Do ASIC chips really matter for cryptocurrencies?

Latest update time：2018-05-15

Reads：

Source: The content comes from Leifeng.com, thank you.

Editor's note: The author of this article is the chief developer of the blockchain-based cloud storage platform Sia. About a year ago, he and some members of the Sia team founded the cryptocurrency ASIC manufacturing company Obelisk. The establishment of the new company gave him a deeper understanding of the digital cryptocurrency market. In this article, he shared the relationship between ASIC chips and digital currency players.

All chips are ASIC

Many people think that computing chips are divided into three categories: CPU, GPU and ASIC, but in fact there is only one type of chip in the chip field - ASIC. Inside Nvidia, Intel and other companies, they all call their products ASIC, and the so-called other types of chips are actually just another name for ASIC.

If you use a scale of 1 to 10 to measure the flexibility of a chip, Intel's CPU represents '1' and ASICs for Bitcoin represent '10'. On this scale, the flexibility of a GPU is '2'. If designers have the ability to design chips from 1 to 10, when designers move from '1' to '10', they lose flexibility but gain performance, and of course they also reduce the workload of design and development because of sacrificing flexibility.

But the reason why we don't see products between highly versatile GPUs and completely inflexible ASICs is that usually when giving up enough flexibility and not choosing a GPU, you only have a very specific application in mind and are willing to sacrifice every bit of flexibility to maximize performance. In addition, it is much cheaper to design a completely inflexible ASIC, which is why we generally don't see products with flexibility in between.

Of course, there are also special cases of products between GPU and ASIC, such as Baikal mining machine and Google TPU, which can cover flexible applications while having better performance than GPU. The case of Baikal is particularly interesting because it is enough to eliminate the basic same chip used in a large number of digital cryptocurrency mining - GPU.

The ability of chip designers to create flexible chips, from slightly flexible to highly flexible, each with only a slight performance sacrifice, has also hindered the widespread use of ASICs.

ASIC algorithmic barriers

Many people have been pessimistic about ASICs for a long time, and for good reason. However, general-purpose computing chips like CPUs, GPUs, and even DRAM cannot really exert their computing performance to meet general computing needs, but general-purpose chips can exert their performance potential if they are only optimized for specific algorithms.

However, the vast majority of ASIC-resistant algorithms are designed by software engineers who make assumptions about the limitations of custom hardware, and these assumptions are often incorrect. Many algorithm designers do not seem to realize that the computing core and memory portion of the chip can be treated as a whole in an ASIC.

Take Equihash as an example. When the GPU performs Equihash calculations, the data must be transferred from the off-chip memory to the computing core for operation, and then the result is transferred to the off-chip memory. However, the operations that Equihash needs to perform on the data are very simple, and the computing core and storage can be combined together, which means that most operations can be performed in the same place, greatly reducing the loss of data transmission and greatly reducing the transmission time. As you can see, people have produced very efficient Equihash ASICs in the past year.

So it was not surprising when Bitmain released a powerful ASIC for Equihash. However, our internal research showed that the performance of the Bitmain ASIC was actually 5 to 10 times lower than our research results. There could be many reasons for this, but we think it is quite reasonable to expect more powerful Equihash ASICs to be launched in the coming months.

In addition, ethash (Ethereum's algorithm) has a loose design and is the most ASIC-resistant algorithm we have studied so far, so it is not as easy to design an ethash ASIC as Equihash. It can also be seen from the products on the market that many people can still use GPUs with less powerful performance very well.

It should be noted that all those who support ASICs have always underestimated the difficulty that hardware engineers need to design around specific problems with flexibility even on a limited budget. But in the end, for any algorithm, custom hardware engineers will always find a way to beat general-purpose hardware.

Hardfork Barriers to ASICs

Equihash is an algorithm with three parameters. Zcash mining is based on a specific choice of these parameters, and any simple hard fork (Leifeng.com note: hard fork, mainly refers to the compatibility issues caused by sudden changes in the Bitcoin protocol) may change one or more of these parameters. ASIC can provide a basic architecture for Equiahsh that can meet the parameter requirements of any hard fork. This means that based on this flexibility, the adjustment of algorithm parameters is not enough to challenge the chip, and ASIC will have a significant improvement in speed and efficiency compared to GPU.

Of course, many people do not realize that resilient ASICs are possible, and expect a small hard fork to destroy the value of ASICs on the network. Sometimes a small hard fork may be enough, but just as an algorithm may be resistant to ASICs, it is possible for ASICs to become resistant to a hard fork, especially when the change is small.

Secret ASICs with huge returns

Monero developers are committed to keeping the common structure of the PoW algorithm, and because of this commitment, we believe that Monero miners can beat the hard fork with less than 5 times the performance. A few months ago, a group was exposed for mining Monero using secretly developed ASICs. However, it is reported that they have been developing these secret ASICs since the beginning of 2017, and the return on investment of these secret ASICs is enough to make the group more than enough money.

