This story was told by one of the leading developers of Sia, cloud-based storage of the blockchain. A year ago, he and several members of the Sia team have created Obelisk, a company producing ASIC miners for mining cryptocurrency. Their first “ashiki” will begin to leave the Assembly line in two months. The story with the immersion into the world of cryptocurrency mining next first-person.
One of the reasons we launched the Obelisk, was that we felt that developers of coins in General, very poorly understood world of mining, and the best way to understand it would be to get your hands dirty and bring to market a miner.
Since then, as we launched the Obelisk, we learned a lot about mining, GPU, ASIC, FPGA, ASIC-resistant, mining farms, electricity and many other things, as you would know the developers of cryptocurrency. We won’t tell all that we know, but share information that I believe will be useful for designers of cryptocurrency and other members of the cryptocurrency community.
What is ASIC-resistant?
Initially, we are extremely suspicious of such thing as ASIC resistance (or ASIC-resistance), and journey into the world of equipment has strengthened our suspicions. Hardware is an extremely flexible thing. Universal computing devices such as processors (CPU), graphic processing units (GPU) and even DRAM, having great potential which is underestimated, so that they can be used for General computing tasks. Basically, when designing hardware, most algorithms can be significantly optimized by just removing this whole community and focusing on a specific task.
The vast majority of ASIC-resistant algorithms have been designed by engineers who started in their assumptions from the limited custom hardware. And these assumptions usually are wrong.
When a new ASIC for Equihash?
Equihash is perhaps the easiest target, because many people were firmly convinced in the algorithm. Today we have a year like know how to do very effective ASIC under Equihash.
The most important is to do the sorting memory. Many designers are not aware of algorithms that ASIC can combine computing and storage elements of the chip. When the graphics card with the GPU performs calculations on Equihash, she has to go all the way to memory outside the chip to deliver the data to the computational kernel to manipulate these data, and then send the modified data back into memory.
The manipulations that you need to exercise with the data in Equihash simple enough so you could just combine the memory and computation together and thus produce a large part of manipulation in one place, which significantly reduces the amount of energy needed to move data back and forth, and also reduces the processing time. Which in turn increases the efficiency and speed.
Needless to say, we weren’t surprised one bit when Bitmain has introduced a powerful ASIC for Equihash. Actually, the “asik” Bitmain less productive (5-10 times) than it should be, based on our internal research. That could be a lot of reasons, but in General it is reasonable to assume that in the coming months there will be a more powerful ASIC-system for this algorithm.
We also had the outline for Ethash. It is clear that Ethash not so easy ASIC, as Equihash, but as we see in the case of existing products, outdated GPU, it is possible to get around. Ethash this is definitely the most resistant to “isicam” algorithm we have seen.
In the end, you can always create special equipment that will work best General purpose machinery. I can’t not to mention that everyone with whom I discussed ASIC-resistance, underestimates the flexibility with which engineers hardware bypass some problems, even with limited budget. For every algorithm there is always a path that can elect engineer and exceed the common equipment. This is a fundamental limitation of General-purpose equipment.
Resistance to hardforum
Many people believe that computing can be divided into three categories: CPU, GPU and ASIC. Although this is the category that in General the public sees in the world of chips there is only one type of chip: ASIC. Inside of Nvidia, Intel and others call their ASIC products. That underlines how flexible ASIC essentially.
I would like to appreciate the flexibility on a scale from 1 to 10. At one end, where 1, we will deliver Intel. On the other end put bichenovii ASIC. Designers are able to create chips that will be placed all over the scale. As you move from 1 to 10, you lose flexibility, but get the performance increase, and substantial. You also reduce the number of required efforts for design and development, sacrificing flexibility. On this scale the graphics processors will have the flexibility of 2.
Generally speaking, we don’t see designed products that fall somewhere between the GPU and it is not flexible ASIC, because usually by the time when you’ve had enough refused flexibility to move away from the GPU, you already have a specific application in mind and you are willing to sacrifice the remaining flexibility for the sake of maximizing performance. In addition, it is much cheaper to develop it is not flexible ASIC, which again explains why not so many products stretched in the middle of the scale.
