Direct Die Cooling?

[TheStapleGunKid]

Hey guys, this is my first post, but i've been scanning the boards for a while now. Anyway to get to my question, i was just wondering what your veiws on Direct Die cooling are?
I currently am running a TB 750 oc'd to 1Ghz. My cooling is taken care of by a nice SunRoc water cooler (the kind ya put the blue bottles on) And a homemade waterblock(which consists of a part of a pvc pipe end, and a flat peice of copper, as the base).
My idle temps are about 15C, and under full load is goes to 28C. So my next step was going to be running the water right over the cpu die, but i was just wondering if the stuff on the cpu (the green glue... or blue) is in fact water tight, as i'd planned on applying epoxy to the bridges etc....
Has anyone tried this, with any success? And were the gains worth the effort?

Thanks

[phiber]

as far as i know....direct die cooling is worth the hassle...nor the risk IMHO i have heard that over time the covering of the DIE will wear away and that the surrounding structure can absorb the water...therefore shorting the CPU

another thing is..for direct die cooling to be effective you would need to get a fast pump about 350GPH so water travels over the CPU alot more.... there is also the fact that you have to seal it and stuff...so i dont think its worth the hassle....

[WarriorII]

15C/28C ? Why bother?

I mean really. 15/28C

Go ahead rub it in some more why don't ya.......


Good Luck.
:cool:

[The Overclocker]

my cpu is direct die cooled, i covered the bridges with epoxy to prevent water from getteing to them, my temps are idle 24 and load 24, the ambient is about 20, your temps are quite good already.

[TheStapleGunKid]

Hey the overclocker, how long has your setup been running? And the reason i want to switch to direct die cooling, is the nonexistant delta, between idle, and full load. The reason my temps are as low as they are, is because the water is cooled, in the chiller.

[Aesik]

I have yet to see any direct die cooling that made any significant difference from water cooling. The main problem being is that you drastically reduce the effective suface area over which heat transfer can occur.

[Neco]

Originally posted by Aesik
I have yet to see any direct die cooling that made any significant difference from water cooling. The main problem being is that you drastically reduce the effective suface area over which heat transfer can occur.

Maybe it's time for The Overclocker to direct die cool an Athlon XP 2000 + :beer:


I'd love to see the results from that vs. normal watercooling..


don't look at me to buy the CPU tho, I'm always broke.... :D

[MiseryQ]

My first water cooling experiment was diec die and while the setup at the time wasn't optimal, I still was'nt happy with it...

The 2nd block I made perfomed better before I even "tuned" the system...

[UserName]

The components on the CPU need not be sealed, in my opinion.

Aseik comment is correct in that there is less surface area. He did not mention the 2 thermal barrier jumps needed in a water block design compared to one in direct die. Thermal jumps are very expensive in tearms of heat transfer.

The jury is still out on direct die cooling. Very little work has been done here. In general all of the DD setups i have seen are very poor in design and still seem to be about the same in performance. We are seeing volumous direct die cavities when very small DD cavities would be indicated.

It seems clear to me that a decent DD design aught to kick the pants off a WB setup.

[Aesik]

Just some food for thought...if direct die cooling is all that and the extra surface area is not needed from a water block, why not direct die cool with a concentrated stream of air? We wouldn't need heat sinks at all for air cooling, just a high-speed nozzle!

[Yes, I'm playing the devil's advocate and just trying to get you all to think ;) ]

The very reason we use water blocks, or heat sinks for that matter, is to increase the surface area over which heat transfer can occur. As Username noted, there is a thermal resistance created at every interface or material change. That's exactly why we spend the extra time and bucks to lap our heatsinks and meticulously apply Artic Silver or some other material. These are all attempts to minimize the temperature drop across the material interfaces.

If the thermal resistance is minimized, there is more to be gained by increasing the surface area for heat transfer than is lost at the interface. This is a must in air cooling. Now with water cooling the convection heat transfer coefficient is greatly enhanced by the increased thermal conductivity of water over air, and it just may be possible to pull off some decent direct die cooling (as can be witnessed by a few around here who have done it to some degree of success). However, and this is my opinion, I believe we are better off trying to optimize the design of a water block and work on decreasing the thermal resistance between the silicon die and the water block.

[UserName]

Are you kidding?

