# Bitcoin mining network supply-side modelling

### A fundamental approach

Cayle Sharrock, April 2016
Nimbus Technologies
and Coinbase Inc.

# Introduction¶

In this paper, we model the supply side of the bitcoin mining landscape in order to draw some inferences about the economics of mining difficulty and an estimate of the capital employed in the bitcoin mining network.

## BTC market data¶

The models make use of a lot of BTC market data. Here is a summary of the data that is collected and used in the models:

Data Source Description / contents
mining_distribution Blogs Estimates of geographic hashrate distribution
hardware Web BTC mining hardware specs
electricity Web searches / news Regional electricity prices.
difficulty_history Blockexplorer Mining difficulty history
network stats Quandl Daily average network hashrate, BTC supply, block count
BTC price Quandl Weighted average BTC price over 8 exchanges

The BTC market data is updated daily. Mining distribution, electricity and hardware tables are updated as new information becomes available. A summary of some of the data used is given in the Appendices.

# Methodology¶

The ultimate goal of the model is to be able to generate a cash cost, or supply curve for the bitcoin mining network at any instant in time. The cash cost curve as used here is the total cumulative hashing capacity of the network, ranked in order of increasing cost to operate. Both operating cost (cash cost) and capital costs (fully absorbed cash cost) are estimated to this end. The utility of presenting the data in this way is that it becomes easy to see at what bitcoin price a given operator becomes unprofitable; and rationally, should shut down.

The process of generating these curves is somewhat involved and is described next.

## Algorithm¶

### Hardware cash costs¶

Firstly, and independently of the actual BTC market dynamics, we calculate the running and capital costs of all hardware in the Mining Hardware database (Appendix C) over the date range of interest.

Costs are calculated in \/BTC with the following \begin{align*} C_e &= 0.024 \cdot W P_e \\ C_c &= C_e (1+\alpha) \\ C_d &= \frac{P_h}{365\cdot L} \\ C_r &= r C_d \end{align*} whereC_e$is the daily electricity cost,$C_c$is daily operating cost (electricity + overheads),$C_d$and$C_r$are depreciation and return on capital resectively.$L$is the lifetime of the hardware, typically 1.5 years,$r$is the ROI expected, typically 15% for investment purposes, but a 0% return will be used here to illustrate the break-even point.$\alpha$is the overheads fraction. Electricity costs are estimated from a range of online sources, including regional electricity price publications and prices quoted by mining farm operators in news reports and press releases. While significant effort has been made to use reliable figures, there is significant uncertainty in the values used here. The error in cash cost estimates is proportional to the error in the electricity price. Of course, electricity is not the only input cost in bitcoin mining. Maintenance, staff, insurance and network fees also contribute to the cost of producing bitcoin. These costs are not modelled explicitly here, but are rolled into a single, constant, overheads factor (set at 45% for the calculation presented here). This is clearly an oversimplification; overheads will be lower in China than in the U.S.A. or Western Europe for instance, where labour is far cheaper. Some attempt to take this into account can be made without modifying the model by regarding electricity price as an effective total variable cost and adjusting the electricity price database accordingly; but this was not done in this set of calculations and should be borne in mind when reading the discussion to follow. The model builds up a supply network by simulating addition of every piece of mining hardware in the network. To facilitate calculations, it is more convenient to represent costs in terms of BTC produced rather than a daily cost. This is done for each piece of hardware using \begin{equation*} r_{b,i} = 24 \cdot 3600 \cdot s_i \frac{H}{B} \end{equation*} where Symbol Description$r_{bi}$is the bitcoin rate for the ith machine$s_i$the hashing speed of machine i H the expected number of hashes to mine a block (as a function of the given difficulty for each day in the time range of interest) B the number of BTC released per block (as a function of time -- 25 up until July 2016) The cost (cash and fully absorbed, which includes depreciation and ROI) formulae in terms of$/BTC mined are then

\begin{align*} \hat{C}_c &= \frac{C_c}{r_b} \\ \hat{C}_f &= \frac{C_c + C_d + C_r}{r_b} \end{align*}

### Global hardware supply calculation¶

An estimate of the number of mining hardware rigs available on each day of the time series of interest is calculated using the following algorithm.