Two examples of products between GPU and ASIC miners it is the Baikal and Google TPU (tensor processor Google). It was chips that can encompass a flexible range of applications, being much more efficient than GPU. The Baikal case is particularly interesting because he is good enough to make GPUs obsolete for a large number of coins while still using the same chip. These chips are also flexible enough to catch up and hardforce.
Strategy to make hardforce to turn off the ASIC of the network loses power with each use, because the designers of the chips have the ability to make chips, which are flexible and slightly flexible, very flexible, and every increase in flexibility results in a small performance hit. Monero developers were forced to maintain the overall structure of the algorithm PoW and we therefore believe that it is possible to make a Monero miner, able to survive hardforce and lose performance not more than 5 times.
Equihash is an algorithm with three parameters. Zcash for mining happens to be one particular selected option and any naive hardwork from zcash for order to reset the ASIC is likely to include changes to one or more of these parameters. We were able to come up with the basic architecture Equihash ASIC, which could successfully follow hardforum choosing any set of parameters. The point is that the underlying hardwork changing the parameters of the algorithm will not be able to break our chip, we need a more fundamental change. Despite this flexibility, we believe that our ASIC will be much faster and more efficient than the GPU. We have not attracted funding for Equihash ASIC, so our designs were left to gather dust on the shelf.
However, the findings once again confirm the ability of the ASIC. I think there are many people who do not understand that the flexible ASIC possible, and I think that small hardforce can turn off the “asik” from the network of miners. Sometimes it can be enough, but just as algorithms are trying to be ASIC-resistant, ASIC might try become resistant to hardforum, especially when the changes are minor.
Secret “ashiki” Monero
A few months ago it was revealed that for mining Monero ASIC was developed in secret. My sources say that these secret ASIC mainily from the beginning of 2017. Almost a full year of mining was held secret until the opening. ROI of these secret “Asimov” was huge, and a group of creators got more than enough money to try again with another ASIC-resistant coin.
Estimated secret “ashiki” Monero holding more than 50% of hash rate yet nearly a year before the opening and no one noticed. During this time, huge reserves of Monero concentrated in the hands of a small group and attack 51% could be implemented at any time.
Hardwork Monero successfully shook off “ashiki”. I do not think that ASIC designers have tried to add flexibility to its “isicam”, but now that Monero has announced the biennial change in the PoW, we can see another round with the advent of the secret ASIC with more flexibility. The award for the block Monero is quite high, so even if 30% of “Asimov” will survive hardwork PoW, to create resistant to the hard forks ASIC still makes sense.
I can assume that there will be another generation of secret “Asimov” Monero, and these “ashiki” will be more conservative and flexible to keep pace with the fork that Monero will be rolling out every 6 months.
Other secret ASIC
Rumors about the secret ASIC is full. People who keep secret ASIC, not inclined to talk about them too much, but in March 2018, we learned about the possible existence of ASIC for Equihash and Ethash, as well as for many smaller coins. We believe that three different groups actively mainile zcash for different ASIC Bitmain before the announcement of the Z9.
We know about mining farms that are willing to pay millions of dollars for exclusive access to the projects of “Asimov” under certain cryptocurrencies. Even a small coin with a tiny market capitalization could bring millions of dollars of profit for someone with exclusive access to the secret machines for mining. Accordingly, around the secret of going to the informal underground mining industry. All this is strictly confidential, and profitable, and even hardforce can’t overshadow the joy of secret miners.
At the moment it is possible with high probability to assume that every coin on the basis of Proof-of-Work pay for a block of more than $ 20 million dollars last year maynila at least one group secret “Asimov”, or will be mine in the next couple of months. The easiest easy way to detect this — return of cards. But as ASIC continues to pour into every coin on the market, it will gradually cease to be a reliable metric, because there will be more coins, which work exclusively of the graphics card.
