ASICs and the roller coaster for GPUs

In early April, the China-based mining entity Bitmain announced the release of a new ASIC mining rig (Application-Specific Integrated Circuit) dedicated to Ethereum mining. As one would expect, big news like this deeply shook certain parts of the Ethereum community. But how game-changing will ASICs mining really be and is there is a reason to be worried?

When ASICs were first introduced to Bitcoin mining in 2013, hash rate increased by an order of magnitude and GPUs became pretty much obsolete.

So far, Ether mining has mainly been done with GPU power. This fact has been incredibly frustrating for gamers, because it is responsible for a massive shortage of GPUs in the market.

While Bitcoin uses a different hashing algorithm (SHA256), Ethereum utilizes a memory-hard mining algorithm (Ethash), which requires hardware with a minimum amount of memory. Even though possible in theory, in practice it seems unlikely that Bitmain‘s ASIC will have significant impact on the network’s hash rate right away - at least not anywhere near the impact ASICs had on Bitcoin in 2013. Without going into details of “Moore’s Law” and the processor-memory performance gap, the required memory reads are still a limiting factor in the Ether mining process, both with GPU as well as ASIC. There is also a big sentiment within the Ethereum community, shared by Vitalik Buterin, that ASICs are not really a threat to Ethereum at the moment. Given these restrictions on memory-reads, it remains to be seen how much stronger ASICs will really be.

But when we look at other crypto networks like Monero, there seem to be indeed strong indications for more powerful hardware (potentially ASICs) that has already been utilized in the mining process. Monero (which also uses a memory-hard hashing algorithm) recently hard-forked to an ASIC-resistant code. The day after that anti-ASIC hard fork, Monero‘s hash rate dropped by around 80%, which led to the speculation that ASICs had been mining Monero for some time already.

Whether this is really the case or not is something you can probably spend hours debating and speculating. But the real question at hand should rather be why or why not it might actually be healthy and beneficial to embrace ASICS/more specialized hardware in general and maybe even find a way for both, ASICs and GPUs to co-exist in the ecosystem.

The obvious argument against ASICs that most developers in the community bring up is the threat of centralization in hardware production. ASIC hardware production is currently in hands of mainly one huge player — Bitmain. Bitmain is not only a mining hardware manufacturer, the company also operates massive large scale mining facilities as well as two of the biggest Bitcoin mining pools (Antpool and BTC.com). The combined footprint in all of these three areas clearly makes Bitmain the heaviest player when it comes to hash power on the Bitcoin network.

Bitmain’s Antminers are estimated to make up 70–80 % of all miners in the Bitcoin network, which is a level of centralization that arguably poses a potential threat to the protocol. This puts Bitmain clearly in a powerful position, giving them potentially enough (direct as well as indirect) power to reject certain protocol upgrades and it also leaves the whole network open to vulnerability in the Antminer software like Antbleed, the malicious backdoor which was uncovered about a year ago.

While it remains to be seen, whether in the long run it is healthy and sustainable for a company to operate under such measures (Jimmy Song making a good point against it), there are clearly more players entering the arena right now. Chip manufacturing is a difficult game to play though, heavily ruled by economies of scale. The bigger the company, the better their hardware, the faster they can grow etc. — which will make it difficult for new companies to grab significant market share right away. But still, Samsung recently confirmed that it is providing ASIC chips to mine Bitcoin, Ether and other currencies for hardware manufacturer Halong Mining. So this could just be the start of an arms race for ASIC hardware, as more and more giant tech manufacturers enter the crypto space.

Other than the current landscape of hardware manufacturing and the respective threat of centralization, there is not really much to say against ASICs — utilizing more specialized hardware for mining has never been bad per se and it’s probably fair to assume that such hardware will always be more efficient and cost-effective than all-purpose ones.

So, fight it or embrace it? The only effective (albeit short-sighted) way to fight it would be to hard-fork Ethereum, which does not really seem to be very sustainable as a strategy. This would end up in a “cat-and-mouse” game where the community would have to change the protocol anytime new hardware comes out. Once again, you don’t have to look further for proof than to Monero. When they added ASIC resistance to their main protocol it led to the messy hard fork of four new projects, each with the pre-fork software open to ASIC miners.

And above all, developers pushing for such a change in protocol to render a certain piece of hardware obsolete actually create, in a way, the opposite effect — more centralization.

But the most important argument for ASICs is that they make a network more secure because they naturally align incentives.

Miners, and companies running mining operations, have actual skin in the game: ASICs can mine one coin and one coin only. Having invested heavily in specialized hardware, miners have all the incentive in the world to NOT harm/damage the network (or even its reputation) just to protect the value/utility of their precious ASICs. If the value of that coin falls in price, the value of the hardware falls right down with it.

Unlike with GPUs, where you can harm one network today and go mine a different coin tomorrow (or even better, just sell your hardware on eBay), with ASICs none of these strategies would work.

With large GPU-based mining operations there is a lot of jumping from coin to coin and mining the most profitable coin du jour is a strategy that is likely to increase overall mining yield quite significantly. But for smaller networks in particular, high volatility in hash rate (caused by large scale miners through “jumping” between coins) does pose a threat as it leads to less stability and security, increasing vulnerability. This was probably a factor in the decision for Sia to actively open up for ASICs last year.

So with nothing to be said against ASICs per se, it seems obvious that the ideal scenario would be a diversified landscape with miners purchasing their ASIC hardware from various manufacturers and thus creating an ecosystem where the longterm utility value of their respective hardware is tightly connected to the value of the specific network/coin they are mining for. Obviously more specialized hardware would move mining even more away from small scale operations in private household basements to rather large scale and professionally run mining operations. However, this is a trend that the ecosystem has seen over the years with GPUs already and in the end could eventually just be the price a crypto-network would have to pay for an increased level of network security: a higher but in it’s outcome still healthy degree of centralization on the hardware layer with well aligned financial incentives on the base protocol layer which should clearly encourage the right behavior if you look at it from a game theory perspective. So in this case the financial consequences of malicious misbehavior serve as a natural safeguard protecting the network from such behavior.

Eventually this probably only shows us that even an ecosystem designed for decentralization and the absence of central power/authority, can be vulnerable to some level of centralization across its hardware, base protocol or application layer once it becomes big enough to attract large scale financial interests. And we can only hope that it is exactly the latter that will encourage further hardware players to start shipping competing mining hardware.

Until then, developers will have to ask themselves whether it is a fight worth fighting or whether to embrace more specialized hardware, maybe even find ways to co-exist with the existing hardware and from there rather focus on more important challenges ahead.

But in the end, none of this could turn out to be relevant at all, at least for the Ethereum community, if and when it transitions to Proof of Stake, which would render ASICs, along with GPU mining, obsolete. The upcoming Casper update is probably paving the way in that direction.