We can direct die cool with a concentrated stream of air. You and i both know it's done all the time. In fact a 5HP compresser and a nozzel would prolly work better than most heat sinks or water blocks. If you added a water mister you could get close to freezing.

So after we all think about it it's a great cooling idea that is used every day and works great in many applications. Much like phase change and LN2.

Or were we supposed to think it wont work?

Now lets think about the cost of a thermal jump. How many degree temp differance must be accepted with a block? What is the closest we can get the block to the die? Within 5-10C? or worse? If we all laped perfectaly and set the block on perfectly? And then how big a jump to the water temp? another 2-5C? So we have a total of 7-15C temp delta with a water block?

How low a temp delta can we get, in therory, with water cooling? if we have a, i guess 132 square mm die what kind of thermal coefecient do we need to keep it within 2C? is that do able with water? Very.

So for the reasons above i think DD coling has not come close to its potential.

[Aesik]

Username, where are you getting all these numbers you're posting? You are tossing out numbers that have no basis. Unless you have some analysis to show where these numbers come from or have some empirical data, they are pretty much smoke in the wind. I've always planned to run some simulations of direct die vs. block cooling, but unfortunately have not had the time to do so yet.

You say that direct die air cooling is done all the time yet I haven't ever seen it applied to something that puts out so much heat in such a small area. If you have some examples, I'd love to see them. I've seen impingment cooling used on ICs that put out a lot less power per area, but never on a modern processor.

*IF* you perfectly lap a copper block and the die and perfectly mounted them, you would get a near-zero degree temperature drop. Of course that's not possible though.

I never said that direct die cooling wouldn't work, others have proven that it is very possible. What I did say was that I don't personally believe that it is the best way to see improvements right now. Water blocks themselves have a long ways to go for improvement as well and I'm spending my time researching this area. The last thing I want to do is discourage others from trying direct die cooling as a lot could be learned from it.

[UserName]

Oh come on.

You were trying to get us to think about a concentrated stream of air for cooling.

You very well know that air moving a some speed over some surface area is equavlent to faster air moving over some smaller surface area. I ment to say this is done all the time for conditions that require it. For example machining or other manufacturing enviroments which require a small surface to be cooled cheaply. The sentance implied it was done with IC's. I know of no IC's being cooled this way. It would certianly not be tough to do. Like Ln2 or phase change.

You very well know that the cost of a thermal jump is very expensive. It is very misleading to suggest to those with out your background, and who value your opinion, that the cost of a thermal jump can be made small. ASII's .7C thermal conductivity can't help here.

You very well know that no amount of lapping will match any two surfaces to anything close to zero. How small can it get, practicaly? What does that mean for a temp delta?

If you have emperical data about the delta between the die and blocks go ahead and post what you got. If you don't, get out one of thoes big books and give us the theory. What numbers would you say are better? If the numbers i stated are all wrong i will confess my error. If they are correct or conserivitive, and we think a thermal coeffecient to beat that delta can be done with water, that we see that the true potential is for DD cooling.

If you don't wish to discourage others from trying DD cooling then a fairer statement of the facts, that included the avantages as well as the disavantages, would be in order, wouldn't it?

[The Overclocker]

Originally posted by TheStapleGunKid
Hey the overclocker, how long has your setup been running? And the reason i want to switch to direct die cooling, is the nonexistant delta, between idle, and full load. The reason my temps are as low as they are, is because the water is cooled, in the chiller.

i have had my setup running since around december.

i dont know if you will get 0 delta between idle and laod, but it will definatly get you closer then what you have now.

Neco: i am going to get a new cpu soon and will probably watercool it direct die again, at the moment money is a problem (i dont have any at all )

Aesik: water has a much higher thermal capacity then anything readliy avablable, added to the fact thet there is almost 0 gap between the water and core, results in a much better temparture. i have used a becooling jagged egde waterblock on my same system with the same parts and only achived 26 degrees celcus load, this waterblock is equial to a maze2 and prefoms very well, but is no match against direct die.

[Aesik]

I wish I had some good numbers for you right now, but I don't just yet. Unfortunately it's not the easiest thing to analyze, but I could probably get and extremely rough idea of the velocity and mass flow rates that would be necessary for effective direct die cooling with air. I have a gut feeling that it would take an insane amount of air to get any reasonable results. Air coming from a high-speed jet nozzle and impinging on a flat surface is going to make hella noise too!

