No products in the cart.
EXPLORATIONS INTO SOLAR MINING
- AGENTP
Introduction
I am new to Bitcoin (class of 2021). I have some unorthodox viewpoints; I am either new and naive or I am new enough to avoid the blindspots which have captured everyone else. I have been running hosted miners since the market bottom in 2022. While mining has been profitable and I continue to run my hosted machines for as long as they remain profitable, in the long term I think mining needs to become unprofitable. For reasons I hope to expound in a future article, I think that mining may need to transform out of the
“business” mindset during the next 10-20 years if it is going to avoid state capture. So in summer 2023 I began exploring solar tech as a way to keep old machines running. I don’t care much for “green energy” and climate alarmism. For me, solar tech is about easily decentralized power sovereignty. I spent this winter getting hands on knowledge with an S9, S17, and S19 for space heating, and planning and building out my first solar mining system. From December to February I ran initial tests on the 100 Acres Ranch direct DC Hashbox, and put up the final installation in April 2024. Unfortunately it is still not running at full power, but I hope to finish the remaining pieces in the next month or so.
Upfront I’d like to make it clear that this build is unfinished, and is too underpowered to truly demonstrate direct DC potential. Hopefully I can expand soon and push the limits. I will describe some things here which may sound critical of the 100 Acres Design. I bought every part of this system myself, no one has paid me anything to review this, so I can write freely. I think direct DC has huge potential in general for solar mining. The 100 Acres Hashbox is a good system. It is built well, it does what it says, and can work really well IF you design and build the system around it carefully. You cannot just slap any type or number of solar panels on the box and expect it to magically give you amazing efficiency because its “direct DC bro!” You have to do it right, and generally that means a large number of low voltage panels. This also means you need space for a lot of panels. There are many ways to do this build wrong (I’ve discovered most of them!).
I understand Bitcoiners don’t care much for credentials, but I have a degree in computer engineering which I think has enabled me to understand some of the intricacies with solar mining, and I’ll try to share what I’ve learned so far here. This writeup will hopefully serve as a useful primer for miners interested in direct DC mining, and as a helpful guide for anyone considering building a 100 Acres Hashbox system specifically. If you are interested in consulting me for more tailored design advice I am on Twitter/X @keegreil or Telegram @AgentP137.
“For your safety”
I would be remiss if I started an Ungovernable Misfits article without the requisite appeal to safety. Seriously, high current DC power is no joke. There are reasons no one builds large solar installations this way. The risk profile and failure modes are entirely different than a high voltage low current system. In the latter, you are more concerned with good insulation, and short circuits causing tripped circuit breakers. In this low voltage regime, connection failure typically results in melted or vaporized copper starting fires. You can have a bad connection overheat and vaporize without ever tripping a breaker. In a typical mining server, the low voltage high current of 200+ amps travels through two connection points and about one to three inches of solid copper bar. In this build, we are pushing 300-400 amps potentially thirty feet or more through at least 8 connections. Each one presents a possibility of failure, especially after repeated thermal cycling. It can be done, and it can be done safely. But it requires a LOT of copper and good solid high quality connection points. I have found that periodically inspecting with a thermal camera is useful as well. In the following picture, the “hot” wires were in fact only barely above room temperature. So it’s good to “calibrate” what you see in the picture by getting hands on a few things. But this method should make any dangerously hot connections easy to spot.
So, if you take this on, do it carefully, don’t skimp on copper, and check and re-check connections often. But you probably shouldn’t bother building it at all. You know, “for your safety.”
A note on compliance
Another consideration is that a fully off-grid system like this can make compliance with your local power regulations much more difficult. You likely won’t have to pull any permits, or report power usage for mining to anyone. There will be no second commercial meter, no building inspectors, and no compliance department to help you through the process. You are pulling power out of the sky and turning it into hash without any oversight or procedural governance. This can be an uncomfortable position to be in for anyone who understands the importance of compliance. Perhaps more concerning for some readers is that solar technology is generally assumed to be a highly compliant form of power. Solar is “green”, “clean”, and “good for the earth”. Installing solar panels on your property may cause your neighbors to assume that you are a more compliant and “green-conscious” person than you actually are.
