I took a deep dive into Equihash to look at the impact of ASICS. Here are the results.

The Equihash ASIC Takeover is Complete Barring a major market rally, Equihash<200,9> GPU mining is done. If you have a great electricity rate, you might have a few more weeks of profitability on your GPU, but margins will be next to nothing and other algorithms are bringing better returns. ASIC earnings are also plummeting, dropping 75% in the last two months. Notably, ASIC profitability today is only a quarter of what GPU profitability was at its peak 9 months ago. Equihash profitability for the last year. Daily ROI is profitability after electricity, and normalized by device cost.The data shown above is a combined result for ZCash, Bitcoin Gold (until the fork), ZClassic, and Horizen (formerly Zen). Profitability is based on instantaneous exchange to USD. A range of device capabilities and electricity rates are considered. Another key takeaway (not plotted) is that anybody who purchased a GPU at prevailing price in 2018 could not yet have payed it off by mining only Equihash coins, and now that profitability is negative, they will probably never be able to pay it off with Equihash. Of course, there are other algorithms bringing profits right now and GPU miners likely read the signs long ago. It is difficult at best to actually estimate the ASIC hashrate ratio, but following the money is easy. GPUs are gone. Security considerations ASIC advocates often conflate high hashrates with network security, but higher network hashrate only improves security when combined with a highly decentralized network. A better measure of security is the total retail cost of the devices mining on the network, called the retail capital here. Retail capital is proportional to how much equipment an attacker or coordinated group of attackers would need to collectively control in order to perform a 51% attack. For example, if the coin has a retail capital of $100M, then a potential attacker would need to purchase $100M in equipment to overwhelm the network. The retail capital is a maximum value because you must assume that the attacker already controls a portion of the existing network hashrate. So the minimum equipment investment needed for a 51% attack is actually half of the retail capital. The attacker’s investment can be brought down further by access to equipment below retail. In the “malicious chipmaker” scenario, a chipmaker covertly manufactures ASICs and uses them to attack an unsuspecting network. To gauge what a malicious chipmaker would have needed for Equihash, profitability has been projected backwards to April, which is about when rumors of Equihash ASICs became consistent. The ASIC retail capital on the release date of the Z9 Mini was $62M. Of course, there weren’t $62M worth of ASICs on the network at that time, but that’s how many would have been required to hash at the network hashrate on that day. As an exercise, if Bitmain has a profit margin of only 20%, then they would have needed a 0.5x0.8x62M = $25M investment to mount a pre-release 51% attack on ZCash. I know of no evidence that any chip manufacturer has engaged in this type of behavior, but there does seem to be a common belief in the community that Bitmain does some heavy profit mining before release. Figure 2: Retail capital for ZCash and Bitcoin Gold.The data shows that the introduction of ASICs substantially lowered the ZCash retail capital, although it is rebounding quickly. ZCash has the highest miner payouts of the Equihash coins, and currently boasts a $211M retail capital. For ZCash, then, a certain level of defense is due to its Equihash dominance. Smaller coins do not enjoy the same benefits. Bitcoin Gold is interesting because it forked to the ASIC-free <144,5> variant shortly after the Z9 Mini was released. By forking, BTG immediately saw a nearly 6-fold increase in it’s retail capital, from $2.9M to $16.8M, with another few million dollars added in the following week or so. The Bitcoin Gold team, still reeling from a 51% attack in May, seems to have made the right choice for the coin’s security. Since Bitcoin Gold was the second-highest paying Equihash coin, their departure also leaves smaller Equihash coins even more vulnerable. If less than a couple million dollars worth of devices can launch an attack on virtually any Equihash coin but ZCash, then an existing ZCash pool with 1% of the network hashrate could easily turn an attack. In short, ZCash is probably secure, but only due to their dominance among Equihash coins. ZCash dominance and the withdraw of Bitcoin Gold from revenue pool leave smaller Equihash coins vulnerable. Based on the data, all Equihash coins except for ZCash should switch algorithms immediately for security purposes. For the sake of decentralization, ZCash should also fork, but