GPU Mining Prospectus: Equihash Is No Match For RAM-Assisted ASICs
Only artificially-swapped algorithms will be left standing
Since January 2018, we’ve seen the release of ASICs for a number of currencies that were previously only mined with GPUs. This article takes an in-depth look at what has occured, and attempts to make predictions about the future of GPU mining at large.
Sia and a number of other, smaller currencies use an algorithm called blake2b. While it was possible for a long time to make a blake2b ASIC, the rapid run-up in the price of SIA made it economical to do so. Sia Labs had planned to launch their own ASIC miner: Obelisk. The Obelisk miner is now selling, but lost all its thunder due to Bitmain jumping the queue by releasing the Bitmain A3. The A3 mines Sia at 815GH/sec and costs $520. Obelisk’s product, the SC1, costs $1699. Bitmain released the A3 in large numbers, rapidly, at low cost. GPU miners on the Sia network were effectively rendered useless over the course of a few weeks.
The Sia community discussed, but did not implement, algorithmic changes that would have rendered the Bitmain A3 useless on the Sia network while preserving the functionality of their in-house ASIC, the Obelisk.
* GPUs on the Sia network have been fully displaced
* Obelisk (a Sia-linked company) hasn’t been able to sell many of its SC1 ASICs due to their relatively high cost and slow release
* The Sia community was ready for ASICs, but Bitmain managed to release a device more quickly than Sia.
* Bitmain effectively captured all of the profits to be made from the displacement of GPUs on the Sia network.
In Bitmain’s usual low-key style, they announced a CryptoNight miner capable of mining Monero. This threw the Monero community into an uproar. In Monero’s case, it was quickly decided that the currency would be changing the algorithm used to mine new coins. It should be noted that the hard-fork was successful and difficulty on the Monero network dropped. But this didn’t come at no cost to the monero community, because we now have:
* Monero Classic
* Monero Original
* Monero V
* Monero 0
…to contend with. All of these coins preserve the functionality of the ASIC miners, and derive their value from the monero transaction history. Weather this is actually a problem is maybe a tougher question to answer.
* Monero was mined for quite some time by ASIC makers, in secret. They most likely made huge profits because they had the fastest, lowest cost equipment. This is proven by the slow but very significant increase in difficulty leading up to the actual release of the miner. A similar difficulty increase occurred on Sia, Ethereum, and Zcash.
* Monero’s decision to hard-fork has proven popular with both miners and the wider user community. It’s price has increased significantly since the fork, and GPU miners are finding it to be the most profitable coin (with significant liquidity) to mine.
* Smaller CryptoNight coins are changing their algorithm, same as Monero did:
* Bitmain and other ASIC makers profited from making CryptoNight ASICs because of forks of Monero.
* Monero is the best coin to mine with GPUs.
The Bitmain E3 is the first Ethereum ASIC miner. It’s worth discussing how it’s constructed and how that relates to the design of ethereum’s hashing algorithm, in order to give the reader an idea of how similar devices will be built for similar algorithms in the future.
The Ethhash algorithm is built to require a lot of memory in order to find solutions. These kinds of algorithms are called memory-hard. Usually, moving data over a high-bandwidth bus is what limits the performance of these algorithms (and makes them hard to implement an ASIC for). Well, the Bitmain A3 is an ASIC loaded with lots of RAM. The ASIC takes care of the repetitive hashing-algorithm part that never changes, because ASIC chips can’t actually adapt to changes like a CPU or GPU can. A controller in the ASIC deals with the RAM in the ASIC and feeds instructions to the ASIC chips. The result isn’t as earth-shattering as the release of ASICs on the Bitcoin blockchain because speed is still limited by the ability to move data over a high-bandwidth bus via RAM and feed it to the processors. In comparison, a Bitcoin ASIC, which mines an algorithm called sha256, doesn’t need to deal with RAM, and so its performance is limited only by the performance of the processors it contains. In short, the E3 is a totally new breed of ASIC that integrates RAM to enable mining of memory-hard algorithms like Ethereum’s, and suffers from the speed limitations of its memory, not its processor.
Additionally, it’s notable that with or without ASICs, mining on Ethereum is set to become less profitable with Ehtereum’s upcoming move to Proof Of Stake.
Ethereum Difficulty Chart: This did not happen because people were suddenly able to buy GPUs. It is the reason people are able to buy GPUs: They’re no longer as profitable to use for mining Ethereum.
Please note that when discussing GPU displacement on Ethereum, we’re discussing a financial move with 15x the impact of GPU displacement on Monero, or 58x the impact of GPU displacement on Zcash. Ethereum mining facilities are extensive and well-established. Those GPUs will not disappear: They’re going to get put onto other chains (Monero is the best candidate) or sold to gamers. But there simply aren’t many GPU-mined chains left!