The contact resistence from cpu--->Artic Silver--->Heat Sink/Water Block will drastically vary from system to system. It will never have a zero temperature drop, but it can be made to be very low. What we are dealing with is what is called the "composite thermal wall" by most texts. From the point of the heat source to the first material (AS) there is a temp drop, then another temp drop through the thickness of the material, a drop across the interface of the first and second materials (AS--->Copper), then a final drop as the heat path gets further away from the concentrated source in the block itself. We have control over the smoothness of the copper/AS interface, the thickness of the AS, and to a lesser degree (only for the brave!) the cpu silicon itself. A lot of work and care will net a system that has a minimilistic drop across the interfaces. I don't have raw numbers for the particular case of cpu cooling in pcs, but I have scads of data on other interfaces. I'd really hate to toss out any numbers off the top of my head right now, but when I get my personal testing equipment set up I plan on working with such experiments. Not sure if BillA has measured both die temp and heatsink temp as I don't have all of his information with me, but maybe he can show us some data.

So, from my viewpoint here are what I see as the advantages and disadvantages of direct die cooling with water:

Advantages:
Improved heat path directly into the fluid
Simpler system overall (less total parts)

Disadvantages:
Much less surface area for thermal transfer
Not sure it's safe for extended use on an organic substrate
Could need a higher overall flow rate

Of course the above statements are totally dependent on the design of each individual system.

The more I think about the possibilities and methods, the more it intrigues me. However, gut feel and very limited analysis still turns me towards improving on waterblocks for now. Too bad I don't get paid to do this and could work on it for 40+ hours a week!

[Aesik]

Overclocker...do you have any idea what your actual volumetric or mass flow rate is? What pump, total length of tubing, elevation difference between highest and lowest point, etc do you have? I'd be really interested to hear more details.

Of course water has a much better coeffecient of thermal conductivity of any other usable fluid. But we aren't really comparing fluids here, we are comparing going from silicon to water, or silicon to copper to water.

Leufken claims "a 20 degree temp drop" from standard waterblock cooling. This I'd have to see since the vast majority of water coolers are not 20 degrees above ambient in the first place.

[UserName]

When you get a job doing it let me and billA know. 'tween the three of us we would spank this bad boy.

the substrate is ceramic right? no more organic then a chem definition that says anything with carbon is organic. Or am i way off base here?

again we are gonna dissagree on the flow rate thing as i think all that is nessary is velocity over the die. but we not gonna talk about that again.

do you reall think you could get the temp differance to lesss than 5-10C?
I know you dont want to throw out numbers... but really, 5-10C on a 2 - 3 barrier thermal jump (i skipped the AS2 barrier).
Yes we can get some sort of flatness, but most wont get it near smooth. and a real production enviroment where jo user slaps that thing on? Do you really think less that 5-10C is possable in the best of hands? What about in joe user's hands?

In a DD setup the joe user is much less an issue.

The big question in my mind is, what kind of water speed do we need to get the die within 2C of the water temp. Or what thermal coeffecient do we need, and what water speed over the die will get us there?

[Aesik]

Yes, I *think* a temp differences of under 5 degrees from the cpu to the closest copper is very obtainable. I got much lower differences going from uni-directional K-1100 carbon fibers to indium to aluminum over a complicated geometry. Going from cpu to copper water block is a much simpler interface that can more easily be conquered.

I'm not sure how accurate of numbers I can calculate for what the need flowrate of water would be, but I'll give it a whirl and see what I can scrounge up.

BTW, if anyone reading this thread is still interested after all my babble, go check out this thread if you haven't already:

http://forums.overclockers.ws/vb/showthread.php?s=&threadid=58781&perpage=30&pagenumber=1

It's funny has hell to read about pepsi and his adventures, let alone it's significance to this discussion.

[UserName]

Do you also remember when most water cooling guys got higher than air temps when they left their machines running?

[TheStapleGunKid]

Soooo... What's the concensus, should i cover just the bridges with epoxy? Thanks for all the replies guys.

[The Overclocker]

TheStapleGunKid: covering the bridges with epoxy only worked for abit, i didn't have a chance to see the epoxy dry though, if you are doinf for a remountable solution then i would try covering the bridges with silicon first or use nailvarnise and reapply it every week.

