Topic: Need advice. Ready to deploy capital into off-grid 300kw BTC mine. Thoughts? (Read 156 times)

I think you are not reading the study you qouted correctly which leads you into making a very inaccurate assumtion of getting a 10% more effiency.

The chart in that study starts at 20c chip temp, that obviouslly is not even remotely sustainable and not a single miner would run at that chip temp, with immersion cooling you are looking at an average 10c difference assuming you are comparing average air cooling vs average immersion cooling, on most miners that is hardly anything, now if you compare a good air cooled miner with fans spining at 100% you won't get too far from the 10c mark.

No way on planet earth you getting 1% for every 1.5c, that is way too much, my miners do not consume 10% more power when it gets 15c hotter outside, on whatsminers we are talking 3400w vs 3500w at best, and that's 40c summer with 100% fans vs 10c winter with 30% fans, that is not even 3%.

With immersion cooling like the C6, you got a 1000w dry cooler running to get your average 10c difference, this almost offsets the fan removal gain

So a 10% more efficiency with better cooling is too far fetched, you need to set up a very terrible air cooling environment vs a super expensive 2 phase immersion cooling + pick a miner that sensitive to heat to get to that 10%.

Keep in mind that even if you magically managed to lower the average 60c temp you get from air cooling down to 30c, your miner won't hash, it would increase the voltage and force itself to consume more power to bring the chip temp to a targeted temp very close to air cooled miners, which is why in any mediocre air cooled farm you don't see fans spining at 100% simply because the miner is happy with the current temps and does not want to bring them down.

If the study you qouted hold any meaningful value, Bitmain would sell the S19j pro as a 24w/th miner but they sell it as a 30w/th gear which if you follow the chart falls under the 60c chip temp which the miner was designed to run at.

The point of immersion cooling is to overclock without exceeding chip temp limit, it is not magic, it simply transfers heat faster than air, if you could push infinite air through the miner the chip temp would be exactly the same as the ambient temp, but because you can't, you use a better heat transfer.

Another pro for immersion is noise control, you get one large fan with large blades spining at a very low RPM vs dozen smaller fans at high RPM pushing against narrow spaces of heatsinks, it also keeps your gears clean so less man power for cleaning, less filters and all, but it does not really boost efficiency, not by a considerable amount that makes it the "best option".

Both true and not true.
The issue is not resistance of copper: It is how the properties of semiconductors inside of the chips change vs temperature and how the circuits are designed to accommodate that.

In this case, the ASIC chips have been designed to run over a very narrow temp range and the chips generally want to be over at least 30C as below that they become unstable - that is why all miners have a pre-heat mode when they are 1st started up. The pre-heat uses a fairly high Vc (chip core voltage) and a slow speed to get them up to operation temp. In the case of air cooled miners, the fans run at min speed. Once at operating temp the miner changes to a lower Vc and higher speed for the chips (and fans).

Based on most mfgr literature it seems the sweet spot is 40-50C. Above that the chips consume more power at any given speed because the semiconductor junctions become more and more lossy - the power used can increase over 25% before you exceed a maximum junction temp of around 120-130C and they self-destruct

That said, with liquid cooling it is a piece of cake to keep that 'sweet spot' temp: Just throttle the fans or heat exchanger on the hot side of the cooling loop. Be aware that when dumping the heat into a river, pond, etc as a sink you MUST NOT throttle the flow through the miners - you use a liquid-to-liquid heat exchanger with the miner side fluid always running at full flow and throttle flow from the sink to control system temperature.

Because it is much easier to cool the chips, more power and higher speed can be applied to them while keeping the chip temps in the sweet spot - that is where the power efficiency increase comes from but - another factor is that the fans used on air cooled each pull 50w or more. With no fans that equates to at least a 200w savings per miner and more if the PSU can also be liquid cooled (no fans as well.

I disagree here. As electrical resistance of copper increases with temperature we know increased efficiency due to reduced temps is true for ALL miners in general. You may have some variation regarding the propotion +1% Efficiency every 1,5°C, but the qualitative statement should still be true. You are right regarding the "heating mode" at low temperatures (can you please link the discussion with thierry4wd ?), but as we go for overclocking that should be irrelevant as long as ambient air temperature is above -20°C. Of course efficiency bonus depends on ambient conditions, but I think +10% efficiency is a fair guess.


You are absolutly right within the same cooling solution, as Hashpower increase = more waste heat = highter temps = lower efficiency.
But i am not sure if Efficiency in truth is related to the temps instead of hashpower, making comparisons between different cooling solutions more difficult.
My guess: overclocking @120% Hashrate might take 125% power consumption with air cooling but only 115% power consumption with immersion cooling due to lower temps. +15% power consumption should be within operating range of PSU. Regarding Antminers the immersion version might be a cheaper alternative (cf. S21 XP vs. S21 XP Imm.)

