It actually looks quite simple. Enough so that I am quite tempted to try it.

Not water cooling, but oil cooling (maybe this part of the forum should be renamed "liquid cooling"?). I know that this topic has been covered on this site before, but this is a particularly cool video in my opinion.

http://www.youtube.com/pugetsys

Hello,

This has been posted about several/many times in the past -- it is interesting, but there are certainly some hurdles to overcome (like rancid oil...). Check it out with the Search function?

Hey Neil.

I did have a search before hand, but didn't find this specific example posted on the forums.

They get around the rancid oil by using mineral oil, not vegetable oil.

The reason I like this particular example is that they show how cheaply and simply it can be done, including the specifics of the parts they used.

That's a fun video. Thanks for the post.

Of course it wouldn't be silent the way the did the HDD, but there are solutions for that.

And the bubbles! Noisy and distracting, but fun.

thanks for the link, nicely made video

i like when he points to 'mild laxative'... thumbs up !!

That's pretty sweet.

Now if only someone would mod their way to a practical 100% do it yourself laptop.

I've seen this:

http://www.mini-itx.com/projects/itx-laptop/

But there's gotta be a better way. It's time for the MTX motherboard standard to be introduced, with MTX cases. Pico PSUs would be perfect for this.

(Yes, I completely pulled the "MTX standard" out of thin air. But standardized cases and motherboards is what's needed before 100% build your own laptops will become a reality)

I don't like those barebones things. I'd really love to have a laptop case built by Lian Li!

Mineral oil eats some rubbers and plastics though. For example, the capacitors in the PSU commonly use rubber rings to seal them, and if the oil eats them, they may bloat and dislodge from the PSU. So it's not without hazard to use mineral oil.

That's interesting info Celoth.. how do you know this stuff??

Here's a detailed discussion of the system, including an update on six months of operation. Good info on temps may be of particular interest to SPCR readers.

Oil-submerged system

Well, I ordered myself a custom built glass aquarium with adequate dimensions. It was surprisingly cheap.. just AU$55

Seems the oil will be more expensive than the aquarium.. cheapest I've found is AU$50 for 20 Litres from an animal supplies wholesaler.

Please keep us updated on this as it is a subject that for me is highly interesting ! I am planning something like this myself, and i would like to try and keep things as small as possible.

So.. updates. I tested the concept with an old Pentium 2 233 in a little plastic tub with 5L of horse laxative grade mineral oil. It worked fine (booted XP from a hard disk. Yes, XP on A P2 233.. pain). I completely submerged the motherboard, and poured oil all over the CPU and PSU. Also, I experimented by removing the PCI video card, pouring oil into the PCI slot, reinserting card, and repowering.. worked fine.

Still waiting on my aquarium; when it is done, I'll pick up 40L of oil (about AU$100), set it up and post the results.

I read somewhere that mineral oil vapors are dangerous because it can accumulate in your lungs, your body can't get rid of it. Not sure about that though.

Fantastic. I'm looking forward to hearing about this too.

Please update us when you get it going, Bruce!

BTW - what sized aquarium are you getting? I'm thinking about that very issue, for myself. I think I want one just big enough for the boards.

-stefan

Instead of mineral oil, how about trying vaseline? It's the same thing as mineral oil (distilled parafin), except it has longer chains. And when the machine is not running, it'll be a solid block, easier to cary to LAN parties. Might need to run it a couple of times before it gets rid of all the bubbles, but it should work nice. Make sure to take off all the fans cause they sure wont be chugging vaseline around.



As a to be health professionnal, i concur with this one.

What a cool idea! I've hadn't heard that proposed before.

The idea of relocating a tank of mineral oil worries me (a spill would be an unholy mess), and this might solve that problem.

Also, I'm guessing that the long-chained vaseline produces fewer vapors than thin short-chained mineral oil.

Vaseline. That's worth looking into.

-stefan

There's a downside though. I think vaseline wouldn't have the cooling benefit of convection since it's a more viscous. But can't tell without trying.

And i thought about the problem of air bubbles, just melt the vaseline before pooring it in the first time around. I remember melting it to make ointments in my potion making classes, ahh, those were the times.

Vaseline sounds funny, but it would be a messy job to change a component later.

Well, I'll be danged.

Tonight I tried melting some vaseline in an old snapple bottle over a candle. I was surprised -- it looks just like mineral oil when melted -- totally transparent, not even a little cloudy.



I could see that convection was happening. Yes, it is more viscous, but it melts at only 100F and is quite fluid at 150F (yes, I stuck a thermometer in there). So, I'm sure there would be a cooling benefit once it has liquified.

However, I'm concerned that chips might overheat at startup. If vaseline (white petrolatum) does not readily conduct heat, then a hot chip will liquify the vaseline immediately around it, but that little pocket of liquidity has nowhere to go, so it just stays there being hot. Convection won't have much cooling effect within a very small liquid pocket.

