Why so few immersion cooled custom builds/Mods?

[Shermand100]

Okay, I know at a glace this could come across as a stupid topic but I would really appreciate it if an articulate discussion came of this...

Cryptocurrency mining hardware is very expensive, and we should look after our rigs as best we can. After all, the build up of dust reduces air flow, raises temperatures, kills blades/miners.

However in my (limited) experience, the components that get killed on the boards aren't often the mining chips as you may think, but the often disregarded power regulating components. Mosfets, capacitors etc. Without this becoming a rant at some manufacturers that under-power their boards, over-stress the power supply circuitry, so it runs fast until it out lives the warranty. Some of us are doing this at a hobby level and don't have the high turnover of constantly getting the newest model of ASIC. We need to make our units last.

So my next point is regarding temperature. For one, people refer to overclocking ASICs and state the temperature that their interface is telling them the unit is at. Often this is wildly inaccurate, giving a surface temp of a PCB or once again measuring the mining chips and neglecting the Mosfets. Yes, when you overclock a mining chip it may raise in temp from 60 to 75, and you'd think with it's upper limit of 125 that'd be fine. But what's the temp of the Mosfets, capacitors etc we're stressing driving this equipment? As I said before it's not often the mining chip goes pop.

So my last point and the punchline really. The power supply components are often smaller than the mining chips in terms of surface area. They are also of unequal heights and this makes it very impractical to fit heatsinks. The designer places them in the airflow as best they can, but as I mentioned, with the enclosure around them this can be partly obstructed especially over time if not maintained. Is the answer not fluid cooled systems, immersion I mean, that give even heat exchange across every component? No matter how small?

I get that could be a logistical nightmare to ship. But I've seen very few hobbyists adapt their miners this way too. Am I missing something? Or if it ain't broken don't fix it?

But lets be honest on a technical level it would be no more complex than assembling a simple frame and dropping it in a fishtank of mineral oil. (A kind of gross simplification).
What about people that want to mine with an S7/S9 at home but can't put up with the noise?

A few thoughts.
Dan
Some interesting examples on youtube.

[1714481572]

1714481572

[1714481572]

1714481572

[1714481572]

1714481572

[1714481572]

1714481572

[-ck]

But lets be honest on a technical level it would be no more complex than assembling a simple frame and dropping it in a fishtank of mineral oil. (A kind of gross simplification).

You know the mineral oil just absorbs the heat anyway and mining hardware generates a lot of heat so you need some kind of cooling for the mineral oil itself equivalent to dissipating the same amount of heat the air cooling on the hardware does. The overall set up would be very expensive; that's the main reason it's not done.

[Shermand100]

Yes a heat exchanger of some sort would be needed too, agreed. It was a simplification and a whole other topic of how to build one.

But on that topic it reminds me of other posts I've read here. People setting up their miners in the garage due to noise/heat and opening a air inlet through the wall, and exhausting through another hole. It just screamed contamination from ingesting bugs, leaves in autumn or anything wanting to nest somewhere warm.  What about the other advantage of having the miner indoors and pumping the oil through hose to the heat exchanger at a more suitable location? I just think that as everyone's circumstances and climates are different depending on where/how they live yet we all have the same air in/out push/pull fan arrangement. Is that really perfect for the majority?


***edit*** the garage example aside, if you look on youtube for "quiet antminer" some people have gone to some extreme lengths to continue air-cooling the miner when perhaps another method could have been adopted.

[NotFuzzyWarm]

Please ref https://bitcointalk.org/index.php?topic=1604125.0 dealing with liquid/immersion cooling a s9.

Mineral oil is NOT a good thing to use:
Only moderate heat-capacity (how well it absorbs/releases heat)
Viscosity highly variable with temp
Even when warm, is still not easy to pump
Attacks many plastics including Nylon (PCIe connectors), PCV (wire insulation) and elastomer (rubber) capacitor seals.