To go on with simplified assumptions to your calculation that would mean that instead of buying 8.4 whatsminer you could buy 7 whatsminer + foghashing C6. In the first generation it is 1.4 whatsminers (=$3.738) vs. foghashing C6 (=$4.500) or in total ~3,5% highter CAPEX / 2 weeks longer till ROI. The Difference is not that much and you have to keep in mind, that due to more scope regarding efficiency you can exceed lifespan of those immersion miners a little longer. In case of a 2nd generation of miners in 3-4 years immersion finally will be more profitable. But i can comprehend your "hit & run" approach.

Not always true, efficiency "can" increase with better temps for some miners, whereby it's the exact opposite in other miners, we had a good discussion with thierry4wd "a fellow bitcointalk member who is an expert in the firmware part of mining" regarding this, for some models, when the temp is cold the miner would increase the voltage to bring the temp to the desired chip temp and thus increases power consumption, so the study posted from Braiins is probably only true for those miners they listed, so to be fair we need to exclude this point and work with the things we know are true for ALL miners.


Hashpower increased = efficiency decreased, this is true for every electronic piece on the planet Earth.


Now obviously, I can cherry-pick numbers to prove my point, but I will try to be as less biased as I can and I would work the CHEAPEST immersion cooling quotes I have, let's take Foghashing C6, it costs roughly $4500 including 200L of coolant (the coolant alone is $1300), keep in mind that if you were to use known brands that follow U.S/EU standards the cost can easily jump 2-3x but let's work with this brand.

The C6 can hold 6 Antminers or 7 Whatsminers , it can hold 8 Whatsminers but they have to be underclocked given that it has a max cooling capacity of 25kw while each Whatsminer does 3.4kw (let's use Whatsminers to make things even better) so you are looking at $4500/7  = $640 per miner, in my previous example I used M60 which costs $2,670, it now becomes $3,310, that's 24% increment.

Now it depends on what power rate you want to use, the M60 would make 9.2$ a day, at 5 cents kWh it would net $5.4 so a air cooled ROI is 2,670/5.4 = 494 days, vs the immersion cooling ROI which is 3,310/5.4 = 612 days, that's 118 days more to ROI.

Obviously, this is just a rough estimate, you are going to get worse numbers when you factor in the shipping cost for the tank, dry cooler, and liquid, it won't be a 1:1 ratio, you will easily increase those 118 days to over 150-200 just by getting a quote on the shipping and handling of a tank that size.

Things that would make the numbers go down would be your power cost, customs and tariffs, shipping cost, and your initial capital, if you have enough money to buy Foghashing BC40 Pro which can hold nearly 400 miners, the cost per miner would go down by A LOT, if you are too small and can only afford their C1 version the cost per miner would be almost 100% the miner cost, so there are many factors to consider, in my above example I tried to be biased towards immersion cooling, I used the cheapest plug-n-play cooling system I know of, I did not take pump lifespan, pipes maintenance, tank leaks into consideration, I didn't use 2k-3k BitCool coolant.

There is also another factor to consider, your hashrate would be cheaper but the cost per miner would be more expensive, you can't have 7 Whatsminers in the C6 and overclock them, but you can potentially buy only 6 and get the same hashrate of 7 miners, so you save $2,670 per tank, obviously the efficiency would drop so your power rate would be the decider here.

You see, it's a dozen numbers involved, it's going to be different for everyone, but ROI will be longer for everyone, and since my general belief regarding mining is that it's more of a hit-and-run business, I prefer to keep my cost as low possible so I can jump to the safe side as soon as I can, other people may treat it differently and they may find that an extra 20-30% more money is worth it.




 

I would be excited to do so @mikeywith... let us assume that we "upgrade" air cooled miners to immersion cooling by immersing hashboards into cooling liquid....

My Assumptions are:
- CAPEX is increased due to oil pan and cooling liquid (dielectric oil), additional oil cooler, pump, lines, etc.  
- Efficiency is increased +10% due to reduced chip temperatures
- Hashpower is increased +20% without the need of new PSU, due to increased efficiency + increased cooling
- OPEX for Fans is reduced -90% if you have some sort of natural water heat sink like river, lake, etc. (-100% fan +10% pump) otherwise it is +/- 0% (you still need fans but can choose larger ones -10% + have liquid pumps +10%)
- to reduce capital allowance over time while maximizing remaining lifetime let us stay with current miner generation (S21s, M60s, etc.)

