Bitcoin mining dynasty change: Tracking S9 and S17 miners through nonce distribution map

As Bitcoin approaches its next halving, the network is going through multiple transitions. In addition to major changes in mining economics due to the reward adjustment, Bitmain’s Antminer S17 miners are also replacing the long-standing S9 series miners and becoming the network’s dominant mining hardware.

It is reported that Bitmain released Antminer S9 in 2016, which quickly became the most popular SHA-256 mining machine on the market. After several years of development, S9 still occupies a large proportion of the market.

The lack of public data on the types of mining hardware used by individual miners makes it difficult to measure how quickly this transition is happening. However, one signal source does reveal trends in mining hardware: the network’s distribution of nonce numbers. The arrangement of these arbitrary numbers, which miners include in the hash of each block, hints at how mining hardware usage has changed over the years.

In our last report on the state of the network, we looked at how we can use random number distribution patterns to spot the rise of ASICs. In this report, we will further explore the peculiarities of Bitcoin’s nonce distribution and the sources behind the streaks in that distribution to investigate recent changes in mining hardware. We will then break this data down by mining pool, providing a deeper understanding of the hardware used by a specific mining pool.

Understanding the Bitcoin Mining Process Through Golden Marbles

Mining is a key part of Bitcoin’s security model and is arguably the most significant improvement over previous attempts to create a digital currency. While mining is fairly complex, the concepts behind it are relatively easy to understand.

From the perspective of a miner, mining a block is similar to repeatedly selecting marbles from a bag without replacement. The number of marbles in the bag is very large, with a large proportion of blue marbles and a small proportion of gold marbles. When a miner takes a gold marble from the bag, he or she receives a reward.

To explain it in more technical terms: Bitcoin miners compete to find a golden random number that, when added to the proposed block header, hashes to below a certain value determined by the network's difficulty parameter. Miners search for this nonce, or arbitrary number that can only be used once, by guessing a value and checking if the resulting hash is below a certain threshold. The first miner to find that value for a valid block and broadcast it to the network gets the right to select and order the transactions in the block, a necessary step to ultimately make those transactions valid.

In return, miners receive a block reward and fees from any transactions included in the block, both of which are earned through special coinbase transactions. Assuming the properties of the SHA-256 hash function remain the same, the distribution of golden nonce for any given block is random and cannot be found except by brute force calculation.

Because the reference to the coinbase transaction is included in the block header, each mining entity is sampling from a different distribution. In other words, each entity is drawing marbles from a different bag, with the same number of marbles in the bag, and expecting the ratio of blue to gold marbles to be the same.

The ratio of golden marbles is determined by the network difficulty parameter (which is automatically adjusted by the network) and is fixed for the relevant period. Today, due to the high block difficulty and random variance, there are often no golden nonce random numbers for a particular block header. In other words, some bags have no golden marbles.

A miner who runs out of random number space for a proposed block typically increments the timestamp of the block to generate a new set of random numbers. That is, when a miner runs out of marbles, they grab a new bag full of marbles. If the timestamp has reached the point where further adjustments make it invalid, the miner must adjust the set of transactions included in the block. Similarly, if a miner runs out of bags in a room, they need to grab more from another room, which is time intensive.

To increase the probability of finding the golden marble in a fixed amount of time, miners can parallelize their calculations, which is similar to grabbing a handful of marbles at a time instead of grabbing one at a time. By using hardware suitable for the task (specifically GPUs and specialized chips called ASICs), it is possible to parallelize the search for nonce random numbers. ASICs can parallelize more efficiently than any other method.

In another form of parallelized computation, several miners coordinate their nonce discovery and agree to split any mining rewards. Groups of miners acting in this way are called mining pools, and their operators typically charge a fee that individual miners accept in order to reduce the volatility of their income.