Much better choice are thermal transfer fluids designed for use with electronics and electrical parts such as http://www.clearcoproducts.com/dielectric-silicone-fluids-sto-50.html

Water-thin meaning easy to pump and flows very well around components, does NOT attack rubber/plastic, etc., very good thermal properties.

[OgNasty]

I'd like to see more immersion cooled mining setups as well...  Having the benefit of learning from someone else's mistakes might give me the confidence to try it myself.

[NotFuzzyWarm]

I'd like to see more immersion cooled mining setups as well...  Having the benefit of learning from someone else's mistakes might give me the confidence to try it myself.

For decades I've worked with immersion cooling all the time in HVDC supplies producing 10's of kwatts and will categorically state the thermal xfr fluid is a key to it. Yes in my case has to withstand 30-50kv but more importantly has to be thermally stable, have excellent heat transfer ability and, most importantly not attack other materials. Off the shelf mineral oil attacks hoses, seals, and electronic components. Due to its viscosity it pumps/flows very poorly.

Yes having a horse trough (preferably a real metal tank) full of miners in a fluid like the one from Clearco should work great. I use it in work and is near-water thin so pumps/flows very well. So far only caution about materials is pure silicon-rubber -- makes it swell.

Since miners like a high temp, say a bath running around 50-60C, that makes for a great delta-T between the fluid and outside ambient resulting in a smaller hot-side air/fluid exchanger needed.

Yes the Clearco fluid is pricey but way way less than Novec is. Plus of course, if left in the open on a very hot day or in case of a drastic thermal event caused by no flow/pump & fans status/overtemp cutouts, several hundred $ (or several thousand $ if Novec) of it does not go away....

[Shermand100]

Ok, so there's a few more forum topics than I thought on this (I searched the forum for immersion cooling, and there was nothing recent. I didn't look for submersion   )

But I don't really want to get hung up on the how to. That seems like a common theme, oil types etc. (Unless the reason people don't do this is complexity.) Just the why "why" interests me in particular. I do this as a hobby and am very prepared to modify things to suit my needs. I have my miners close to where I sleep due to living arrangements. This makes noise my biggest concern. No point me having an S7 if I never sleep again    This has limited me to s3 units and swapping the fans for quieter one's. For me that works great.

But I trying to test the room for if people are happy the way things are with air cooling. How about when bitmain had the C1? It's not quite as drastic as immersion/submersion cooling but tackles the noise problem. Not sure it helps the power regulating components much though.

Does anyone think that as these things are sounding more and more like a jet engine with people posting videos on youtube using blenders as noise comparisons, we're using air as a cooling medium but that's on the edge of still being practical? A step in a different direction

I just feel the life of these units could be extended if this route was taken, (and done properly, but again not bothered with the how yet as in other threads).

***notfuzzywarm posted as I was writing this and it's great to see some expert insight. Can you say if the units you work with benefit from the immersion cooling due to high component count, or due to needing to transfer more heat than air allows, or other reasons?

[NotFuzzyWarm]

2 things drives it: Max kw in smallest size package and moving that heat load somewhere else to be gotten rid of. For us that is a plant cooling water supply running up to 110F, preferably under 95F. Well, that and lack of fans/filters to replace.

Ok that makes 3.
Maybe 4.
What is this? The Spanish Inquisition??! <apologies to Monty Python>

A typical 30kv supply good for 20kw is about 2ft-square-sided cube and that is mainly due to spacing needed due to the high voltage.

[NotFuzzyWarm]

Might as well resurect these old Novec Data Tank vid eh? https://vimeo.com/95819646 and https://www.youtube.com/watch?v=iykkSo75yDo
240kw in a very small volume tank in that last one... Woof.

[OgNasty]

This guy seemed to have good results:
https://www.youtube.com/watch?v=xo0g2dKPAcQ

He had interesting reasoning for why he hasn't done immersion cooling on an S9...

you cannot adjust the Heat fan speed to 0 because of this the firmware will detect that the fans are not operating at the proper RPM and it will go into automatic shutdown the fans can not move at the proper RPM when they are placed inside of mineral oil due to the friction

Seems like it would be easy enough to get around, but another interesting drawback to the auto-tune firmware Bitmain introduced.