Cost breakdown and ROI
The creator of the Hashbox touts efficiency and ROI as the primary advantages of his technology. Having built one, I have major caveats to both. First the numbers. These prices are what I paid, you may or may not be able to find the same prices, but in most cases you should be able to do better than me.
Solar Panels: $2,487.10 (this was not enough panels!)
https://www.ebay.com/itm/335075375898?var=544301079740
This total price is probably about right. On one hand, I overpaid A LOT per panel ($0.69/W), and others can likely do much better. On the other hand, you will need more panels than my 3600W to do well.
Pleb-Tip: PayPal Credit offered me 0% APR for 2 years for opening a new credit line, which I used to finance the panels. This was the only debt I used to build the system. Ironically, this is better financing than the Pubcos can get! Sometimes there are advantages to being small.
I had originally found 3kw of used commercial panels for $820. If you search around the used panel market you can generally find them for $0.25/w, or even less in rare cases. Unfortunately most panels, especially larger cheaper commercial ones, tend to be 30v-48v. I discovered the hard way that the Hashbox requires unusually low voltage panels to make use of all the available power. See the section “PWM vs. MPPT” for a primer on why. Essentially, all the available power above 14V is wasted. I actually hooked up the box to my 48v panels in January. It ran, but with abysmal performance. That’s how I discovered that 12V panels are a better fit. 12v panels are typically marketed for small retail systems like campers and RVs, and difficult to find in large quantities on the used market.
I also did not buy enough, even of the 12v variety, to really get the maximum efficiency in the miner. The reasons for this are complex, see the section on “Understanding direct DC efficiency” for a detailed theory. 6kw is a more reasonable amount of panels to run a single miner and maximum “efficiency”.
Batteries: $2,000.00
https://signaturesolar.com/eg4-lifepower4-lithium-battery-12v-400ah/
The stock system requires 400AH of 12v LifePo4 batteries to operate. These are all generally the same price. Occasionally you can find deals, and you can save some money ordering individual cells from China and building your own batteries, but this adds time and complexity. My first set of batteries cost $930.68 from Walmart. They were sold as “marine grade”. I learned that hard way that does NOT mean waterproof. After a month water ingress killed both BMSs. I went with higher quality batteries from Signature Solar because they have excellent tech support, a 5 (or 10!) year warranty and were only marginally more expensive. I also doubled the capacity to 800AH to reduce strain on the batteries and add system redundancy. If one fails, I can continue to hash on the other while dealing with warranty replacement.
Trackers: $722.38
https://www.ebay.com/itm/124431577755
Trackers are an interesting design option. They are a movable frame with actuators on two axes, so they can automatically turn the panels to face the sun as it moves position in the sky throughout the day. They are more expensive than a static rack, but not by much.
They do add system complexity, but even if they fail they can still work as a static rack as long as the failure isn’t due to catastrophic wind. They offer two advantages. First, they increase total energy production around 30-40%. Second, and most importantly for mining, they smooth out the power curve and allow the miner to run at a more consistent rate throughout the day. Static panels produce very little power in the morning and evening, and too much power around noon. With an isolated system like an off-grid mine, you have to simply charge batteries in the morning if you have too little power in the morning, then overclock the miner like crazy in the middle of the day and keep charging batteries with the excess, then draw that power back out of the batteries to keep running in the evening (with consequent efficiency losses). My hope with trackers is that I could use a smaller total amount of panels, reduce round trip energy losses to/from the batteries, and run the miner at a more consistent hashrate. I also wanted to test them out and collect data so that I can share with anyone else interested in building a solar system like this.
Trackers do best when your location gets a lot of direct sun, and not too much wind. In Oklahoma or Wyoming you are likely better off getting a windmill, as high winds can destroy the tracker. In Seattle, you are better off buying more bi-facial panels which will continue to produce power in cloudy weather. My system is in a location that is fairly middle of the road, somewhere between Arizona and Seattle.
As you can see in the title picture, I opted to build “rockjacks” to mount the trackers onto. This is DEFINITELY not recommended or supported by the manufacturer. The directions specify a poured concrete pad. I wanted the system to be mobile, as there are some trees at my location which will be coming down soon and open up better location options. The rockjack frames are staked quite firmly. Time will tell whether they are staked down firmly enough.