because they have fewer security concerns, and because they are tech leaders and models for the industry, they can take their time and do it right. Energy efficiency ASICs do rule in at least one metric. The energy usage of the network has dropped considerably. ASICS are at least 10 times more efficient than their GPU counterparts. Assuming that the network hashrate will level off in the coming months, then the Equihash coins will have collectively lowered their energy burden by a factor of 3 or more over GPU energy use before ASIC release. Figure 3: Combined daily energy use of ZEC, ZCL, and ZENIf ASICs are 10 times more efficient, you might ask why the energy is only reduced by a factor of 3 or 4. The answer will take a little math, but I’ll save the gory details for the methodology section below. If you look at figure 4, you might be able to convince yourself that there seems to be a minimum level of profitability that the network will level off at. This makes sense, because as profitability drops, people will stop buying hardware, so it’s a self-regulating system. After each major rally in the past two years, the profitability leveled out at about 0.5%. It looks like even GPU miners won’t sink lower than somewhere around 0.5%. Figure 4: ZCash daily ROI curves since December 2016So if we assume that there is a daily ROI of about 0.5% to which the market will self-regulate over time, it turns out you can calculate how much electrical energy the network will burn using (see methodology for exact equation) where C is a factor dependent on some network parameters and the electricity rate, which can be estimated. So if you want to lower the energy burden on the network, you need to raise the device factor, which requires either raising the cost of available devices (not typically what miners want), or lowering the device power. This seemingly obvious statement has some unexpected results. For instance, you might expect that since a GeForce GTX 1080 Ti has much better power efficiency (3.7 H/J) than a GeForce GTX 1050 Ti (2.4 H/J), the network would use less energy if everybody used a 1080 Ti than if they all used the less efficient 1050 Ti, but that is not the case. Using current prevailing prices, which are back near MSRP, the device factor of the 1050 Ti is actually 12% higher than the 1080 Ti, meaning a network of 1050 Ti’s actually uses less electricity for the same ROI. So then do ASICs have better device factors than GPUs? A high-end ASIC has a device factor of about 9 right now, where a high-end GPU is below 3. That’s why the network energy usage only dropped by a factor of 3–4 even though the efficiency is 10 times better. The power efficiency of ASICS has an effective upper limit governed by physics and manufacturing technology, and the device cost has an effective lower limit governed by things like raw material costs. If we pretend that competition has already pushed these factors to their limits, then ASICS appear to be better for energy efficiency by a factor of 3–4. In reality, there will probably be at least some improvements, so this will be a little higher. Equihash then and now Equihash was introduced by Cryptography researchers from the University of Luxemborg in early 2016. Their seminal paper, which uses the acronym “ASIC” more that twenty times, declares that Equihash is an “ASIC- and botnet-resistant PoW with extremely fast verification and very small proof size.” The ZCash team quickly adopted the algorithm. In a blog post, Zooko didn’t mince words when explaining their reasoning. Equihash is a memory-oriented Proof-of-Work, which means how much mining you can do is mostly determined by how much RAM you have. We think it is unlikely that anyone will be able to build cost-effective custom hardware (ASICs) for mining in the foreseeable future…. Nevertheless, we can’t know for certain that Equihash is safe against these issues, and we may change the Proof-of-Work again, if we find some flaw in Equihash or if we find another Proof-of-Work algorithm which offers higher assurance. — ZCash CEO Zooko Wilcox-O’Hearn - April 15, 2016 ZCash has now voted to migrate away from Equihash, but not until late 2020. Other coins are being more proactive. As described above, Bitcoin Gold forked to Equihash<144,5> shortly after the Z9 Mini was released, and remains popular and profitable among GPU miners. It is also now confirmed that ZClassic and Bitcoin Private will also be adopting the <144,5> variant as well. This catch-me-if-you-can approach to ASIC-resistance is almost identical to the strategy that Monero and other Cryptonight coins used in April to shake off their own threat from Bitmain. Some sources reported that Bitmain then attempted to dump their hardware at a discounted price without informing customers that the chips will be