[sidehack]

Anything that'll make a 12V pulsetrain at about 135Hz on the tach line will tell the controller it's got a fan running at 4000RPM. A $1 555 circuit will do that.

[Dibblah]

Or, a flow sensor for the cooling loop would also work. They're quite cheap on ebay these days, but watch the seals.

[sidehack]

Provided the output was compatible (or easy to transform), that'd be a better option.

[jstefanop]

I hacked together a prototype tank for some testing a while back. This is just a simple open loop design (just runs and dumps cold water through the coil). Have some more ASICs coming so maybe ill modify this with a real closed loop design.

https://www.youtube.com/watch?v=p5-BaH8_SAk

[MachineZero]

I would love to heat my pool using some kind of heat transfer between may-sep, then switch to heating my basement on other months.
It'd be nice if it was one system and interchangeable with the latest miner.

Wishful thinking...

[sidehack]

If you just had a dunk tank with a circulator inside it, you could put whatever miner you wanted in there. Pump your fluid out to a radiator in your basement or another exchanger in the pool. If that's too much pricey fluid, use a heat exchanger in the tank and pump water or something through it and out to the radiator. That's gonna raise the overall temperature a bit because you'll have two heat transfers instead of just one, so you'd have to pay a bit more attention when designing and operating to make sure it'd handle full power.

[MachineZero]

If you just had a dunk tank with a circulator inside it, you could put whatever miner you wanted in there. Pump your fluid out to a radiator in your basement or another exchanger in the pool. If that's too much pricey fluid, use a heat exchanger in the tank and pump water or something through it and out to the radiator. That's gonna raise the overall temperature a bit because you'll have two heat transfers instead of just one, so you'd have to pay a bit more attention when designing and operating to make sure it'd handle full power.

Need a parts list and some pictures.

I'm thinking something like

that encases 2 S9s,  feeds into a pool pump and with a flick of the switch exchanges heat into the house.

[NotFuzzyWarm]

If you just had a dunk tank with a circulator inside it, you could put whatever miner you wanted in there. Pump your fluid out to a radiator in your basement or another exchanger in the pool. If that's too much pricey fluid, use a heat exchanger in the tank and pump water or something through it and out to the radiator. That's gonna raise the overall temperature a bit because you'll have two heat transfers instead of just one, so you'd have to pay a bit more attention when designing and operating to make sure it'd handle full power.

Ya on the emphasis I added.
In our laser systems the total piping run for the cooling oil side is only a few feet just for that reason. Way cheaper (less thermal fluid) to xfr from the oil to water using a heat exchanger located in the systems.

My fav heat exchangers are the plate type, damn near 100% xfr efficiency. I use this one https://www.mcmaster.com/#35115k62/=1797t5f to keep 10kw of laser heat load a constant 72-75F using 60F water flowing 5GPM, water solenoid spends about 50% time on & off so could push even more heat in if wanted.

Given the miners like running MUCH hotter - remember, 70C = 158F - - gives us a very large delta-T to work with so cooling the water side just using air even on a hot day should still let you pull a helluva lot of heat out of the system.

EDIT: Hopefully it should go without saying but I must anyway: If you are going to be using liquid cooling and running miners at over about 100F (37.78C) then DO NOT use a plastic tank! Plastic of any type begins to soften/deform at surprisingly low temperatures... Fiberglass good (hmm, a stand alone laundry sink? Already has a nice hole in the bottom for incoming oil returning from the oil/water exchanger...), metal tank is best.

[QuintLeo]

You have to design a board to work with immersion cooling specifically, as a general rule, for it to work properly and reliably - and the up-front investment in the hardware AND the cooling setup isn't exactly low cost.

It can work well once it's set up, though - but it's more "large pro farms" that use it, or something like that BitFury "shipping container" setup.

[OgNasty]

What are the options for purchasing a finished system for immersion cooling?

I've only been able to find this product, and the company didn't want to talk to me about a small residential installation...

http://www.grcooling.com/carnotjet/