The hot liquid vaseline will (at some rate unknown to me) transmit its heat to the surrounding solid (well, gel-like) vaseline and eventually liquify it. I don't know how fast that happens, but if it isn't fast enough, the chip would overheat and get damaged (or shut itself down).

If the whole tank started out as a liquid, I would not be worried -- since I have witnessed plenty of convective movement in the sample I played with earlier.

I don't know where to find thermal conductivity data on vaseline or white petrolatum. An abstract at http://lib.bioinfo.pl/auth:Fallopa,F describes ultrasound experiments using gel, water, vaseline, and mineral oil as coupling agents (for transmitting sound from transducer to subject). It said that the vaseline was the least thermally conductive, but gave no hard figures.

Bleh. None of which proves anything. I just don't know whether solid vaseline can carry away the heat energy fast enough to prevent cpu/gpu overheating before it becomes sufficiently liquid.

And stkris is right -- it would be REALLY messy to work with.

I think I'll stick with mineral oil.

Still - well done trying it out. Methinks the only way to go forward is to test and try...

In answer to somebody's earlier question about the dimensions of the aquarium I ordered:
Height:17" Depth: 10" Width: 14"
Volume: 2380 cubic inches (~=39.01L)

I determined the size by measuring the metal chassis which I will be mounting in there, and adding an inch to each dimension.

Well, i'm not going to try either of them, but if i were, i'd still try vaseline just for fun. Cause i can't see why you'd need to do this asside from it being just for fun. And maybe for the ability to freeze it into a block and cary it to lan parties.
Anyways, here's my two cents about the conductivity problem. There are 2 factors we need to consider about vaseline. On one hand like you said the melting point is low. But on the otherhand, the conductivity is supposedly bad. So if we put a heatsink on the CPU, it will lower the temperature gradient but increase the surface of contact. The lower T gradient is not to worry about since melting point is low, supposing T of the heatsink is still higher than the melting point. Increased surface of contact is a plus since it'll let us melt more vaseline at a time. Not like anyone was going to try to cool their cpu without any kind of heatsink anyways, that seems pretty extreme to me. If you want to make sure, you could always mix in some mineral oil to lower the melting point to somewhere around 25-30C (porportion unknown to me).
The wierd thing with this, if vasoline is really insulant in it's solid form, would be that your computer will be running warmer in winter than in summer!

That would indeed be ironic. But ice and snow has that property - they are surprisingly good thermal insulators (they absorb heat energy slowly).

BTW, I've been learning about thermal transfer. Ordinary single-pane glass, such as found on an aquarium, can have a thermal emissivity of 6.42 W / m^2 K. (see http://books.google.com/books?id=C-avw0 ... tBsAL8gcUc)

The heat loss through glass is proportional to its temperature above ambient and also to the area of the hot glass. The formula is

H = U * A * T

where U = 6.42 W / m^2 K, A = area in sq meters, T = temp difference in centigrade, and H = heat loss in watts (joules/sec).

I have a 5.5 gallon tank measuring 16" x 8" x 9" tall. The sides alone are 432 sq inches, which is about 0.28 m^2.

Suppose the mineral oil in the tank reaches 65 C (about 150 F), and the room air temp is 20 C. That's T = 45 C, and so H = 6.42 * 0.28 * 45 = 81 watts.

Hmm, that's not really very impressive. That's like just a CPU's heat production, isn't it?

Anyway, that is just conductive heat loss from glass to air. It doesn't allow for radiative heat loss, which I'm told is negligible in most computing applications (but maybe it matters here?)

-stefan

Yeah, glass isn't the best conductor. I think we can only compare this versus a full passive air cooling solution. Air is less conductive, and less capacitive, though through convection it exchanges with the room more.

I suppose U is specific to glass. A metal bin might have a higher U. Everyone has done it with glass or a sealed plastic case, for fun. But metal would have been better. I think transformers on electric posts are submerged in some type of mineral oil too, and the case is metal, with heatsink fins to increase A.

I just can't put this subject down. I found out about radiative cooling, at this URL: http://www.engineeringtoolbox.com/radia ... d_431.html

It turns out that radiative cooling is related to the 4th power of temperature, so the cooling effect is superlinear. The higher the temperature, the more radiation matters.

The radiation emission of a hot black body in cold surroundings is

q = s (Th^4-Tc^4) A

where q is heat loss (watts), Th is the temperature of the hot body (kelvin), Tc is the temperature of the cold surroundings (kelvin), A is the area of the interface (sq meters), and s is the stefan-boltzmann constant, s = 5.67 x 10^-8 W/m^2 K.

In my aquarium example, the oil was 65C (150 F) in a 20 C (68 F) room. We had Th = 338.15 K, Tc = 293.15 K, A = 0.28 m^2. So the radiative heat loss comes to 90.33 watts.

However, glass is not an ideal black body. An ideal black body has an emissivity of 1.0. Glass is close, with an emissivity of 0.92. So, the glass would emit only 83.1 watts.

Combine that with the 80 watts of conductive heat loss, and you have 160 watts of heat loss from the sides of the tank.