The 100 Acres Hashbox: $2000
I understand the price has dropped to $1800 recently as more orders have enabled better relationships with parts suppliers. The box consists primarily of a large metal utility box, a monstrous supercapacitor, four 100A charge controllers (to take in the power from the panels), a relay to power on/off the miner, a large copper power bus, and a Raspberry Pi which is running the 100 Acres Ranch proprietary algorithm to control miner hashrate according to the system’s available power and voltage. He also includes a couple exhaust shroud options for ducting the heat in case you are reusing that (you should!). It also includes two sets of 4/0 AWG wires for the batteries with high quality hydraulically crimped end connectors.
Other Extraneous Items: >$500
The stock Hashbox requires low voltage and extremely high current, which means big thick cables and a LOT of copper. Copper is expensive, and efficiency losses are high if you have long wire runs. It’s best to keep the box as close to the panels as possible. My trackers are 30 feet apart, so 15 foot cable runs from each side. I estimate I’m losing 8% of the power just in the wires, but that 8% would have been lost anyways because it’s in the excess voltage range that the PWMs can’t make use of. Planning and doing just the solar cabling is probably the hardest part of this build. You have to think about how you want to connect to the panels, where to combine the wires and parallel the power, what gauge wire to use, etc.
I spent at least $100 in bolts to attach the trackers to my custom rockjacks. I had to buy large wire crimp tools and connectors, and deck screws for the rockjack frames. You will need some kind of stand for the Hashbox to sit on, as its air intake is on the bottom. Also do not forget a container for batteries, unless you are smarter than me and get waterproof ones.
Lastly, you may desire fuses at a minimum. Standard solar installations use nice “combiner boxes” which parallel the different strings and include circuit breakers and lightning arrestors. Since this system is isolated from the grid, I went with cheaper individual panel fuses, a 30A fuse on each panel. Give this article a read: https://www.windynation.com/blogs/articles/how-to-properly-fuse-a-solar-pv-system. In short, when connecting panels in parallel like this (to get maximum current and low voltage), there are fire risks if one panel develops an internal short circuit. This could happen due to manufacturing defects or an unlucky hail stone. If one panel shorts, ALL the other panels wired in parallel will dump ALL of their current through the shorted panel. On this build this potentially means 200-300A from your other panels and another 400A from your batteries (which are wired on the same bus) all trying to go through one pair of 10 AWG wires to the dead panel. If I do get a short on a panel, the fuse should isolate the dead panel and protect the rest of the system.
Total Cost and ROI discussion: $7,709.48
I have not included the cost of the miner because in this particular case I think it is the wrong way to look at ROI. Normally in Bitcoin mining, ROI calculations are concerned with miner cost and power cost. Assuming a fixed power cost, how quickly will a given machine pay itself off. With a solar system, what you are really building is a power plant. There are no ongoing costs, unless you have to rent the land or pay maintainers. All the cost is upfront capital. So a better way to think of it is in comparing it to the alternative, for example paying for hosting. You can typically buy hosted power for $0.08/kWh. Instead, hypothetically, how much would you pay a hosting contract that was all upfront cost? How much would you pay upfront for free power for 20 years, if that power was intermittent and only lasted 4hrs/day?
In this model, you are paying to build the power plant. Most of the system components should last 20 years, at least in theory. Because it produces free power, you run older cheaper miners and swap them out every few years.
100 Acres Ranch often describes this system as the “fastest ROI”. I believe this is from a quick back of the envelope calculation of $5/day revenue, and $2000 cost for the box ~400 days. Obviously, there is a LOT more cost to one of these installations than just the hashbox, unless you happen to already own a large low voltage solar installation (which NO ONE on earth does, because NO ONE builds large 12V systems).
