Novec 7000 Project [immersive evaporating cooling]

[Cablez]

I am not sure how I missed this thread of awesome but kudos to you ATG.  This is an interesting project so keep us updated.

[1714097811]

1714097811

[DeathAndTaxes]

For your gas to water condenser look into brazed steel plate heat exchangers. I've used the ones sold by dudadiesel as heat exchangers in my cascades/autocascades and other refrigeration systems with no ill effects. Just be careful if you are condensing liquid such that you don't end up with a liquid trap. This is a bigger issue with oil for compressors in refrigeration systems though and luckily I used oil carrying refrigerants to minimize this problem.

I think there may be some confusion.  I looked up the consdenders made by dudadiesel and they are liquid to liquid.



In open batch immersion cooling the primary working liquid (Novec or Flourinert) boils and the gas rises.  It isn't piped anywhere it just rises until it contacts the "cold" (cooler than boiling  point) condenser which causes it to condense, and "rain" back into the batch as a liquid.   I guess I am missing something or you misunderstood.

[Gomeler]

For your gas to water condenser look into brazed steel plate heat exchangers. I've used the ones sold by dudadiesel as heat exchangers in my cascades/autocascades and other refrigeration systems with no ill effects. Just be careful if you are condensing liquid such that you don't end up with a liquid trap. This is a bigger issue with oil for compressors in refrigeration systems though and luckily I used oil carrying refrigerants to minimize this problem.

I think there may be some confusion.  I looked up the consdenders made by dudadiesel and they are liquid to liquid.



In open batch immersion cooling the primary working liquid (Novec or Flourinert) boils and the gas rises.  It isn't piped anywhere it just rises until it contacts the "cold" (cooler than boiling  point) condenser which causes it to condense, and "rain" back into the batch as a liquid.   I guess I am missing something or you misunderstood.

Ah, for that a large coil would be sufficient and the flourinert will just drip off the coil's fins. My mistake, I was thinking of something else.

[DeathAndTaxes]

Any updates?  My immersion fluid should be arriving soonish.  Interested to compare notes.  I don't have any FPGA and am waiting on ASICs so while I could use a GPU as a test load instead I am going to use a bank of these ...



At 12V they will pull 8.3A and dissipate 100W.  By varying the number that are powered it will allow me to run some tests on a non-hashing prototype.

[Cablez]

Cool let us know how it goes..........with pictures.

[bearsworth]

I've been thinking of making something similar. I wish Novec 700 was much cheaper, but its out of the budget for most. I would go with mineral oil. the only downfall is that it takes a while to clean it off, i think the green revolution uses it. Also what about a cooling system where the oil goes through a heatsink cooled by water from an outside maybe portable pool?

[DeathAndTaxes]

I've been thinking of making something similar. I wish Novec 700 was much cheaper, but its out of the budget for most. I would go with mineral oil. the only downfall is that it takes a while to clean it off, i think the green revolution uses it. Also what about a cooling system where the oil goes through a heatsink cooled by water from an outside maybe portable pool?

Mineral oil is an option however I will be working with either Novec 7000 or Novec 7100.  One advantage that ASICs potentially have is extreme energy densities.  Compared to a run of the mill server or GPU rig you have a higher energy usage relative to the size of the system. 3M has provided me with some data on experiment where they cooled 4KW (that would be >8 TH/s using 28nm chip @ 0.5 J/GH) using just 1 Liter of Novec 7000.  That level of energy density is probably not possible (at least not without custom designed boards) but it shows the fluid is capable of incredibly energy transfers.  The challenge is finding a solution to minimize the amount of working fluid required relative to the power consumption and hashrate.

Still mineral oil is an interesting idea however while the working fluid is cheaper everything else has higher cost and complexity.  I already watercooled a server rack of GPU rigs so I need to bump up the extremeness to have any fun.

[DeathAndTaxes]

Cool let us know how it goes..........with pictures.

Will do.  I had a dream last night the UPS guy dropped the bottles and all that expensive Novec ended up on the ground.    I probably will start a new thread but aTG first experiment gave me the inspiration.  I am just wondering if he has made any more progress.  From where he stopped I see two solutions, the first being a larger radiator and the second using Novec 7100.  Still I would prefer to see Novec 6000 working as it will result in lower board temps.

[polarhei]

So its $1k to cool this container. Sinks and fans will be cheaper. What is the business model for the fluid ?

The business model is close to the LED lamp, or the glass-fiber bus. (In Chinese, "用者自負")

Before this, I have checked different things including using vegetable oil to control the total cost. However the drawback is messy cleaning but the fluid, includes the hidden cost so if using this, then you no longer need to do messy thing repeatedly, the fluid is always ready for setting up without problem.

I now am interested with chemical as the business becomes more mature than before. Also if not for mining, I can use this for professional drawing without problem.

[DeathAndTaxes]

So its $1k to cool this container. Sinks and fans will be cheaper. What is the business model for the fluid ?

The goal isn't to fill the container and for it to be economical you are going to want to stack the head loads very tightly.

Most engineered fluids are ~$50 per Liter.  1L = 1000 cm3.  So we are talking $0.05 per cm3.