Bitcoin Nonce random number distribution

Bitcoin's difficulty parameter is adjusted every two weeks so that, if the amount of computation performed on the network remains constant, a new block is produced every 10 minutes on average. This feature ensures that the network will continue to operate even if the computing power may change significantly. In a fully competitive mining market dominated by parallel computing miners, we would expect that the distribution graph of golden nonce random numbers over time should look like a uniform distribution. But surprisingly, this is not the case.

The non-random distribution near the left side of the graph can be attributed to mining by iteratively testing values ​​starting from 0. If a miner is mining with a CPU as an individual without parallelization and thus has no chance of colliding with other members of the pool, then this strategy works just as well as any other strategy because the distribution of nonce for each new block is independent. The disappearance of this pattern coincides with the introduction of GPU miners, which parallelize the computation.

Near the right side of the graph, there is a streaky area with very few nonces. As far as we know, this anomaly was first discovered by Twitter user @100TrillionUSD in January 2019. The area is marked below.

Shortly thereafter, a BitMEX research paper explored the strange pattern, speculating that the anomaly was due to the controversial mining optimization technology AsicBoost.

There are two variants of AsicBoost: 1. Implicit AsicBoost (cannot be observed with certainty on the chain), 2. Overt AsicBoost (can be observed explicitly on the chain). The BitMEX research team discussed both variants, but was particularly interested in the effect of implicit AsicBoost, which became almost impossible to use for non-empty blocks with the activation of Segregated Witness (SegWit) in August 2017. Of course, the researchers were unable to confirm their speculation.

In our October 2019 report, Issue 23, we took a deep dive into Bitcoin’s nonce distribution and noted a streaking pattern. Since then, the streaking pattern has gradually disappeared, and recently mined blocks appear to be more randomly distributed.

However, the anomaly in the random number distribution does not seem to be directly related to AsicBoost. Implicit AsicBoost became unavailable in 2017, and the first firmware update to support public AsicBoost was released in October 2018, but the streak in nonce distribution between these two dates is clearly visible. Furthermore, while the use of public AsicBoost remains high, this anomalous pattern is no longer visible in newly minted blocks regardless of whether explicit AsicBoost is present.

Another possibility is that the pattern in the nonce distribution may be caused by the way Bitmain's Antminer S7 and S9 miner families sample nonce numbers. This pattern may be caused by a side effect of optimization and is ultimately harmless to miners and the network.

When looking at all nonce values ​​on the network, the streaky pattern first becomes clear in late 2015, which coincides with the release of the S7 in late August of that year and the fulfillment of orders in late September.

The Antminer S9 was announced in late May 2016, with the first buyers receiving their orders in mid-June of that year. Soon after, the stripes became narrower as the S9 replaced the S7 as the dominant miner on the Bitcoin network at the time.

The recent breakdown in the pattern coincides with the transition from the S9 to the Antminer S17, the dominant miner on the network. Although the S17 was released in April 2019, miners continued to use the S9 until recently due to mining economics.

Summary Analysis

By stratifying the dataset by miner per block, we can take a more granular look at nonce distribution.

We know that miners of blocks are usually identified by tags in the coinbase data field of the block, and these identifications are voluntarily provided and may be forged. Miners do not need to leave information and can choose to leave another mining pool's tag instead of their own. Therefore, in some cases, there is even an incentive for these misleading behaviors, so we should recognize the shortcomings of this method. However, this technology has now become an industry standard, and although many miners choose not to leave identification codes, we do not believe that large-scale forgery is occurring.

Once we have categorized blocks by miner, we can incorporate this information into our graph of Bitcoin nonce distribution.

We can also look at the distribution of nonce random numbers across mining pools. Even in this case, unusual patterns are still visible. Take a look at the chart below, which shows blocks mined by Antpool and BTC.com (both owned by Bitmain), as well as ViaBTC.

The streaking pattern is much clearer in the distribution of nonce for Bitmain-affiliated pools, which suggests that these pools had a higher proportion of S7 and S9 miners during the relevant period, which is expected given the association between the pools and the manufacturers of these miners.