At the time of writing my system has only 4 of 12 panels actually hooked up, the others are waiting on a repaired PWM controller and my finished shunt power monitoring modification. So I cannot truly say how much revenue I will generate, and I don’t have enough data yet to even provide a good guess. But as a VERY tentative teaser, I can say that I’m producing roughly 3kWh per day, so at full capacity I should make 9kWh/day. With MPPT modifications I may be able to increase that to 10-12kWh/day. If the Bitcoin network pays my miner $0.08/kWh (roughly current breakeven) then my revenue might be around $1/day, and my ROI on the entire system is more like 21 years, not including brief increased revenue during bull markets. Since I already have the hashbox and batteries I can increase production by simply adding more panels, which helps to dilute the sunk costs of hashbox and batteries. With that my system cost could easily surpass $10k, but my revenue should be closer to $3/day with the increased panel production and increased miner efficiency, and ROI closer to 9 years, not accounting for bull runs. With a newer gen miner like a kPro or XP I may be able to push that revenue up further, but I’d then have to consider the cost of the miner as well.
In conclusion, to build one of these systems and do well you have to go big. You cannot buy 3.5kw of panels for a 3.5kw miner and expect good results. Here’s what you need to do to build the stock system and do well if ROI is your primary goal (it shouldn’t be):
- – Access to SUPER cheap, low voltage panels. I overpaid.
- – Access to enough space for 7kW of panels. This is a LOT of space.
- – Think in Amps, not Watts.
- – Understand these terms: Voc, Vmp, Isc, Imp. Isc is the key one for the stock build(Short Circuit Current). These are standard solar panel specs listed on every solar panel. Production in perfect sun will be pretty close to this number. Use Imp (Max Power Current) for a slight underestimate in perfect sun.
- – Understand voltage drop along wires, especially with high currents, and how to calculate it.
- – Price your panels in $/Amp (Isc) instead of $/W (Read “PWM vs MPPT” to understand why). I paid $11.50/A. Others have done much better.
- – Aim for at least 300A of rated power output which is enough to run a miner with a small overclock and top off a battery at the same time. More is better. The box can handle up to 400A, so maxing out the box might produce enough for two lower power miners like the kPro, doubling your revenue.
– My total cost so far of around $8k is too low. A higher cost of $10-12k would allow for twice as many panels, and likely 4x the revenue due to increased efficiency at higher power levels. This is VERY counterintuitive, see the section “Understanding Direct DC Efficiency” for more.
Additions and Modifications
What I’ve described so far is essentially the stock build. I’ve made and am making certain additions and modifications to the stock box to better suit my purposes. I haven’t included their cost since they are highly customized and mostly related to remote monitoring and data collection.
First, I added another Raspberry Pi and special USB data cable from Victron. On this I am running the open source version of Victron’s solar monitoring software, Venus GX. This lets me remotely monitor system voltage and current through the miner, and control the relay which powers on the miner. I can also get historical and forecasted solar irradiance data for my system’s location.
My tracker controllers and motors are powered from the same system batteries.
I also tested using a Bitaxe to provide a gentle 15W of steady heat to keep the batteries warm throughout the day and night. I need to build a better enclosure, but this did work just fine. It also added a nice level of redundancy, in that if the Bitaxe was still hashing, then the local internet must still be working.
I am in the process of adding a few more things:
- – DC shunts to measure amperage production from the two trackers. I’d like to run experiments with locking one and letting the other track, to get real world comparison data on using a tracker vs. fixed rack in various weather conditions.
- – MPPT controllers: See the later section on why. I’d like to add 2-3 MPPT charge controllers to replace the PWM variety that come in the stock box. These are more expensive, but can pull more usable power from the same panels. Again, I’m mostly interested in collecting real world data and to trial whether they are usable for powering miners. Hopefully I’ll add enough to power half the panels, and leave the other panels using the stock PWMs, and then gather comparison data.
– Voltage or current monitors on the stock PWMs. These are VERY cheap, $50 charge controllers that are primarily used to limit the voltage coming in from the panels to something safe for the miner and batteries. I’ve had several glitch out and turn off. There is no way to know one is dead except physically going to the box and testing voltage on each solar terminal. Since my system is remote, I’d like a way to know quickly if/when one dies so I don’t lose out on that production.
Pulse Width Modulation (PWM) vs. Maximum Power Point Tracking (MPPT)
First, math.
Power (Watts) = I (Current or Amps) x Volts
Everything hinges on understanding this.
