ASIC Resistance and the Demands of Decentralization June 28, 2020
Cryptocurrencies

By Jonathan Laird

Originally published on The Block, April 17, 2019

Quick Takes

  • The creation of specialized mining hardware (ASIC miners) is a natural and expected development as the cryptocurrency space matures.
  • Commodity hardware miners (such as GPUs) encourage Satoshi’s original vision of egalitarian mining but may increase the ease of 51% attacks due to widespread availability.
  • ASIC miners encourage investment in specific cryptocurrency networks but tend to concentrate power in large commercial mining farms.
  • Both commodity (GPU) and specialized (ASIC) miners can benefit or hinder the goal of decentralization. 

In the original Bitcoin white paper, Satoshi envisioned egalitarian mining with commodity hardware, what he described as “essentially one-CPU-one-vote.” At the time, miners could use regular CPUs to solve blocks, and Satoshi believed that using common hardware was a good idea. In December of 2009, when user SmokeTooMuch suggested on BitcoinTalk that developers create a GPU Bitcoin miner, Satoshi said,

The average total coins generated across the network per day stays the same. Faster machines just get a larger share than slower machines. If everyone bought faster machines, they wouldn’t get more coins than before.

We should have a gentleman’s agreement to postpone the GPU arms race as long as we can for the good of the network. It’s much easer [sic] to get new users up to speed if they don’t have to worry about GPU drivers and compatibility. It’s nice how anyone with just a CPU can compete fairly equally right now.

The first Bitcoin GPU miner is believed to have been created by ArtForz in July of 2010, while the first publicly available GPU miner was released in October of that year. From there, the cryptocurrency arms race continued to escalate. GPUs, with their greater parallel processing, were much more efficient miners than CPUs. But computing SHA-256 hashes did not utilize all of a GPU’s resources. Field-programmable gate arrays (FPGAs) could be programmed to compute hashes even more quickly than GPUs with fewer wasted resources, but FPGAs were more of a steppingstone to application-specific integrated circuits (ASICs). Unlike CPUs, GPUs, and FPGAs, ASICs cannot be reprogrammed. Their functions are baked into their silicon, so they can be extremely efficient at one task but useless for anything else. Once ASICs took over in early 2013, all other Bitcoin mining hardware quickly became unprofitable.

Today, the cryptocurrency industry utilizes a diverse range of hardware. Some coins rely on ASICs. Some rely on GPUs. Others use alternative consensus algorithms. For those using some form of proof of work, decentralization faces two competing interests. On the one hand, egalitarian mining like Satoshi originally envisioned encourages more people to contribute to the network. Powerful CPUs and GPUs are common in gaming rigs and high-end workstations, and the economies of scale of Nvidia, AMD, and Intel ensure that the hardware is generally available. Therefore, the cost of entry to mining is relatively low and can be a way to profit off otherwise idle hardware. One key tradeoff, however, is the risk of a 51% attack. The widespread availability of commodity hardware makes obtaining hash power easier for an attacker, which is why some of the smaller GPU-mineable coins, such as Vertcoin and Ethereum Classic, have experienced chain reorganizations.

On the other hand, the relative rarity of ASICs makes acquiring more than half the hash rate difficult for an attacker, so ASIC miners could be seen as capital investments in specific networks. Those investments encourage ASIC owners to continue participating in the network rather than jumping to whatever coin is more profitable. However, the long lead times and high prices of ASICs increase risk and tend to concentrate hash power in large industrial operations that can benefit from economies of scale.

In the last couple months, two major cryptocurrencies have made significant moves regarding ASIC resistance. On March 9, 2019, privacy-focused Monero executed a hard fork to change the proof of work. Monero developers had committed in 2018 to change the mining algorithm regularly to discourage the creation of ASIC miners. The last hard fork saw the hash rate drop from a peak of about 1 GH/s to a stable level around 270-330 MH/s. Hash rates typically drop after a hard fork because some miners fail to update in time, but the large drops experienced by Monero after each hard fork are suspicious. In February of this year, MoneroCrusher wrote an analysis in which he concluded that 85% of Monero’s hash rate was due to ASICs. According to his nonce analysis, each ASIC was likely running at 128 kH/s, roughly equivalent to about 65 AMD RX Vega 56 GPUs. Given the 70% drop in hash rate with the algorithm change to CryptonightR, the Monero community’s concerns about ASICs appear to have been justified again.

Ethereum, while still moving slowly toward proof of stake, has taken another tentative step toward changing its proof of work from Ethash to Programmatic Proof of Work (ProgPoW). ProgPoW aims to level the playing field between ASICs and GPUs by more completely using the resources of consumer GPUs. As a result, a ProgPoW ASIC would need to be closer in design to a common GPU. The Ethereum community seems even more divided over ASIC resistance than Monero’s was, and the opposing views highlight the tension between the various concerns of decentralization.

Bitcoin, as is often the case in the cryptocurrency space, led the way with the first ASICs. They are now so firmly entrenched in the Bitcoin ecosystem that a proof-of-work change seems highly improbable. Entire businesses have been built around Bitcoin ASIC mining, and those companies are unlikely to support any algorithm change. Many other cryptocurrencies have followed suit and simply accepted ASICs as an inevitable progression of technology. Dash, for example, uses the X11 algorithm, which was originally intended to resist ASICs, and Bitmain released the Antminer D3. Now Dash mining is completely dominated by ASICs. 

Proof-of-work cryptocurrencies which wish to resist ASICs have two options. They can change algorithms to make ASIC development unprofitable, which runs some risk of giving developers too much power over a decentralized system, or they can level the playing field between GPUs and ASICs. Monero chose the former. Ethereum chose the latter.

Other projects have developed some interesting alternatives. Grin, for example, uses dual proofs of work in order to discourage ASICs long enough for multiple companies to bring specialized hardware to market. Chia, the cryptocurrency project cofounded by Bram Cohen, the creator of BitTorrent, aims to use proofs of space and time instead of proof of work, leveraging the vast amount of unused storage space to make specialized hardware unprofitable.

The most anticipated alternative to ASIC resistance is proof of stake, which is a form of virtual mining. Ethereum developers have been working on proof of stake for years now, but implementation is still elusive. While virtual mining has the advantage of using practically no power, the cryptocurrency community still vigorously debates whether a network can be secured through a self-referential system.

FPGAs are also likely to make a comeback. As more projects seek to discourage ASICs, the flexibility that FPGAs offer will become important again. After any proof-of-work change, an FPGA will need a new bitstream for the new algorithm, but after that programming, the chip can begin mining again.

ASIC resistance will likely remain a divisive issue in the cryptocurrency community for the foreseeable future. Unfortunately, every option has both advantages and disadvantages. But over the last ten years, Bitcoin and the cryptocurrency industry that it spawned have demonstrated that with or without ASICs, proof of work can reliably provide a means of reaching consensus in decentralized systems.