* Many in the Ethereum community demanded a change in algorithm, similar to Monero’s approach. Vitalik Buterin said on twitter that the best course of action was to do nothing.
* The E3 (and others like it that are soon to be released) are sufficiently cheap and fast enough to fully displace GPUs on the Ethereum network.
* Ethereum Classic, the leading Ethereum fork, has also declined to make a change to its algorithm.
* Because Ethereum did not change its algorithm, it was not forked like Monero was. If it had, it most likely would have forked, leading to parallel chains with each of the ERC20’s on Ethereum running on them as well.
* In addition to Bitmain, two other manufacturers are working to release Ethereum ASICs.
* According to Darius Samani, a miner and equipment reseller, the batch 2 pricing of the Bitmain E3 is seen by many as a gamble, due to Ethereum’s upcoming move to Proof of Stake.
Zcash (and its equihash brethren)
While there’s no conclusive evidence that ASICs have been mining Zcash, other than an increase in hash rate similar to that seen on Ethereum, Monero, and Sia, there is absolutely conclusive evidence that ASICs are now possible for Zcash’s memory-hard mining algorithm, called Equihash. Within the Zcash community, there has been a heated debate on ASICs:
* Whether to avoid them altogether (by changing to a new algorithm for mining coins)
* Whether to bind them to Zcash only by ensuring that Zcash’s implementation of equihash is sufficiently different from that used by other Equihash coins
* Take no action and let the market decide
Zcash’s leadership has been accused of meeting with ASIC manufacturers in private on its forums, though these rumors are unsubstantiated. Modifying Equihash to ensure that Zcash can only be mined on Zcash equipment would stop miners from switching from blockchain to blockchain to get better profits. Zookoo, Zcash’s founder, says that this will provide better security. Obviously, community members with large investments in GPU mining are very concerned about these developments, and several have stated their intention to move their hashpower to Monero.
Equihash has proven to be a fairly popular algorithm. It is used by:
* Bitcoin Private
* Bitcoin Gold
The important thing to note about Equihash in the context of this article is that ASICs are either already running on these networks, or will be very soon. Equihash was once thought to be the pinnacle of ASIC-resistance technology, but it is no match for new RAM-assisted ASICs.
* ASICs for Equihash are not an “if.” They’re a “when.”
* There’s circumstantial evidence that these ASICs already exist
* Zcash leadership is working to ensure a smooth entry for ASICs
* It doesn’t seem like Zcash is likely to swap algorithms like Monero
Grin: Introducing the Cuckoo Cycle
Grin is a promising new (unreleased) cryptocurrency that will be mined using something made by John Tromp (Yes, the “Tromp miner” Tromp) himself, called the Cuckoo Cycle. For great detail on the Cookoo Cycle, please see the link below:
Essentially, the Cuckoo Cycle is another memory-hard algorithm, that was expected to be ASIC-resistant. With the new RAM-equipped ASICs, however, it seems that even the Cuckoo Cycle will not stand the test. In the below Readme, there’s a discussion by Mr. Tromp about how the Cuckoo cycle will in fact be easy to accelerate with a RAM-equipped ASIC device. The speed limitation would still come from memory: the ASIC chips are likely to be able to solve much faster than memory as able to provide them with problems to solve.
* Grin is a blockchain that had planned to be extremely ASIC resistant
* New-style ASICs have made it apparent that Grin will fall to ASICs very quickly
* One of the best minds in POW design built Grin’s POW, and thinks that it will be implemented on ASICs rapidly
From the looks of things, the only way to be ASIC-resistant is to swap out the algorithm used for mining new coins like Monero did. With regards to the debate on centralization, it can clearly be seen that Monero’s swap requires its own type of centralization: the centralization of the coin around a leading development team. Even currencies that don’t yet exist, and were thought to be highly ASIC resistant, are now planning for the introduction of ASIC mining.
For currencies that don’t want to routinely swap out their mining algorithm, ASICs (weather memory-assisted or not) are set to dominate from this point forward.
* Today: Global collapse in demand for GPUs for mining
* Today: Cryptocurrencies that partner with ASIC manufacturers, like bytom+bitmain or obelisk+sia
* 3-6 Months: New entrants into the ASIC marketplace overall
* 3-6 Months: Teams copying Monero’s ASIC-resistance success by announcing that they will swap out their algorithm regularly. I also predict that these future coins will eventually become seen as too unpredictable, and possibly too centralized.
* 6 Months: Cryptocurrencies that include ASIC implementation information to aid the wide release of compatible ASICs (Grin seems to be moving this way)
* 12+ Months: RAM-Assisted ASICs integrated into mobile phone, laptop and desktop processors. (RAM-Assisted ASICs use much less energy)