See this guide for a more in-depth overview of these two different methods of driving solar panels:
https://www.victronenergy.com/upload/documents/Technical-Information-Which-solar-cha rge-controller-PWM-or-MPPT.pdf
In brief, solar panels are kinda strange. Photons smash into the panel and free up electrons which can then flow through wires and do work for you. Panels will produce a relatively constant number of electrons, i.e. a constant current almost no matter what voltage they are at up to a point. So you can pull 10A off a panel at 1 volt and produce 10 Watts, or you can pull the same 10A of the same panel in the same sunshine at 48 volts and produce 480 watts. In a chart, this relationship looks like this:
There is a special point on the curve where the maximum power is produced, called appropriately the “Maximum Power Point (MPP).” The ideal way to pull power from a solar panel is to drive it precisely at that MPP, i.e. to vary how much current is pulled off the panel such that the panel voltage remains at the MPP. This creates our first problem, those watts are now probably at a weird voltage that is too high to safely charge batteries or run a hashboard. So we have to do DC to DC buck conversion, which takes high volts and low amps and turns it into low volts and high amps. So now you can take the 10A and 48V from our example and produce 34A at 14V (480W/14V), minus efficiency losses of 1-15%. This maximizes the number of Watts produced, but the second problem is that the circuitry involved in doing that voltage conversion is really expensive. Quality MPPTs that can handle 100A are around $600, over 10x the cost of the PWM controllers used in the 100 Acres Box. Four of those would more than double the cost of the Hashbox.
This cost is the main reason why 100 Acres opted for PWM controllers. A PWM controller is essentially just a voltage limiter circuit. It limits the solar panel to a voltage safe for charging batteries. It pulls whatever amps it can off the panel at that voltage. They are cheap because they are simple, they don’t do any voltage conversion. Another advantage is that raw current production is relatively unaffected by panel temperature, so production is more consistent throughout the year (except for clouds). See the second chart above to understand this relationship. The downside is they will produce fewer total Watts than an MPPT. My low voltage panel’s MPP is around 18V. Chopping this down to 14 means I’m only losing maybe 20% of the Watts that an MPPT controller could produce. If however you try to use 48V panels (like I did at first), the PWM will chop that 48V MPP down to 14V, essentially leaving 70% of the rated power on the table.
This is unfortunate because the solar industry in the US at least seems to have commoditized around large, 500W 48V panels. Huge solar farms, subsidized by the fiat system and further manipulated by the insurance industry, regularly throw away perfectly productive 48V panels. This was my first set of panels, cast offs from a large commercial farm. They were extremely cheap, but the stock 100 Acres system cannot make use of those panels efficiently. You have to deal with voltage conversion to take advantage of those.
My understanding is 100 Acres Ranch is working on future versions and upgrades that will work on higher voltage systems. The holy grail would be a 100A buck converter that can adjust output voltage to follow commands from the miner control board. In the meantime, you can try to source low voltage panels, or you can try something new (see “Future Ideas and Experiments”).
Understanding Direct DC miner Efficiency
Everything you think you know about miner efficiency is wrong.
The standard model of miner efficiency assumes that there’s a sweet spot for efficiency somewhere a little below stock hashrate. For most machines you can underclock a bit and keep them profitable a little longer, or during bear market bottoms. Overclocking a bit reduces efficiency, but can generate more revenue per machine, so may make sense in certain business environments like maximizing the use of limited rackspace. This perspective is only true of standard mining off of AC power using the stock power supply. Direct DC is entirely different. To understand why, we have to break down the mining machine into its components and understand how the PSU and Hashboard efficiency curves interact. After that, the promise of Direct DC mining will make sense.
AC to DC Power Supply Efficiency
Below are charts of several Bitmain PSU’s AC to DC conversion efficiency at various power levels. The general shape of this curve holds for almost all voltage conversion methods. There is a sweet spot that is actually somewhat below the designed operating current. Reducing hashrate actually raises PSU efficiency, to a point. Conversely, increasing hashrate reduces PSU efficiency, and is likely the primary reason that miners assume that overclocking reduces J/TH efficiency.
DC Hashboard efficiency
The hashboards themselves are an entirely different story. From my own tests on a 96TH S19 jPro, I’ve observed that:
– Hashrate is essentially linear with frequency (unless voltage or temperature are too low and the trons can’t get through the logic gates fast enough, causing unhealthy chips )
– Amperage is linear with frequency too, except for variations from temperature.