For every day:

• Determine the most energy efficient available hardware. The frontier equation for determining the 'most efficient' hardware is variable, but it is a strong function of J/GHs and \$/GHs. There is typically a clear winner though, regardless of the actual formula used. • Calculate the marginal hashrate,$\Delta H$, the increase from the previous day's calculated capacity and the historical average hashrate for the current day (or the actual hashrate for day 0). • The previous day's supply is$\sum s_i n_i f_i $, where$n_i$is the number of machines that have been produced for each hardware type, and$f$is the fraction of those machines that should be running based on their cash cost and the actual BTC price (i.e. miners shouldn't run at a loss, but they may temporarily -- several empirical relations were used to estimate the form of$f$). • If$\Delta H > 0$, then calculate how many machines are required to produce the marginal hash rate. Update$n_i$for the current day. • Proceed to the next day Note that for Day 0, the marginal hashrate is the network hashrate. This will give the wrong hardware distribution for Day 0 and indeed until that hardware becomes obsolete. For this reason, the start date for the calculations was always chosen to be early enough that GPUs were still the dominant mining hardware. By the time that GPUs contributed an insignifiant percentage of the total hashrate (Q2 2013), one could assume that the hardware distribution provides a reasonable picture of the overall mining supply landscape. ### Participation curves¶ Several formulae are available for the participation function,$f$, but the one that seems to match behaviour the best is given by \begin{equation*} f = 1 - e^{-\beta\frac{P_b}{\hat{C}_c}} \end{equation*} where$P_b$is the BTC price. This curve for different values of$\beta$is plotted in Figure 1. Our empirical investigations into matching hashrate with hardware investment suggest that$2.5 < \alpha < 3.5$. A value of 2.95 was used for most of our calculations. Note: For ratios above 1, f is always 1. In [2]: # Compute participation curve b = np.array([2, 3, 4]).reshape((1,3)) x = np.linspace(0, 1, 50) f = 1 - np.exp(-b.T * x) pyplot.plot(x,f.T) pyplot.legend([2,3,4]) pyplot.xlabel('Cash cost - Price ratio') pyplot.ylabel('Operating fraction') pyplot.title('Figure 1 - Sample operating fraction curves using an exponential participation model')  Out[2]: <matplotlib.text.Text at 0x7f1217d3f290> In [3]: # Generate supply curves start_date = '2013-01-01' end_date = '2016-03-30' btc = bitcoin.BTCMarket() supply = bitcoin.MiningSupplyGenerator(start_date, end_date, None, overheads=0.30, market=btc, lifetime=1.5, roi_rate=0)  In [4]: # Calculate global network capacity supply.calcHardWareCosts() supply.calculate_capacity('exponential', alpha=2.95);  # Discussion¶ ## Hardware distribution history¶ The first thing the model allows us to visualise is the evolution of the mining hardware landscape. The mining hardware supply was estimated over the period from January 1, 2013 to April 1, 2016. This period encompasses several BTC price bubbles including the major one in H2 2013. Hashing capacity has been added to the network at an exponential rate over almost the period under consideration. During the bubble of 2013, hashrate addition was super-exponential, representing an unprecedented investment in bitcoin mining hardware capacity (See Figure 3). Figure 2 displays the hardware distribution over time as the hashing difficulty increases and the hardware improves. The charts in the left and right columns show the same data, with the only difference being that the left hand column has the hashrate on a log scale. In addition, the top row plots all installed mining capacity (whether it is running or not) and the bottom row plots only those machine that are (or should be) running, according to the participation model,$f$, employed. Several interesting features of the plots to note are: 1. How quickly mining equipment moves from dominating the landscape to becoming insignificant (12 - 18 months). 