Conclusion:
- If you are able to realize immersion cooling < +20% CAPEX you are more profitable from the beginning + you have benefits for the future as cooling solution is much more suitable to be reused compared to "old" asic miners

---

@mindrust: i guess that is more true for some kind of fluctuating energy source like renewable energies. With fossile fuels like oil i guess you can care for continuously power supply and therefore minimize the need for batteries as a buffer.

Sounds like another failed business.

Offgrid means you’ll be running miners on batteries. I am yet to see a successful farm that operates on batteries and making profits.

Have you done the math?

I’ve done it many times before and everytime I did it, the ROI was always in negative meaning the business was losing money.

Maybe there is something I don’t know or maybe I suck at math. I’d like to see you prove me wrong.

Overclocking means more CAPEX, you buy a normal air cooled miner, point it to the window, it starts hashing, you want to overclock it you are going to invest more in cooling, be it water or immersion.

It is true that in the long the run you will be able to buy more hashrate for a lower price, save on space and noise but to me that is a terrible idea.

Mining is a very risky business, it is not a very long term kind of project where you can keep feeding it for years and expect the results to come in the future, it is not a "know-how" based business that needs much experience, all it takes is one rich firm that would invest a few billions in mining with zero experience and put you out of business, so every day of ROI counts.

The one year estimatation is probably inaccurate if you were to count everything, it is probably more, and while you may think investing in cooling is more of CAPEX, it kind of hold no value when you have to sell it, so it is more of OPEX, nobody is going to buy your used liquid and pumps for any good price.

If you want to run the numbers, we can, but tell me which type of cooling you want to plot.

@mikeywith: can you share your assumptions/experiences and explain your thumb calculations regarding overclocking?

To my mind it should be possible to reduce time to ROI. I guess you might have immersion cooling solutions in mind, but even than you have to notice much longer live span of cooling solutions. The +1y might than be true for the 1st generation of miners but for the 2nd or 3rd one using the same cooling solution ROI again would be shortened!?
Additionally from what i've read here, there should be roughly about 1% more Efficiency every 1,5°C less chip temperature, that should pay in to ROI too.   

I would not buy used gear, no matter how cheap it is. It’s cheap for a reason. The last solid gear that could run trouble-free for 4–5 years was from the S9 era. Newer models have a much shorter lifespan and are very risky to buy second-hand.  

Given that your power rate is good, your main focus should be the lifespan of the gear. You need a "set-and-forget" kind of setup, unlike those with expensive power rates who have to chase efficiency every few months by selling older gear and replacing it with newer models.  

I would also advise against overclocking because it requires a massive investment in infrastructure, which could easily push your ROI back by an extra year.  

I would buy air-cooled Whatsminers, five spare PSUs, and 20 fans, knowing that I would be very unlikely to spend anything on maintenance for 2–3 years. MicroBT makes rock-solid gear. I have managed and owned thousands of mining units in a country with a hot climate, operating in farms with only average cooling. I still have 4-year-old Whatsminers that have only rebooted a couple of times due to power outages—a very low failure rate compared to every other brand on the market.  

Get M60s. They cost about $15/TH, consume about 19W/TH, so that would be 178TH × $15 = $2,670 plus shipping. The last quote I got for DDP shipping per unit was $550, so the total cost would be just under $3,300 per unit, shipped to your door.  

With 350 kW, you should reserve 10% as a buffer for cooling, leaving you with 315 kW of usable power, which supports roughly 100 M60s and gives you a solid 17 PH.  

You don’t have to go full blast immediately. If you don’t have the capital, maybe start with 20 gears and then scale up by reinvesting your mining income every two weeks.  

To be honest, I didn't really get involved with mining until last year because it's regularly uneconomical due to the high energy prices in Germany.

But I have an engineering background with professional experience in fluid dynamics and cooling solutions. As far as immersion cooling is concerned, there is definitely still a lot of potential, especially in data centers with ever-increasing power density, the trend is clearly moving in this direction.

The question that drove me was whether high-end cooling solutions can increase the performance of new miners so much (or reduce the acquisition costs por TH) in order to operate economical mining in Germany with renewable energies and battery storage. I therefore started to gain a foothold in mining about six months ago and started a few test series.

My current setting is an S19j Pro + Luxor firmware for the air cooling and efficiency tests and two old S9s for tests of immersion cooling with water (still in the planning stage). The following threads may be of interest to you (let the browser translate them):
https://bitcointalksearch.org/topic/m.64374433
https://bitcointalksearch.org/topic/m.64607783

If you want me to do some special experiments, want to support my experiments or are looking for professional advice on cooling solutions for your project, feel free to write me a PM.
 

Translated with DeepL.com (free version)

Ok, very cool, thanks for the response!