Aesik: i have a no name pump that is 800lph, it has a lift height of 1.5m. i use just over 1 m of vinal tubing in my system and it spans the whole of the antec midi tower (the dual fan becooling rad is at the top and a large tupperware container for the res at the bottem) all tubes are 3/8inch, i don't know the aculal flow rate but i think it would be very low

[TheStapleGunKid]

Overclocker: I plan on only doing this once, in that i'm going to epoxy the block right on the processor, so there won't be any going back. If i epoxy the bridges, and let them dry, that should hold up, shouldn't it?

[The Overclocker]

make sure the L1 bridges are done with something more perminent then pencil, such as silver paint. then it will work

[Ichelo351]

just thought i would throw something in here,
alittle while ago i got a high powerd fan to replace my current cpu fan. its runns at aproxmantally 21,000 rpm (yes that wasn't a typo, its really 21,000 rpm (and also if you thought a delta 7,200 rpm was loud.....)) the jet of air this thing put out was incredable, and i decided to see if i could do a direct die cool with air. so i tried it, and i was able to do it. i got into windows, and then that was good enough for me just cause i wanted to see if i could do it. i was running my duron 800 at 880. so it is possible but windows was not running nicely and was getting display glitches, i concluded that the processor die had no where near the amount of surface area needed for that type of cooling. so yeah there is my expierence with it.
-Aaron

[dream caster]

TheStapleGunKid: did you read Bong Cooling W/Direct Concact Advice! (http://forums.overclockers.ws/vb/showthread.php?s=&threadid=58781)some topics down in this same forum? It has many pertinent points for you, the most important would be to use marine epoxy to cover briges (it says that common epoxy not labeled as marine epoxy will omnly last some months ... and also that marine epoxy is too difficult to buy in less than 50gallons amount :( )

Aesik: some months ago i did calculations using the rated heat conductivity of Artic Silver II and the adviced thickness for well laped heatsinks and got a delta of about 8°C; took the data from their web page.

about the possibility of not using heat compounds: see N8's web page (http://www.wsu.edu:8080/~i6735189/cpu/news.html). It is posible, he works with very special lapping machines, I do not think is posible with hand lapping.

PS the topic here in this forum is the same Aesik linked!

[UserName]

so if DD can get better than 5C it's better than WB's

does anyone know why you have to seal your CPU in DD cooling. I have never done it and not had a problem, but then i'm blessed and highly favored.

:p

Silicone;

I'd just put sillicone sealent on it. no way you gonna get it off

[RandyM]

Aaron, you are a brave man. Were you able to monitor temps while conducting your super "air turbine" cooler test on that bare Duron die? Guess you would've mentioned it if so.

It sounds like a nozzle might actually work. If the compressed air came from a source tank already roughly at room temperature, the air will from the nozzle will be a little cooler than ambient, adding a little more benefit. Unfortunately, unless your PC is near an auto shop or factory, this mode of cooling probably isn't practical!

[JoeCrappa]

What kinda SunRoc water cooler are you using TheStapleGunKid? I might look into purchasing one of those. looks like an easy and somewhat cheap method for cooling.

[TheStapleGunKid]

Joe: It's some old Brown monster...from like the early 80's, my friend who works for a water distributing company, found it for me, in the company warehouse. Any water cooler would do, i just have the thermostat cranked as cold as it'll go, and it'll form ice in the reservoir. I'm surprised more people havn't tried this, as it is extremely easy to do. The only prob would be if ya had to go out and buy a new one...as i think they go for around 200$ Canadian. Anyway, it's works great, and it's nice and quiet.

Just so this is clear, it's one of the water coolers you'd find in an office...or whatever, which you put the blue bottles on top.

[pudgy-duck]

Originally posted by TheStapleGunKid
Overclocker: I plan on only doing this once, in that i'm going to epoxy the block right on the processor, so there won't be any going back. If i epoxy the bridges, and let them dry, that should hold up, shouldn't it?