2. How the state of the art equipment becomes the dominant hashrate contributor within 3 - 6 months of its release. This means that the most efficient hardware only about a year; 18 months if very lucky; to produce a return on investment (and R&D costs) before the arms race and exponentially rising difficulty level overtakes it and renders it scrap metal. In [5]: # Get total and operating mining capacity hrDataRunning = supply.getHardwareDistribution('reciprocal', alpha=2).sum(axis=1, level='Product') hrDataAll = supply.getHardwareDistribution('all').sum(axis=1, level='Product') # Plot configuation netHash = supply.network_hashrate fig, axes = pyplot.subplots(nrows=2, ncols=2, figsize=(16, 16)) #axes[0].plot(hrDataAll.index, hrDataAll) axes[0, 0].set_yscale('log') axes[0, 0].set_ylabel('Network capacity(GH/s)') axes[0, 0].plot(netHash.index, netHash, 'k:', label='Actual historical') hrDataAll.plot(stacked=True, kind='area', ax=axes[0, 0])<-> axes[1, 0].set_yscale('log') axes[1, 0].set_ylabel('Network operating capacity(GH/s)') axes[1, 0].plot(netHash.index, netHash, 'k:', label='Actual historical') hrDataRunning.plot(stacked=True, kind='area', ax=axes[1, 0]) axes[0,1].set_title('Mining capacity, including mothballed equipment') axes[0, 1].set_ylabel('Network capacity(GH/s)') axes[0, 1].plot(netHash.index, netHash, 'k:', label='Actual historical') hrDataAll.plot(stacked=True, kind='area', ax=axes[0, 1], legend=None) axes[1,1].set_title('Mining capacity, operating equipment only') axes[1, 1].set_ylabel('Network operating capacity(GH/s)') axes[1, 1].plot(netHash.index, netHash, 'k:', label='Actual historical') hrDataRunning.plot(stacked=True, kind='area', ax=axes[1, 1], legend=None); print 'Figure 2 - The evolution of the hashing network hardware distribution from 2013 to 2016'<->  Figure 2 - The evolution of the hashing network hardware distribution from 2013 to 2016  ## Capital investment in Bitcoin mining hardware¶ Figure 3 presents a minimum capital investment into bitcoin mining hardware. It is estimated by summing up the published retail prices of mining hardware rigs for each piece of hardware added to the supply landscape, as given in Figure 2. The value represents a minimum because • For each day in the calculation, only the most efficient available hardware is used to account for the entire marginal increase in hashing capacity. • If less efficient hardware is actually installed, its cost per GH/s is necessarily higher than what was estimated here. • There is no such thing as "free mining". Any other hardware contributing compute cycles to BTC mining, whether knowingly or not, including things like illegal zombie networks are contributing greater-than marginal-cost hardware to the network. This necessarily represents a higher capital investment than what we've allocated. The fact that the beneficiaries of the mining effort haven't paid that capital is irrelevant. This graph is useful to estimate what it would take for an independent malicious attacker (let's say a rogue government or terror organisation) wishing to execute a 51% attack on the network would need to spend asuming they have no hashing capacity now. As of 1 April 2016, this value stands at approximately \$2.35 bn, assuming the rest of the world stood still. It is an interesting thought experiment to consider how the world would respond if it relied on bitcoin for its financial infrastructure today, and North Korea decided to try and build a mining farm to bring the network down; perhaps something for a future post.