In 2015, the percentage of blocks mined by unknown entities dropped significantly, a result of the block size wars, during which many previously anonymous miners began to identify themselves on-chain to express support or opposition to block size increases. Today, the vast majority of miners can be identified by hash power. The streaky patterns are faintly visible in the nonce of blocks mined by unknown miners, as is their gradual disappearance.

in conclusion

The Antminer S9 has been the most used mining machine on the Bitcoin network since its release in 2016. Although Bitmain released the S17 last year, the S9 remained economical for a period of time, but given the continuous increase in the total network hash rate and changing market conditions, this model of mining machine is being phased out.

At the same time as miners have transitioned from the S9 to the S17, the streaky pattern that was previously a defining feature of Bitcoin’s random number distribution has disappeared. The origin of these mysterious streaks, which appear in a seemingly random space, has been the subject of speculation. The timing of the streaks’ visibility lends credence to the theory that the lines are an artifact of mining hardware, specifically the previous S9 and S7.

Nonce data allows us to measure the scale and speed of this shift in a way that would not otherwise be possible using public information alone. By leveraging the traces left by S9s in nonce sampling, we can estimate the proportion of these miners on the network. Separating this data by mining pool provides unique information about the efficiency of miners’ operations, which we will cover in a future report.

Network data weekly insights

Bitcoin (BTC) mining has shown signs of a healthy recovery over the past week, with the network hashrate growing by approximately 7.3%. Inefficient miners may have begun to capitulate and be replaced by more efficient ones, which is good for the long-term health of the network. For more information on miner economics and the impact of the recent difficulty reduction, see the 44th State of the Network report.

In addition, the number of BTC active addresses also showed positive signs of recovery this week, with its data increasing by 6.3%. In contrast, Ethereum (ETH) active addresses moved in the opposite direction, with a year-on-year decrease of 13.4% last week.

Network Data Highlights

Since the Bitcoin market experienced a major crash on March 12, the number of addresses holding small amounts of BTC has been increasing.

Over the past 90 days, the number of addresses holding between 1/10 billion and 1/100 million of the total BTC supply has increased by about 6%. Similarly, the number of addresses holding between 1/100 million and 1/10 million of the total supply has increased by about 4%.

Starting around March 12, the growth rates for both increased significantly. This could indicate that adoption is growing as new users begin to acquire relatively small amounts of BTC.

Data from Coin Metrics

The amount of ETH held on exchanges has increased over the period of our research, while the amount of BTC held on exchanges has decreased.

Among them, the number of ETH increased by about 5%, while the number of BTC decreased by about 3%. The decline of BTC was mainly related to the rapid decrease in the supply held by BitMEX.

The chart below shows the 30-day change in the total amount of ETH and BTC held on the following exchanges (Bitfinex, Binance, Bitstamp, Bittrex, Gemini, Huobi, Kraken, BitMEX, and Poloniex).

Of all the exchanges we studied, Bitfinex has seen the largest growth in ETH supply. Over the past 30 days, the amount of ETH held on Bitfinex has increased by about 17%, while the other exchanges have seen growth of less than 10%.

Market data analysis

Over the past week, cryptocurrency markets have begun to climb higher and are beginning to recoup the losses incurred on March 12. Our previous research found that the March 12 sell-off was driven by short-term holders and was influenced by the BitMEX liquidation spiral. There is broad consensus that this sell-off mirrors that seen in traditional markets, while the increase in Coinbase user activity suggests that retail investor sentiment does not seem to be affected.

The debate over whether Bitcoin is a "risk asset" or a "safe haven asset" continues. Some market participants have observed that Bitcoin's "safe haven" narrative has been undermined over the past month, which has had a negative impact on institutional investors' willingness to enter the space.

And by studying Bitcoin and gold, we can provide some evidence that this narrative has some truth to it and may be stronger than ever.

According to data, the correlation between Bitcoin and gold has reached an all-time high in the past 30 days.