Epoxying the block right on the processor isn't necessarily a permanent thing. I have built two Direct Die water blocks using the same T-bird CPU. All you have to do is cut the top off of the water block, heat up the epoxy with a 1500 Watt hair dryer/blower and the Epoxy will soften up. You can then peel the epoxy right off the cpu. Once the epoxy is off the cpu, heat the sides and the block will pop right off the cpu. Which is one thing that really bothered me. The water block's job is to absorb heat. It was disappointing how heat affected consumer grade epoxy. Pump failure would be bad no matter what. But with a regular copper water block, the mass of the metal would provide somewhat of a thermal reserve. In direct die cooling, with a PVC water jacket, there would be little or no thermal reserve. In the case of pump failure, I think you would get instantaneous heat higher than the stated operating range of 120 F for consumer grade epoxy.

I recently discontinued Direct Die cooling, and salvaged the T-bird from the water block. Other than the fact that the four "pads" are missing, you'd never know it had been encapsulated twice.

I am just a dumb communications technician, and I haven't been educated in thermal dynamics, so all I can tell you is what I experienced. When I encapsulated the cpu, I bonded a thermistor in contact with the side of the cpu die. Not as good as an on die diode, but about the best you can do with a T-bird. In my opinion, the most telling statistic about how good ANY water based cooling system is working, is the delta T between the CPU temp and the temp of the water incoming into the water block. For my purposes, I considered the temp of the water in the resevoir to be the same as the temp entering the water jacket. The best I could get with Direct Die cooling was a delta T between cpu temp and incoming water temp of 16 F to 20 F, idle and load respectively. I used an evaporative bong for water cooling. Typically my temps would be as follows. ambient room temp of 69 to 70 F, water temp in the resevoir 59 to 60 F (10 F below ambient), and cpu temp of 77 to 78 F. So, my cpu was running only 7 to 8 degrees above ambient, but my cooling water HAD to be 10 F BELOW ambient to get this. In my installed system, the flow rate was a Measured 120 US Gal/hour. Maybe increasing the flow rate would have helped. In the first water block, I expoxied a Lexan window in the side so I could observe the turbulance of the water. I would pull the pump up so it would suck air for a second, and observe the bubbles in the flow. Plenty of turbulance. The surface area of a Duron is smaller than a T-bird, and therefore should be harder to cool. I don't know how theoverclocker gets such good results, and he uses a radiator which should never have water temps below ambient.
I saw a picture of his water block, and its dimensions were not very different from my water block. Maybe we should all copy his setup, he might have stumbled on the perfect setup. But I have never seen him state how he measures his 24 C.

Aesik made the statement:

I have yet to see any direct die cooling that made any significant difference from water cooling. The main problem being is that you drastically reduce the effective suface area over which heat transfer can occur.

In another thread, I earlier made about the same statement after my experience with Direct Die Cooling. The heat transfer coefficient with DD is about the best you can get, BUT on a very small surface area. And I think the very small suface area limits the effectiveness of Direct Die cooling. theoverclocker himself made this statement:

i have used a becooling jagged egde waterblock on my same system with the same parts and only achived 26 degrees celcus load, this waterblock is equial to a maze2 and prefoms very well, but is no match against direct die.

He now has 24 C with his direct die cooling. In my opion, 2 C is not worth the risks that I think exist in long term exposure of both epoxy and the die surface to water. Again, I have no formal education in thermal dynamics or hydraulics, and obviously my results were not as good as theoverclockers. I can only state what my experience was. To sum it up, I tried it and I am no longer using Direct Die cooling. The gains are not worth the risk. The risk, in my opinion, is the bonding agents. I have seen people trying to bond PVC to Ceramics (water block to the cpu),vinyl tubing to PVC, PVC to copper tubing, etc. Throw the universal solvent (H2O) into the mix and in the long term (over months) I think this is risky. Even riskier if you throw in antifreeze or water wetter for a radiator, or biocides for a bong.
Again, I'm not a ceramics expert, but to my knowledge ceramics are poreous. This is hard to grasp for me, but no harder than the fact that window glass is actually classified as a liquid. Feels like a solid to me, but still classified as a very viscous liquid. So, I'm not sure about the long term affects of submersing a poreous substance (the CPU) in H2O.

I am not trying to discourage any one from experimenting with Direct Die cooling. If the Wright Bros, Thomas Edison, and others had not persevered, look were we would be now. But Pioneers usually pay a price. I decided I was not willing to pay that price. After others have perfected this, maybe I will try it again.

IMHO
Pudge