One observation here is that \$2.35 bn represents 36% of bitcoin's current market capitalisation of \$6.5 bn. Our estimates suggest that capital accounts for anywhere between 35% and 70% of the fully realised cost of mining bitcoin, suggesting that society is getting fair value at least from the current bitcoin price at its level of \$420/BTC. Another noticeable observation about Figure 3 is the remarkable addition of hardware over Q1 2014, where at least$1.0 bn (with a B) was added to bitcoin mining operations in 3 months. Sources suggest that large sections of China's IC fabrication units were turned onto ASIC manufacture during this period, looking to cash in on four-digit BTC prices. This is an awesome demonstration of the flexibility of the sector given the right economic incentives.

In [6]:
# Calculate cumulative capital expenditure<->

Out[6]:
<matplotlib.text.Text at 0x7f120fc3b910>

## Cash cost curves¶

The most useful aspect of the hashrate supply model is the ability to plot cash cost curves for bitcoin mining at any instance in time.

Figures 5 and 6 show the curves at various points in the interval January 2013 to March 2016.

Figure 5 shows some cost curves in and around the major price bubble that occurred late 2013. The blue lines indicate the estimated cash cost (electricity and overheads) for the entire network on the given day. The green lines include the cost of paying back capital, given an 18 month lifetime for mining hardware.

(Note: The green curves are 'bumpy' because of the low granularity of the cost assumptions used in the model. Since only the the most efficient machines are used in the marginal supply calculations, there are only a dozen or so hardware types-- and subsequently, prices -- used in the model. The electricity consumption, being a more continuous variable and the major input into cash cost, results in a much smoother curve.)

Three snapshots are given: October 1, 2013 (when the BTC price was \$132), two months later on November 30, 2013, when the price had exploded to \$1135, and May 2014, when it had come down to \$454. There are several observations to make about the grpahs in Figure 5: Firstly and probably most significantly, the bubble ruined things for miners, probably forever. Prior to the bubble, at \$130/BTC, miners were making absolutely huge margins. Fully absorbed cash costs were under \$50 for almost anyone running any kind of ASIC resulting in capital returns in the region of 300% - 500% p.a. These margins obviously wouldn't have lasted for ever, but the price bubble brought a lot of attention to the sector and we saw an astronomical investment in mining hardware over the following 6 months as was shown earlier. This resulted in the super-exponential addition of hashing power to the network, which drove the difficulty and subsequently the cost of mining up massively. At the peak of the bubble (only 60 days later), the fully absorbed cost of mining a bitcoin had risen two to three times. This was fine as long as prices were north of \$1,000, but this situation was about to change.

By May 2014, the price had fallen to \$450 (after briefly dipping below \$300). By this time operating costs for the majority of miners were in excess of \$50 per bitcoin, but the difficulty had risen so fast that Moore's Law could not keep pace and the relative capital cost of ASICs was suddenly very expensive. At this time, depreciation costs of all but the most efficient ASICs were around 80% - 90% of the total mining cost. Miners were barely able to cover the costs of their equipment, let alone produce a return. We now found ourselves in a position where it made sense to continue mining if you had already bought hardware (the bitcoin price was comfortably above the operating cost), but mining, like its traditional counterpart, was now a capital intensive and risky endeavour. This meant that very little further investment could be made while the bitcoin price remained range bound, but the difficulty would not decrease because cash costs were still relatively low. The gold rush of Bitcoin mining was largely over. In [7]: # Collect data for 3 interesting dates 2013-2014<->  The difficulty continued to rise steadily, though at an almost linear rate over the course of 2014, even as the price fell further to settle at around \$250. New hardware started to come onto the market, including the AntMiner S2 and 3 series, reducing the capital intensity of mining and allowing some continued investment to take place. This of course, continued to drive up the base cost of BTC production and for the first time, we see a significant portion of the network operating at, or just below break-even by February 2015.

Figure 6 shows some more cost curve snapshots from the period 2015 - 2016.