Now imagine you have a KNC board you wish to cool.  This isn't a recommendation however it is the type of board that would be a good fit because it has a high energy density.
The board is roughly 15cm x 15cm.  Now put the board on its side and put a 1 cm spacer between it and the next board.  Do that for 10 boards.   You end up with a cube roughly 15cmx15cmx15cm.  The fluid would fill the space between the boards a rectangle roughly 1cm x 15 cm x 15cm.  So there will be 225cm3 of fluid between the boards.  It is actually less because the surface mount component will displace some of the fluid.  225 cm * $0.05 = $11.25 per board.  Now that is just the cooling fluid you will need condenser and if the condenser is water cooled, pumps, tubing, and either a chiller or dry tower.

However this simplistic example should show how while the cost per gallon is high the goal is to use as little as possible.  It also allows incredible energy densities.  In the example above 10 boards (1,000 GH/s) occupy a space of 3375 cm3 or 3.4 cm3 per GH/s.  Now you still need space for the vapor height and the condenser so lets triple that to roughly 10cm3 per GH/s.

You could never air cool KNC boards with only 1cm on clearance between boards.  KNC stock design is for 4 boards (400 GH/s) in a 4U rackmount space.  So roughly 100 GH/s per 1U.  A datacenter rackunit is 19" x 24" x 1.75" = 798 cubic inches. To use the same units that is ~13,000 cm3 or 130 cm3.

10 cm3 vs 130 cm3.  A 13x improvement in density.  While you wouldn't need to rackmount these you could fit about 50 TH/s of KNC gear (90KW) in the same space as a datacenter rack.  If Cointerra or HF can deliver 4x the efficiency it would be more like 200 TH/s per datacenter rack. 

[fran2k]

Very interesting info. Thanks

[BetterWebSys]

Hi

No need for air cooling, no need for a water cooled cpu block, if you are serious about increasing your mining density then immersive cooling solutions are for you.

Miners currently use a 4U foot print with Air Cooling or GPU block cooling solutions. When space is a limitation or you just want to increase your hash per square metre then our solution is your ideal choice.

We are developing a gravity fed two phase cooling system with Novec used in an aluminium extruded immersion tank.
The cooling system is modular, hot swappable with built in safety and redundancy with minimum maintenance required.

By using Novec as the coolant we can place the mining boards about 1mm apart, increasing mining board density substantially.

Useful hot water out of the system with a temperature of around 50C.

The only moving mechanical parts are the water pumps in the base sucking the water out of the system.

The idea is to maximise the mining board density.

Imagine how many KnCMiners you could fit into this type of system.

GPU's can also be catered for.

We are based in the UK, I have been a lurker for some time on this forum and started mining around May 2013.

Interested parties oem's please pm me.
Investors please use seedco.in, pm me for details.

[Meccup]

Yeah that is another option for working fluid.  IIRC they used Novec 649 (boiling point of 50C).

I've been researching this product and is a different family of 7000, this is a acetone that can break down the coating of the wires, we are studying other solution is to use the Novec 7100 has a voiling point to 61º and mix it with the 7000 to reduce it, this tells 3M is possible.

Who told you that Novec 649 is acetone? It is Fluoroketone, tested to be inert enough for such purpose.
Also I have official 3M response that Novec 649 is recomended fluid for immersive cooling.
You can check here
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=4509386&url=http%3A%2F%2Fieeexplore.ieee.org%2Fiel5%2F4476516%2F4509347%2F04509386.pdf%3Farnumber%3D4509386
http://solutions.3mdeutschland.de/wps/portal/3M/de_DE/Novec/Home/Product_Information/Products/?PC_Z7_RJH9U5230OLLB0I4HB8HRT3OC0000000_nid=F55Z1XTKWXbeQQBXSJ1LVVgl

[BetWithCoins]

250$\gallon omg...

Im getting quotes of $1100/gallon

[Cheshyr]

250$\gallon omg...

Im getting quotes of $1100/gallon

3M sells direct in the US for about $250-$300 USD per gallon.  Mine was delivered in less than a week from order date.  Either someone is trying to rip you off, you pissed someone off, or you're international and there are other things going on.

[zulover]

Has any one put any thing iin to a product or solution regarding this?

[antirack]

Has any one put any thing iin to a product or solution regarding this?

Like this?

Here are the preliminary recommended board dimensions for immersion cooling. 

Anything goes with immersion, final decision up to the board designers discretion.

https://docs.google.com/a/allied-control.com/file/d/0ByWHHc0u_thNdzB3c2hvVzJkcTQ/edit

This is for the DataTank designs, which will be sitting in various data centers and ASIC hosting providers in the future.

http://www.allied-control.com/datatank

[Bicknellski]

Building a board to fit those.

[antirack]

You don't have to.  You could put a Cointerra 4U Terraminer in there (it's 19-inch).

Or alternatively just fill the complete tank with their own structure/idea, the inside dimensions (19-inch) are 450mm x 2200mm.

That's just a recommendation by the designers so that the miner doesn't need to always redo everything once the next hardware is around the corner. And to up the density.

Whatever fits in a bathtub will fit in one of these. They are 19-inch bathtubs. Why should a few dozen mining device manufacturers always need to work on their own boxes, cooling, infrastructure when all he market needs is delivering boards on chips? That's why CPUs and DDR memory comes in a standardized form factor. Makes sense?

[Bicknellski]

Ya we have to and we are hoping to get it tested in HK at Allied Control.

Our group is not going to rely on Cointerra. LOL. We got some other plans for the variety of chips we already have in or will have in hand soon.

Hammer Whiteface
AM BE200 Whiteface
Minion Whiteface
"R" - Whiteface

Very interesting times indeed.