Because of the heat loss is related to the 4th power (superlinear), small increases in temperature can lead to big increases in heat loss.

At 75 C (167 F), the radiative heat loss is 106 watts.

At 85 C (185 F), the radiative heat loss is 132 watts.

One downside to oil-submerged cooling is that ALL components in your system are experiencing the same temperature, so your cooling requirements are dictated by whatever is the most sensitive component. I don't actually know what is most sensitive -- cpu, GPU, vid rams, DIMMs, NB, etc. Anyone know?

Okay, suppose you have a 5.5 gallon aquarium (16"x9"x8"), and suppose all six faces (including top and bottom) of the aquarium could be allowed to contact air and radiate freely (maybe give it legs, like a table). That's 688 sq inches = 0.444 sq meters of surface area.

Based on conductive and radiative heat loss models I've dug up on the web (see previous posts), here's the the heat loss rate in a room that's maintained at 22 C (71.6 F):


So, this tells me that if I have some big honking system that sinks 250 watts into this tank, it is going to settle at a temperature of over 85C, which seems pretty hot. A smaller system that generates only 150 W of heat will settle just below 65C.

As I alluded to last post, I don't know what temperatures are safe for the various devices (GPU, CPU, vid rams, DIMMs, NB, etc).

One thing I have not addressed is the rate at which the system approaches the final temperature. That would be dependent on the thermal mass of the oil-filled aquarium, which relates how much temperature rise results from absorbing a given amount of energy.

Mineral oil has a heat capacity of 1.67 J/gK, and specific gravity of .87 g/cc. 5 gallons of it would be 18927 cc, which is 16.466 kg, which is a thermal mass of 27.5 kJ/K.

So, a 250 watt system running for 1 hour (3600 seconds) produces 250*3600 = 900000 joules of energy, which would change the temperature of that thermal mass by only 32 C. And that assumes the tank loses zero heat during that hour! In reality, as soon as the tank gets above ambient temperature, it starts losing some of its heat. To model the temperature rise as a function of time might require a differential equations solver, or something.

The puget systems people mentioned that it took a long time for their system to reach a stable temperature.

If you've read this far, I salute you, and I apologize. :-)

Again, does anyone know the typical safe operating temperatures for the various devices (GPU, CPU, vid rams, DIMMs, NB, etc)?

-stefan

Well i drink to your health mate.
There is still that one last step you might as well make after gathering all the variables. Combine the constant heat production with the heat loss (due to conductivité with air and radiation) and derive it for time infinity and give us a temperature!
CPU, GPU, RAM and those transitors should tolerate to something high enough. Typical safe figure they give for CPU is about 90C. Just a wild guess but i'd worry more about electrolytic capacitors boiling and poping open before.

mobile cpus normally rated up to 100C but desktop cpus a lot less, e.g.

http://www.amdcompare.com/us-en/desktop ... 3800IAA5CU

ST is right, the electrolytics are probably most temp-sensitive component, most are only rated to function up to 85C and for every 10C increase in temp their useful lifespan halves, so your board will die very soon at 85C.

ST, the table in my previous post gives the temperature after infinite time. The table gives the tank's instantaneous heat loss rate as a function of instantaneous temperature. When the heat loss rate (of tank to air) matches the heat production rate (of the GPU+CPU+etc heating the tank), then the temperature stops rising. Consequently, the table also tells you the final temperature for a given heat production rate. Note - this table applies only to a 16x9x8 glass aquarium filled with mineral oil with all six faces exposed to air.

ST and jaganath - I didn't know about the fragility of electrolytic capacitors. Thanks for cluing* me in on that. Another source tells me that the electrolytics are the most common cause of electronics failure.

Even if I stay within the temperature limits, is there a danger of the mineral entering the caps via their vents and mixing/melting the material within? I don't know if these vents are hermetically sealed, or if they are actual holes. Isn't the electrolytic an actual fluid? If so, I would guess the vents would have to be sealed.

This all points to what may be the Achilles heel for oil-submerged cooling: you are subjecting your entire PCB to the heat generated by the chips, and consequently you must be mindful of the PCBs weakest link, which in this case may be the caps (w.r.t. temperature). More conventional water cooling (water blocks + tubes + pumps + radiator) are able to absorb the heat where it is produced ( the chips ) and take it out of the system before it can reach other parts of the board.

-stefan

* wow, the spell checker accepted "cluing", but not "electrolytics".

Sorry, i just thought if heat production was constant and heat loss was exponential, that the temperature would end up lower. Was confused there, lacked the imagination. Indeed, it is rather the rate that's going to be slow but the T wont be different at t infinity.
Well the only thing we'll be betting on in this case is that the rate will be slow enough that i'll just turn off the computer before it stabilizes... Which is quite a dangerous assumption.
How much does a system dissipate anyways? It can't be dissipating ALL the electricity that goes into it; it's not a heater, it's a computer that does work. What was T of the guys who did it in the video? I remember seing Tom's Hardware try this with vegetable oil and they were pretty happy with their temperatures.