The price remained below \$300 for the first three quarters of 2015, resulting in a sort of stasis for the mining landscape. The cash cost curve seemed to remain fairly stable, with the most efficient miners producing at under \$100/BTC with a steady increase in cost to around \$300/BTC at the 85% capacity level. However, this apparent calm belied considerable turbulence beneath the surface. The apparent equilibrium was only achievable due to a constant turnover of hardware, with the increase in ASIC speeds and efficiencies just barely keeping up with the difficulty increases in the hashing algorithm. With approximately 25% of the network barely covering costs, there were bound to be casualties. Over the course of the year several hardware manufacturers went bankrupt 1, 2, 3. In late 2015, too late for some, the bitcoin price rallied to \$500 and provided some breathing room to mining operators. However, the price rally again brought in a rush of new mining capacity. Whereas in 2015, some older equipment could still eke out enough money to cover costs, it seems that this time, only the strongest (read AntMiner S7 and equivalents) are able to survive. As of 1 April 2016, our estimate is that over 70% of the network is driven by S7 (or equivalents) and are comfortably covering their capital costs at a price of \$420/BTC. However, second tier ASICs, such as S5-class machines are at the break-even point. In [8]: # Collect data for 3 interesting dates - 2015/16 dates = ['2015-02-01', '2015-06-01', '2016-03-30'] titles = ['price bottoming out', 'price stable', 'Q1 2016'] topY = [500, 500, 500] method = 'exponential' alpha = 2.95 n = len(dates) hwDist = supply.getHardwareDistribution(method, alpha, strip_unused=True).sum(axis=1, level='Product') ph = supply.btc_price fig, axes = pyplot.subplots(nrows=2, ncols=n, figsize=(n*8,15)) for col, date in enumerate(dates): hw = hwDist.loc[date,:] cc = supply.getCostCurve(date, method, alpha) cc['cumHash'] *= 1e-6 price = ph[date] labels = cc.index.drop_duplicates() axPie = axes[0, col] axCC = axes[1, col] axPie.set_title('Hardware distribution %s\n %s -$%3.0f/BTC' % (titles[col], str(date), price))
hw.plot(kind='pie', ax=axPie)
axCC.set_title('Cash cost curve %s' % (titles[col],))
axCC.set_ylim([0,topY[col]])
cc.plot(x='cumHash', y=['cashCost', 'facc'], ax=axCC, legend=False)
axCC.set_xlabel('Cumulative hash (PH/s)')
axCC.axhline(price, color='b')
# Plot actual network hashrate
hashRate = supply.network_hashrate.loc[date]
axCC.axvline(hashRate.Value, color='g')
#      #Annotate curve with hardware
#      maxH = cc['cumHash'].max()
#      for label in labels:
#         row = cc.loc[label].iloc[0]
#         lx = row.cumHash + 0.1*maxH if row.cumHash < 0.8*maxH else row.cumHash - 0.1*maxH
#         axCC.annotate(label, xy=(row.cumHash, row.cashCost),
#                     xytext=(lx, row.cashCost - 0.1*topY[col]),
#                     arrowprops={'facecolor': 'k',
#                                 'shrink': 0.05,
#                                 'width': 1})


## The halving¶

This has some ramifications for the state of the mining supply landscsape for the second half of 2016. The current mining hashrate is dominated by a single class of hardware, leading to a very flat cash cost curve. This makes the network very vulnerable to sharp changes in the BTC price, or cost of production.

There is also very little news of new hardware appearing on the horizon, with the exception of the Bitfury 16nm ASIC, slated for H2 2016. It is possible that other manufacturers are operating in secret; it would be extremely surprising if Antmain, for example, did not have a successor to their very successful Antminer S7 series up their sleeves.

Nevertheless, as it stands, the cash cost curve of 30-03-2016 shows that a halving in price to \$200, or roughly equivalently, a doubling in cash cost, would put 30% to 70% of the current hashing capacity out of business. This is a very wide margin, and is due to the very flat cash cost curve brought on by a single dominant participant. In July of this year, the bitcoin reward will halve, bringing this very scenario into existence. Miners will only receive 12.5 BTC per block mined, instead of the 25 they currently receive. Overnight, most of their costs will double. We suspect that what happens next is largely a function of the price at the time. \$400 is a fairly critical level now. Above that, the roughly 600 PH of capacity that will be producing at or around that price will be able to continue operating, meaning that the network will perform as usual and the world may never even notice that a halving ever happened.