I have not been factoring overclocking into my calculations, I will certainly do that moving forward.

Do you recommend a particular firmware or mining pool management software for this? I've been informed that the new Bitmain machines can no longer be flashed with 3rd party firmware, and the entire control board needs to be replaced?

I haven't had the opportunity to play with any immersion cooling rigs, but after seeing this https://shop.canaan.io/products/avalon-miner-a1366i-119t-immersion-cooling?VariantsId=10186 , I am interested!

Were those sites I listed the official company shops for those manufacturers? There are a lot fake sites!

I appreciate your insights, and I will post follow ups as I get more information pertaining to the set up and costs of the oil site side of things.

Thanks!

Congratulations on your successes, you've definitely done a lot of things right!

---

Low power costs are not only an argument for older / less efficient miners, but also for overclocking.
This requires good heat transfer, whereby you can basically say:
Air cooling < water cooling < immersion cooling < 2-phase immersion cooling

Water cooling should be easier to implement at the beginning, mainly because water is cheap and easy to handle as a coolant. But the increase in performance is easy to estimate, which is why this is usually already priced into the mining hardware or the upgrade water cooling plates. in addition, the plates are miner-specific and probably cannot be reused in the next generation.

Immersion cooling, on the other hand, is more of a long-term investment, as you can simply replace mining hardware and reuse the cooling solution. As miner prices are linked to the hashrate delivered and this is heavily dependent on cooling, it may be attractive to buy air-cooled miners and convert to immersion cooling, as the price per asic-chip and therefore TH-potential might be lower.
On the other hand, immersion cooling requires some kind of dielectric oil as a coolant, which is much more expensive. Therefore a setup with two circuits may then make sense: the miners in an immersion cooling tank together with a heat exchanger water-cooled to cool the oil. This means more complexity and therefore higher initial costs. 

Translated with DeepL.com (free version)

Ok, so I've been mining BTC since 2015, and done well by having a HODL philosophy, coupled with a trading strategy focused on selling hype and buying panic.

I started with S9's and R4's from Bitmain that I purchased with BTC, directly from Bitmain.com and it was always a good experience.

When I was ready to re-up my investment around 2019-2020, I didn't go with the S-series from Bitmain (Everyone at the BTC conferences I went to said the S17 was not super reliable) and so instead I mined with the Avalon A1246 Series from Canaan. While not the most efficient, those things were flippin' tanks and very robust- I am a fan.

I didn't reinvest in hardware during the crypto winter of 2021-22 because hardware prices were affected by the "irrational exuberance" of the previous market surge.
Around late 2022 early 2023 my hookup on grandfathered electricity rates for the commercial property I own expired, and they jacked me up to $0.17 or higher during "peak hours".

I shut my 2 PH/s setup down, dumped all my rigs on eBay, and put the money into BTC and SOL. GekkoScience sticks are rad, (SideHack is the man!) and I am in the trenches solo mining with the degens 24/7, but I am itching to be back in the game with higher stakes!

All last year I worked on finding a new setup, and I am almost ready to deploy a respectable amount of CAPEX on an off-grid setup out in the oil fields in my geographic region.

It has been an endeavor, as the oil business is rather predatory and the potential to get taken advantage of is high- but the numbers so far look solid, and I am ready to commit capital in the next 60-90 days when I finalize a contract with the mineral rights holder.

The electricity cost is very low, so the business model is built around high volume of older/less efficient machines - I am not opposed to buying used, well-maintained rigs if anyone knows a reputable vendor. A lot of the rigs on eBay are used and abused to the point that I probably would not buy them.



Which brings me to the main point- what are your opinions on these vendors?

I was hoping to work with UpStream Data, as Steve Barbour is rad, but I am boot strapping this and self funding, so I am looking at powermining for the container because it is a bit more cost effective, and I'll be sourcing the genset locally.

https://www.powermining.io

I've worked with https://altairtech.io and they are awesome-

Also, from what I can find, these are the official sites for the main ASIC manufactures, can anyone confirm the MicroBT and Canaan sites?

Bitmain (Antminer) –      https://www.bitmain.com/
Canaan (AvalonMiner) –  https://canaan.io/ 
Is this legit?                   https://shop.canaan.io 
MicroBT (WhatsMiner) –  https://www.microbt.com/

 Revolve Labs                https://www.revolvelabs.ai                      These guys are ~ 5 hour drive from my setup

Any thoughts on the rack mounted MicroBt stuff, or immersion cooling in general? I've never done either, so I'd appreciate insight.
The way things are looking, I'll have space for ~160 3500w rigs, with a total capacity of 350kw, but I can expand to 500kw if needed.

What do you think?