Energy use: computational effort vs cooling

[gkonstan1]

Hi.

This question may have been discussed at length, however after a few searches I can't find it. If someone knows a good thread please point me in the right direction.

How much energy is used for air cooling vs for conducting the computational efforts for mining? As a percentage. (ie/ chipset vs fans)

The reason I'm asking is because I expect air cooling to use a lot more energy than liquid cooling, and I would like to find out by how much. I'm imagining a large mining farm is running thousands of little fans, and then also a few very large fans in the facility. In total, the fan energy consumption is probably not negligible.

There may be a good case to use less energy by running a liquid cooled geothermal system. The energy costs for pumping the water around are very low, and you can use the ground to dissipate the thermal energy. This configuration would be very low energy use relative to air cooling, and could be done almost anywhere because the temp below ground is cooler than the surface. It might be possible to cool enough that you could overclock the mining machines. If you have a small river near by then it gets even better in every way (less capital time/cost to set up, and cooling performance is better).

Ideally, the energy spent is utilized to conduct the calculations, and minimal for cooling. I'm curious how beneficial a liquid cooling system would be. For example, if the air cooling energy is only 0.5% of the total energy than a liquid cooled solution is not worth the efforts. Although, if air cooling is 40%+ of the energy consumed then liquid cooling is most likely very beneficial.

Thanks,
George.

[philipma1957]

Okay we burn 100kwatts

and we use 19000cfm fans input with 18000cfm fans output

we cold wall separate so there in a cold aisle in and hot isle out.


1200 watts 9000 cfm.  the input fans
800 watts 10000cfm


400 watts 5000 cfm.    the output fans
1200 watts 13000 cfm


total of 3600 watts of fans and we have burned as much as 130kwatts in 95f summer temps. at this point we are a bit too warm.

so 3%

so if you are in a hot country 3600 watts of fans in [95f  or  37c ] barely works to cool 130kwatts of gear. with decent hot- cold aisle setup.

[gkonstan1]

Thanks for the info Phil.

3% is small.

That 30kW (30%) additional load in the summer is getting pretty high. Do you have any thoughts on where that comes from?  I mean, I wouldn't expect a fan to pull more power (maybe it does). However, if the chipset temp increases maybe it uses the additional power?

Also, are you running additional fans in the building or is this strictly the small fans on each miner. IMO, if you have extra fans in the building those should be included in this estimate.

If the 30% additional energy consumption is due to atmosphere temps, then a liquid cooled system may be appropriate for hot areas that have cheap power. For example, the desert where you could use solar energy and then couple that with geothermal cooling.

[philipma1957]

We ran more gear and gear loses some efficiency.

Our

btc ph was 550+1400 = 1950ph
ltc  was. 10gh

now  btc ph is 1700ph
ltc gh is 7gh

we swapped out older gear for some gpus and some gear died.

and we purchased some gpus to replace the ltc asics and btc asic they do not use as much power as the btc and the ltc that is off the system.

In the winter all those fans are not on so we may use 3000 watts right now.

We do have good hot/cold aisles.

[alh]

While it might be interesting, I am not aware of any current generation mining hardware that readily accept a change from air to liquid cooling in any reasonable way. All the current generation hardware has many dozens of chips spread across 2-3 boards. Each chip usually has a heatsink stuck to the chip. The past generations actually used glue which made it difficult if not impossible to remove the heatsink. I expect the current generation is constructed similarly, though somebody that has actually dismantled one would confirm or refute my claim.

The mining hardware of today does NOT have a small number of large chips for the actual SHA-256 calculations. It's all about parallelism, both within each chip and then across the board. I'd be stunned if Bitmain (for example) would have any interest in producing a variant of the mining board with a liquid cooling plate instead of a heatsink per chip.

I assume you were talking about some kind replacement for heatsink when you mentioned liquid coolling, or did you have something else in mind?

[gkonstan1]

If the benefit of liquid cooling is high enough, I would expect to be able to convince a mining machine OEM to design/build a liquid cooled version (ie/Bitmain or another supplier). I'm sure these companies have some customers that are large enough to warrant a completely liquid cooled design, should someone place a large enough order.

Conversely if they don't entertain it (which would surprise me) that could justify a full in-house design from one of these big mining farm companies. For the amount they spend in buying machines, I'm sure the cost would be negligible to them to create some in-house prototypes for testing. Long-term, I expect someone to design/build their own miner and use it exclusively themselves rather then sell it to others to run. There probably are some mining farms that already do this. (Do you know any mining farms that make their own miners?)

More importantly, I would expect this switch to liquid cooling to be accompanied with a full architecture change on the machine and the farm such that both would look very differently: no tubular case and the chipsets could be stacked.

chipset - LC heatsink - chipset - LC heatsink - repeat.

one larger power supply to support the larger number of chipsets

These could be made in reasonable size heights, perhaps 100mm tall for 4 chipsets, or 500mm tall for 20 chipsets.

The liquid cooled versions could be approx 10X physically smaller than their air cooled equivalent which further reduces the costs for a farm (smaller building)

[alh]

To find out more about current mining manufacturers, you might want to consider posing your questions and reading what you can in the various posts in the Mining -> Hardware  sub-forum adjacent to this one. There is significant overlap in folks, but it's not 100% and you might well find additional response.

Best of luck.

[gkonstan1]

Excellent, thanks. I didn't see that sub-board. Appreciate it.

[wndsnb]

Bitmain already announced a liquid-cooled miner, the AntRack, last year.

It was discussed in this thread: https://bitcointalk.org/index.php?topic=5287628.msg55547933#msg55547933

Also, I haven't personally seen the guts of a S19 yet, but I've been told that they no longer have individual heat sinks, but large heat sinks that spread across multiple chips. If that's the case then it could be possible to swap out the heatsinks for liquid cold plates to do a liquid-cooled conversion.

[philipma1957]

the s19 and s19 pro could fit water blocks.

whatsminer m31 m30 could fit water blocks.

both would want different cases but it could be done.

I think whatsminer m30 and m31 would be great with water blocks.

[wndsnb]

You'd probably be fine with the current case if you just remove one of the fan plates. There are some companies that specialize in making custom cold plates, I wonder what quantity you'd need to make the $ work. In hot climates, you might save 50-75W per miner.

75W would only be 657Kwh/year, so not likely to ROI on energy usage. That's only $33 a year for 5c power.

[mikeywith]

I am afraid you are a little bit too late, Bitmain already has their Antrack , Canaan also have their own, both solutions eliminate the need for air cooling altogether.

The liquid cooling will most likely consume a bit less energy than air cooling, but its initial cost could (most of the time) make it a lot less attractive, it's like solar vs grid, it's cheaper to run solar, but not everyone does so.

This whole mining business is probably going to come to an end in a few years, so unless someone is gambling with investor's money or have no clue about what they are doing - they would want to minimize their cost as much as possible, if I would ROI the liquid cooling investment in say a year or less, it would make sense, but if that takes me 5 or 10 years, I rather not,  so in other words, if I save $100 on energy per month but have to pay $10,000 for the liquid cooling infrastructure, it means I have to wait for 8 years to start receiving the rewards of my investments, do I want something like that? hell, no.

There MUST be other reasons added to the small potential energy-saving to justify the cost, those include but not limited to noise reduction and better cooling, but if I don't care about the noise and I know the chips I have will likely die "economically" faster than "physically" then even the better cooling aspect is worthless, of course, to someone who has a lot of free energy but he must keep the noise level very low, they would want to have a liquid-cooled farm.