– Most importantly: The Y intercept of Amps vs. Freq isn’t 0. There is overhead current required from just having a board powered on. “Powering Off” a hashboard in Luxor OS firmware, which I did my testing on, does not remove this overhead current. Even putting the miner in “curtail” mode with fans off still drew about 160W.
The implication is that if voltage is held constant, either because that is your test condition or because you have no control over the voltage at all (as in the 100 Acres Hashbox), then you get the best efficiency by running as high a hashrate as you can get away with without glitching out the chips or burning up your boards.
A note on temperature
The following chart is a test run I did with constant (very low hashrate), slowly drawing down battery voltage. Blue line is volts, red line is amps. The key things:
- Most of the test, fans were held constant at 50%
- In the blue circle I set fans to “automatic”, which favored 10-20%
- In the red circle fans were 100%.
This is a pretty counterintuitive result, but aligns with data gathered by Braiins on S19 chip temperature (https://braiins.com/blog/impact-of-temperature-on-efficiency-of-antminer-s19-models). At manual 50% fans the chips were nice and cool, around 34C. When the auto mode reduced fans and allowed the chips to warm up, amperage draw increased even though voltage was relatively constant. Hashrate was constant too, so I was actually losing a LOT of efficiency by allowing them to warm up.
What this also means is that given the same voltage, you will be able to push more amps through the chips (and can do more hashes if the frequency is increased accordingly) if you allow the chips to warm up first.
I believe the reason this occurs has something to do with the fact that semiconductors have a negative temperature coefficient. “In semiconductors, the energy gap between the conduction band and valence band decreases with an increase in temperature. The valence electrons in the semiconductor material gain energy to break the covalent bond and jump to the conduction band at high temperatures. This creates more charge carriers in the semiconductor at high temperatures. The higher concentration of charge carriers decreases the resistivity of the semiconductor. As the resistivity of the semiconductor decreases with an increase in temperature, it becomes more conductive. A semiconductor exhibits excellent conductivity at high temperatures.
How I currently understand miner efficiency in the DC regime
Now understand the PSU and the hashboards together. While the hashboards are more efficient at higher hashrates because they dilute the overhead of just powering the board on, this seems to be more than offset by efficiency losses in the PSUs themselves. This is why everyone assumes that overclocking reduces efficiency.
Next, eliminate the PSU. This is how miner mining energy efficiency works on a direct DC system. It is not simply that you gain 10% by eliminating AC to DC conversion, or 20% by eliminating both the inverter and PSU as in a standard solar off-grid system. If you have the amperage available (because you used a LOT of good low voltage panels), and if you allow the chips to get warm enough to easily flow the current, you can run the hashboards at far higher frequencies than normal without pushing the stock PSU deep into the far right portion of its power curve and offsetting all your hashboard efficiency gains. The catch is, you MUST have the raw amps to do it. You cannot build an underpowered direct DC system and expect to do well simply because you’re underclocking it. In fact with direct DC the more you underclock, the worse your efficiency gets.
This means that there is a minimum viable size to get a stock Hashbox running well, and that likely means 300 -400 amps worth of panels, or about 5-10 kW rated power (depending on panel voltage), not 3.5kW. A single hashboard version would likely perform very well, and would make the total system size and cost much more accessible. I understand a single hashboard version is under development by 100 Acres Ranch and look forward to them releasing it.
In my tests, I’ve been able to achieve about 10% better efficiency than a stock jPro. My system does not yet have enough power installed to overclock and really test the limits of Direct DC or the theory that I presented above. The best I’ve observed so far was 25.6 j/TH with the 100 Acres autotuning, at 12.67V and 75TH, or around 13% better than stock efficiency. Not earth shattering, and is basically accounted for by eliminating the 10% loss in the PSU and a bit of tuning/undervolting. Given that the 100 Acres box does not do any voltage conversion whatsoever, the important efficiency metric here is actually Amps/TH, not j/TH anyways.
Further, I think it’s an open question whether j/TH or amps/TH are even useful metrics when trying to maximize a particular intermittent power system. Really what you want to maximize in an isolated mining system like this is hashes per day, or hashes per hour of available sunshine. Sometimes lower efficiency might make sense if you need to burn off excess energy, and storing it in batteries would incur efficiency losses itself. Perhaps a better metric for solar mining would be total shares submitted per day, or total hashes per day. Then normalize that for total daily irradiance. Systems designed for Arizona vs. Michigan almost certainly require different optimizations to squeeze every available hash out of the power available.