It is estimated that Monero’s secret ASICs accounted for almost 50% of the hash rate for a year before being discovered, but went unnoticed. During this period, a large part of Monero’s issuance was concentrated in the hands of a small group, and a 51% attack could also be executed at any time.

Monero's hard fork seems to have successfully impacted ASICs, ASIC designers are trying to build flexibility into their ASICs, and now that Monero has announced twice-yearly PoW changes, the future may see secret ASICs with more flexibility. Since Monero's block rewards are high enough, even if ASICs only have a 30% chance of surviving a PoW hard fork, it may be worth it.

My guess is that Monero miners will build new secret ASICs and that these ASICs will be more flexible and follow the hard forks that Monero pushes every 6 months.

Other Secret ASICs

There have been many rumors of secret ASICs for other cryptocurrencies. While people who have secret ASICs tend not to talk about them, as of March 2018 we have heard of secret ASICs for Equihash and Ethash. We believe that before the release of the Bitmain Z9, all 3 different groups were working on Zcash with different ASICs.

Since miners are willing to pay millions of dollars to have products designed for a particular cryptocurrency, it is possible to have a dedicated ASIC for even a low-grade cryptocurrency, providing the owner with millions of dollars in profit. As a result, an informal underground industry has built up around secret mining. The secrecy also means that it is disconnected and operates mostly based on rumors and previous relationships. Still, it is a very profitable industry, and even with things like the Vertcoin hard fork, the setbacks suffered by secret miners pale in comparison to the rewards of success.

One thing is certain, every coin worth over $20 million has at least one set of secret ASICs currently mining it, or will have secret ASICs mining it within a few months. The easiest way to detect if you have secret ASICs is GPU returns, but as ASICs continue to infiltrate every coin on the market, this will no longer be a reliable metric as there will not be a mining farm large enough to maintain GPUs as a benchmark.

The ASIC game has become such a competitive one because the rewards are big enough.

ASIC chips are not important

Most mining startups seem to be very focused on the chip itself, but from what we’ve seen the chip is actually less than half the story, but if you have the best chip in the world you’re not going to be a competitive manufacturer.

As a miner, the goal of the day is to do as much hashing as possible for as little money as possible, and faster and more energy-efficient chips mean more hashing rate for less money spent on the chips. But miners don't just spend money on chips and electricity, they also need to spend money on PCBs, controllers, Ethernet, power supplies and power management, fans, chassis, racks in data centers, etc.

Therefore, the chip is only one part of successful mining, and failing to see this may end up losing money because of the chip. In fact, this is one of the reasons that led to the failure of Butterfly Labs. They designed a high-performance chip that could generate hundreds of watts of heat, compared to the typical power of Bitmain chips, which is about 6 watts. But Butterfly Labs had to struggle with expensive, cutting-edge, unreliable cooling systems, which made their powerful chips late to market and too expensive to operate.

Bitmain is undervalued

People tend to underestimate Bitmain, they have the most money and dominate because of economies of scale. Of course, they have the shortest time to market, have the best chip developers for cryptocurrency chips, and they have innovated in dozens of places to reduce costs and improve efficiency from places most people don't know about, which makes them dominant. Therefore, Bitmain can ensure that it is top-notch in every iteration.

Also, I think a lot of people underestimate Bitmain or think that because they play games, they can't continue without cheating. But that's not the case. They play games because they can optimize their business on the other side and get away with it. Everything they do is highly optimized. If we want to understand mining, we need to realize that it's an impressive, skilled, sophisticated entity that controls most of the mining today.

in conclusion

All of this suggests that mining is for the big players. The more money you spend, the more advantage you have, and there is no easy way to change this fact. At least in traditional consensus, a large entity that generates and controls the majority of the hash rate seems to be more or less the ultimate winner, and the best case scenario is that there are 2 or 3 major players in the same position.

While this is bad news, it is not the end of the world for Bitcoin or other cryptocurrencies. Hashrate decentralization is a good thing, but there are also many other incentives and mechanisms that can keep monopoly manufacturers in lockstep. The Bitcoin/Segwit2x situation is a good example, with over 80% of the hashrate publicly supporting the activation of Segwit2x, but the overall move failed.

There are plenty of other tools available to cryptocurrency developers and communities to counter hostile hashrate, including hard forks and community splits. ?Hashrate owners know this, so they are careful not to do anything that would cause a revolt or threaten their healthy profit flow. ?Now that we know there will be large concentrations of hashrate, we can still continue as developers and inventors to work on structures and schemes that are secure even if hashrate is funneled to a few places.

Today is the 1587th issue of content shared by "Semiconductor Industry Observer" for you, welcome to follow.