreader is better than direct die cooling (if you're still not getting why, you really should've taken that second semester physics course, or at least a course on thermodynamics back in college, or take one now, it's never too late.)

(2) Secondly, increasing flow will always yield better temps. However, at some point, as you continue increasing flow rate (assuming no additional heat dump from your pump) you will approach an asymptotic limit of the heat dissipation capacity of your system. As such you will forever approach that limit, but you will never reach it (just like in Zeno's paradox), and furthermore, as you increase flow rate, the gains become infinitesimally smaller and smaller as you increase flow, and as a consequence they will become harder and harder to measure (i.e., gains will first be in the -1C range, then in -.1C range, then -.001C, then -.000001C, etc.), and it appears from Martin's testing that at approximately 2 gpm, in his lab, with his equipment, at his altitude above sea level, at his ambient temp and humidity, the gains begin to become too small to measure with his equipment at that point, and I believe from memory that Martin's equipment is only accurate to to the tenth or hundredth of a degree Centigrade. In other words, there are still gains in heat dissipation to be had above 2 gpm, but they are just too small to be measured with Martin's equipment, because he is approaching the asymptotic limit of his system. Now your system may be different, you may see more substantial gains when surpassing 2gpm, but you may not. But either way, you will still be getting better heat dissipation, you just maynot able to measure them. But one indisputable benefit you will get from higher flow rates is faster responsiveness to changes in heat.

(3) And thirdly, the over-looked, but most interesting take-away nugget in that article, at least for me anyway, is that the IBM engineers (who are no dummies) like the idea of "cooling" their processors with "hot" water, as their water starts off at 60C! They are actually heating up the water to start it at 60C! That to me is astounding. The science of it is great, because it clearly demonstrates an important but often overlooked point of water "cooling": that there is no such thing as "cold". That's right sweetcheeks, you read it right, there is no such thing as cold, there is only heat and the relative absence of heat. And the IBM boys know that water can "store" heat from 0C up to 100C, and as such their water cooling system is effective at removing heat from their processors even when the water in the system starts at 60C. Now, why exactly they are doing this, that is a bit of a mystery. I do not believe for one minute that IBM is "going green", and I do not trust their stated goal of wanting to use the water to heat the building. Gimme a break, this is IBM we're talking about, and IBM has never shown any concern for the environment in its 100+ years of existence. (If you doubt me, take a drive up to Armonk, New York and check out IBM and their employees. When you are there, don't be frightened by all of those identical white male robots with white shirts and red ties coming out the front door at closing time, those aren't robots, they're IBM employees.) I think it has more to do with the fact that they want to control the temp range of the water for some reason, and they have determined that the 60-65C range is optimal for their purposes, and have tried to spin it as some kind of earth-saving propaganda, which is ridiculous. But the science is not. I applaud them for their science, but condemn them for their false justification. Regardless of my personal feelings, this shows that you do not "need" to have water that is 2C above ambient to have an effective water "cooling" system. You can still do a great job at 40-45C above ambient!