To push on this, even 17 j/TH efficiency measured directly at the shunt before it enters the hashboards is mostly a party trick, especially if the whole system is only making barely 200TH out of 19kW installed capacity (to use a hypothetical example). How much energy is lost in the wires due to the low voltage and high current runs? How much energy is never generated because the PWMs are driving the solar panels at a suboptimal voltage? How much energy is lost when charging and discharging batteries? While j/TH efficiency and power rate are the primary determinants of grid based mining profitability, they are at best secondary tools when optimizing a solar miner.
The limitations of batteries
The batteries are by far the most annoying part of this entire system. They are necessary to smooth out power fluctuations from the panels when clouds or other shadows cut-off the wireless photonic power transmission system from the nuclear fusion reactor in the sky. They buffer the power and allow the miner to stay turned on and hashing instead of constantly rebooting.
Unfortunately, batteries are expensive, can be finicky. Most importantly in this particular system, the batteries are wired directly to the main system bus with no voltage conversion. Normal LifePo4 voltage at rest is 13.1-13.2V. If you are producing more power than you are burning through the miner, the PWMs will push the voltage up towards 14V and charge the batteries at the same time. Once they are charged it holds voltage there until production decreases.
If you would like to burn precisely your production and not excessively charge the battery, then you must throttle the miner to maintain 13.2V. That voltage may or may not be appropriate from an efficiency standpoint for the amount of power you would like burn off.
Perhaps the most difficult scenario though is with a very underpowered system. When the sun comes up and pushes voltage high enough to trigger the relay and turn on your miner, it will immediately start burning more power than is being produced. The difference comes from the battery. All batteries sag in voltage when power is drawn out. The amount of voltage sag depends on the size of the power draw relative to the size (in AH-amp hours) of the battery bank. On my 800AH system voltage can sage as much as 1V if I’m pulling 150-200A from the batteries. This means that the hashboards must now run at 12.2V instead of 13.2V or 14.2V. Underclocking will reduce the power draw on the batteries, helping them to last longer. But, this also increases system voltage, decreasing my j/TH and Amps/TH efficiency. Even a large battery bank is unlikely to last more than a few hours.
One solution is to simply charge the batteries fully with the miner off, then run them down at a high hashrate to maintain good efficiency through the miner. However, running batteries hard like this actually creates efficiency losses in the batteries themselves during the energy round trip, further hurting total system efficiency.
The only real solutions in this underpowered case are to physically disconnect 1-2 hashboards or increase the number of panels. When doing direct DC with batteries and without real voltage control, the best mode of operation is to run the boards really hard as long as you have the power to keep voltage high.
Future ideas and experiments
This first foray into solar mining has been extremely enlightening and rewarding, if not exactly profitable yet. What I would really love to see however is exploration of smaller sized systems with a grid tie in.
Using MPPTs to drive the panels, while expensive, would open up panel sourcing options and reduce the total number needed, further reducing the total size. Making use of the stock PSU for grid tie-in could eliminate the need for batteries, reducing system cost and complexity and allowing finer control of system operating voltage.
I have successfully run the 96TH jPro underclocked to 40TH using an APW3 to provide 250W of grid power with the rest provided by the direct DC solar. Measured at the wall, I was netting 6 J/TH, at least while the sun was shining. Crucially, to do this I didn’t need an expensive inverter, nor did I need batteries. I was able to rely on the grid tie to buffer clouds and occasional power dips, while using the solar to provide the bulk of the power for hashing. An untested extension of this idea would be to get rid of the Loki modification and allow the stock PSU to control system voltage according to control board firmware commands (rather than the battery/PWMs). This would make underclocking far more efficient.
In the slightly more distant future, a collection of Bitaxes could throttle their power use far more quickly than a stock Bitmain hashboard, further simplifying system design and eliminating the capacitor and possibly even the MPPT controller. Stock Bitmain hashboards cannot throttle power up and down quickly without potentially damaging themselves. This is due to their circuit design of long chains of series connected chips on separate voltage domains.