If however, the price in mid-July is below the $380 mark, almost anything could happen. In a worst-case scenario we could see a dramatic fall-off of hashing capacity. This would lead to disruption in confirmation times and general disarray in the bitcoin network, affecting confidence and causing the price to fall further. The difficulty would be adjusted downward two weeks later; bringing all the mothballed capacity back online again for two weeks. This might not force the price back up, but the backlog of transactions would start to clear until the difficulty is again ratcheted up in the following adjustment. Thus we may see a period of difficulty and hashrate oscillation with a downward-moving price until 1. Most of the hashing capacity has been replaced by next-generation chips (and the current set would be obsolete anyway). 2. The bitcoin price and network settle down to a new equilibrium value in the \$275 to \$325 price range. We have developed a forecasting algorithm to try and predict how this scenario will pan out, and may release those results in a future post. # Conclusion¶ Bitcoin mining went through a golden period in 2012 - late 2013 where costs were low and the bitcoin price (even at \$120) was far higher than the fully absorbed cash cost of mining. Those early adopters are smiling now.

Since 2014, mining has been in a ruthless arms-race, where only the most efficient ASICs stand a chance of providing a decent return on capital before being usurped by the next generation of chip.

Currently, about 70% of the network hashing capacity is produced by a single generation of chip. This could make the network very vulnerable to widespread disruptions following the halving in July if the bticoin price cannot maintain its present position above \$400, with substantial further downside pressure should the price fall below that level come the mid-July halving. # Appendices¶ ## A. Mining distribution table¶ Estimated regional mining distribution Date China Iceland Europe US 2013-01-01 0.000 0.000 0.200 0.800 2013-06-01 0.000 0.050 0.195 0.755 2013-08-01 0.050 0.050 0.200 0.700 2013-12-01 0.100 0.050 0.250 0.600 2014-06-01 0.150 0.060 0.350 0.440 2015-01-01 0.160 0.060 0.400 0.380 2015-03-01 0.180 0.100 0.440 0.280 2015-06-21 0.580 0.100 0.150 0.170 2015-10-01 0.600 0.100 0.150 0.150 2016-04-01 0.63 0.16 0.1 0.11 ## B. Electricity prices¶ Estimated wholesale prices in$/kWh

E. Europe China Euro Iceland US
2010-01-01 0.063 0.1056 0.0462 0.0677
2011-01-01 0.064 0.076 0.11531 0.0546 0.0682
2012-01-01 0.067 0.0798 0.106128 0.52 0.0667
2013-01-01 0.071 0.086 0.110262 0.05 0.0684
2014-01-01 0.065 0.088 0.085644 0.043 0.0701
2015-01-01 0.056 0.080 0.089964 0.043 0.05
2016-01-01 0.056 0.080 0.089964 0.043 0.05

Price can decrease in parts based on news articles or other information inferring bitcoin farms receiving favourable rates.