virtually worthless within weeks. The ASIC-resistance of Equihash relies on “memory-hardness.” While <144,5> is more ASIC-resistant than <200,9>, if enough coins adopt it, a chip would almost certainly be developed. What the Bitcoin Gold team has done is more important as a symbolic gesture than as a hardened defense. It does buy some time. More importantly, though, is the demonstration of how easily a coin can shake off the threat of ASIC incursion, with GPU miners hopeful such rebellions will result in an unwillingness for ASIC manufacturers to chase the moving targets. There are other defenses as well. Bitcoin Interest(BCI) is a newer coin that initially chose Equihash for it’s PoW. When ASICs appeared, BCI switched to the new and untested ProgPOW algorithm. On a personal note, I would say that the fork was handled in a skilled and professional manner. The team kept users updated and answered questions continuously through their Discord channel. From a more technical perspective, the fork went really smoothly, and BCI is among the most profitable coins for GPU mining right now, so… ProgPOW is promising. It was developed by the ifdefelse team which includes Kristy-Leigh Minehan, the creator of the ETHLargementPill. ASIC-resistance is achieved because the algorithm uses virtually all parts of the GPU. If an ASIC manufacturer wanted to create ProwPOW hardware, they would essentially need to clone 80% or more of the GPU, which is the opposite of what ASICs need to do. The efforts of ifdefelse and BCI have not gone unnoticed, with discussion of ProwPOW heating up in a recent Ethereum Core Dev Meeting. So should everybody switch to ProgPOW? Maybe. Regardless of the answer, GPU miners can take some comfort knowing that there are new and promising avenues of ASIC resistance being researched. Does it matter? For coins such as Bitcoin, ZCash, Ethereum, and Bitcoin, which dominate their algorithms, ASIC-resistance might not matter much. For newer and smaller coins, though, ASIC-resistance might be critical for security. Now take a wider view and ask what’s best for blockchain as a whole. What does the blockchain ecosystem need in order to boost the formation of a healthy, benevolent, and transformative industry? One of the most consistent opinions you’ll hear is that blockchain needs mass adoption. GPU mining has been the first taste of blockchain for many. Tens of millions of gamers own GPUs, and the investment is not cost-prohibitive for most. GPU mining provides a consistent cultural pressure, like a blockchain missionary spreading the good word. It’s quite the tragedy that gamers can’t mine Bitcoin anymore. Does anybody believe that crypto is better off now that gamers can’t earn Bitcoin while they sleep? On the other hand, ASICs can drastically lower the power requirements for PoW blockchains, which needs to be part of the conversation. For the foreseeable future, their will exist two classes of PoW. In the long run, though, can the ASIC dodgers continue to stymie the voracious chipmakers and keep alive, at least in part, an idealist’s vision of the blockchain? Methodology The network hashrate is pulled directly from the respective blockchains using a Python script and the BlockNinja module. The daily hashrate is calculated using the chainwork and timestamp associated with each day’s first and last block. Daily coin prices in USD are the daily closing market values from coinmarketcap.com, which tracks hundreds of exchanges and provides weighted averages. Table 1. High and low profitability device parametersThe actual parameters used for the daily ROI ranges are given in table 1. Additional equipment overhead is not considered. “Low” and “high” sets of parameters result in the bottom and top lines on the daily ROI chart. Only ZCash is considered in the calculations of energy usage. A device-independent method of calculation is the goal. First, the device cost is normalized against the hashrate to give a relative cost, ρ. The requisite gross daily earning to achieve a given daily ROI (R_d) can then be calculated. So a device with a given relative cost and power efficiency needs to earn at least the gross daily earning (E_g) to reach the requisite daily ROI. The total payout of the network, Q, is fixed by the exchange rate, block time, and block reward as The total number of the devices on the network can be calculated two ways. One by dividing the network hashrate by the device hashrate. The other by dividing the total daily fiat rewards by the minimum gross device earnings associated with the assumed daily ROI. Setting the two techniques equal, and recasting gives where H_n is total network hashrate and is now independent of any particular device hashrate. Total daily network energy usage in Watt-hours is then given by Connect with the Raven team on Telegram