## C. Hardware List¶

Product Advertised Mhash/s Mhash/J Mhash/s/\$ Watts Price (USD) launchDate
FPGA 25200 20.16 1.64 1250 15300 2013-01-01
Avalon Batch 1 66300 107 52.34 620 1299 2013-02-01
BFL SC 5Gh/s 5000 166 18.24 30 274 2013-03-01
Avalon2 300000 3075 2013-04-01
ASICMiner BE Blade 10752 129 28 83 350 2013-05-05
ASICMiner BE Sapphire 336 130 17 2.55 230 2013-06-01
BFL Single 'SC' 60000 250 46.18 240 1299 2013-07-01
Avalon Batch 2 82000 117 54.7 700 1499 2013-08-01
Avalon Batch 3 82000 117 54.7 700 1499 2013-08-01
KnC Jupiter 500000 400 80 600 4995 2013-11-01
KnC Saturn 250000 400 66 300 2995 2013-11-01
ASICMiner BE Cube 30000 150 55 200 550 2013-12-01
BFL SC 50 Gh/s 50000 166 50 300 984 2014-01-01
BFL SC 25 Gh/s 25000 166 20 150 1249 2014-01-01
BFL SC 10 Gh/s 10000 200 50 2014-01-01
Blue Fury 2500 1000 17.8 2.5 140 2014-01-01
BFL 230 GH/s Rack Mount 230000 500 400 2014-02-01
HashFast Sierra 1200000 909 169 1320 7080 2014-03-01
HashFast Baby Jet 400000 909 71 440 5600 2014-03-01
AntMiner S1 180000 500 800 360 299 2014-03-01
Klondike 5200 160 260 32 20 2014-03-01
bi*fury 5000 1176 24 4.25 209 2014-03-01
Twinfury 4500 1174 20 3.83 216 2014-03-01
AntMiner U1 1600 800 55 2 29 2014-03-01
CoinTerra TerraMiner IV 1600000 1066.67 2100 1500 2014-04-01
Bitmine.ch Avalon Clone 85GH 85000 13 650 6489 2014-04-01
HashCoins Zeus 3500000 1436 2400 10999 2014-05-01
HashCoins Apollo 700000 1436 600 2499 2014-05-01
AntMiner S2 1000000 900 442 1100 2259 2014-06-01
ROCKMINER R-BOX 32000 711 500 45 65 2014-07-01
NanoFury NF2 3700 740 74 5 50 2014-07-01
AntMiner U2+ 2000 1000 115 2 17 2014-07-01
KnC Neptune 3000000 1429 231 2100 12995 2014-08-01
BTC Garden AM-V1 310 GH/s 310000 954 1003 324 309 2014-08-01
AntMiner S3 441000 1300 1154 340 382 2014-09-01
AntMiner S4 2000000 1429 1429 1400 1400 2014-10-01
ASICMiner BE Prisma 1400000 1333 2333 1100 600 2014-10-01
ASICMiner BE Tube 800000 888 2500 900 320 2014-10-01
ROCKMINER T1 800G 800000 800 2462 1000 325 2014-10-01
BTC Garden AM-V1 616 GH/s 616000 951 1760 648 350 2014-10-01
ROCKMINER R4-BOX 470000 1000 2238 470 210 2014-10-01
ROCKMINER R3-BOX 450000 1000 2250 450 200 2014-10-01
ROCKMINER R-BOX 110G 110000 917 1250 120 88 2014-10-01
Avalon3 800000 2014-11-01
HashCoins Apollo v3 1100000 1100 1000 599 2014-11-01
HashCoins Zeus v3 4500000 1500 3000 2299 2014-11-01
Spondooliestech SP20 Jackson 1700000 1545 1299 1100 1309 2014-12-01
Black Arrow Prospero X-3 2000000 1000 333 2000 6000 2015-01-01
HashFast Sierra Evo 3 2000000 909 294 2200 6800 2015-01-01
BFL Monarch 700GH/s 700000 1428 508 490 1379 2015-01-01
Black Arrow Prospero X-1 100000 1000 270 100 370 2015-01-01
AntMiner S5 1155000 1957 3121 590 370 2015-02-01
Spondooliestech SP35 Yukon 5500000 1506 2460 3650 2235 2015-03-01
Spondooliestech SP31 Yukon 4900000 1633 2361 3000 2075 2015-03-01
AntMiner U3 63000 1000 1658 63 38 2015-03-01
AntMiner S7 4860000 4000 2666 1210 1923 2015-10-01
AntMiner S7-8 4860000 4000 4050 1210 1200 2016-01-01
AntMiner S7-16 4860000 4000 7476 1210 650 2016-02-01
BitFury 16nm 10000000 10000 6500 3000 1538 2016-06-01