The game in the ASIC become so popular, because at stake is a lot of money. Even small coins can be worth tens of millions of dollars, which is more than enough to justify the production cycle with high risk.
Influence on the manufacturer
Manufacturers, publicly sell ASIC, Bitmain seems less susceptible to such things as hardforce for ASIC than users. For example, Sia, we estimate that Bitmain brought the A3 to the market less than $ 10 million. 8 minutes after the announcement of the A3 Bitmain has already received more than $ 20 million of sales for equipment for the development of which has spent $ 10 million. Bitmain recoup their investment before the miners come to their cars.
In this case, hardwork did not affect Bitmain. Bitmain got money on Sia and the developers there’s nothing you can do. It was the same with Monero miners that Bitmain announced. Bitmain did not represent the miners, while Monero is not announced hardwork, and yet sold enough old equipment to consumers, offsetting costs and making a profit.
The game in mining is heavily dependent on manufacturers. They get control over production equipment and know about the state of the industry more than anyone else. The miner’s profitability is largely dependent on variables that the manufacturer controls, without telling anybody.
In the case of miner Halong Decred we saw that they “sold out” with an unknown lot size of 10,000 dollars. It was then noted that more than 50% of the reward miners have gone to another address that is associated with Halong, meaning that they kept the most part can see and arrived at home. Our research equipments for mining suggests that the total cost of the production equipment is less than $ 1,000, meaning that anyone who pays $ 10,000 for it, the manufacturer makes a huge profit, allowing him to make 9 more pieces for myself. In addition, the buyer has no idea how many miners have been sold and what will be the difficulty upon arrival the “asika”. The buyer trusts to the manufacturer.
If cryptocurrencies like Sia was monthly remuneration in blocks of $ 10 million, and on the shelf are “asik” capable mined 120 million dollars, one would expect that the company will earn $ 120 million on their “asiko”. But from the manufacturer there is a way to earn a lot more.
In the case of A3 Bitmain, a small party of miners was sold to the public with very fast shipping time, say, less than 10 days. Soon after YouTube started to go live with people who bought the miners and quite a hit over 800 dollars a day on them. This gave rise to the mania around A3, and the second batch of Bitmain sold much more successful.
Although we do not know how many were sold pieces of A3, we expect the profit that the company received from sales of the second batch was much higher than the possible reward for the block from mining with the use of the party A3. It turns out, Bitmain sold mining equipment worth hundreds of millions of dollars, knowing that the reward for the block is not large enough to discourage users have the money, even with free electricity. And this is not the first time something like they did with the miners Dash. We call it “flood” (flooding) and this is another example of the dangerous asymmetry that exists between producers and consumers.
In the end, equipment manufacturers for any mining company to sell machines to print money. Well-funded company, maximizing profit, you will sell print money machine only for the money that exceeds the profit from self-printing. The buyer should understand why the manufacturer sells to the miners, not the mines itself.
There are several reasons the manufacturer, it would be advisable to sell the machine for printing money and not printing them myself. The first is the capital. Manufacturing is an expensive process that requires time. If the manufacturer does not have enough money to create their own equipment, it makes sense to sell this equipment, and the money from the sale to send to production. It all comes down to what the manufacturer sells future profits to make a revenue today, and it is quite a common transaction in the financial world.
Another reason why the manufacturer can sell a typewriter for money instead of their use — energy costs. If the manufacturer had certain agreements on electricity, somebody else, with cheaper electricity or data center better, would be willing to buy the equipment at a price higher than the estimates by the manufacturer. Of course, most manufacturers have access to a good prices for electricity, and if your electricity is not free, you will hardly pass.
Finally, the manufacturer may be other reasons why he might want to make quick money instead of make long-term investments in equipment. But this does not apply to the mining of cryptocurrencies, because the life of miners is less than two years, and for business it’s not a very long period for investment return.
The flexibility of the manufacturer
In the world of the production of traditional chips for everything, from the start of development before the chip leaves in 2 years. In the case of miners Sia and Decred it took us 13 months from start of project to receipt of the product. If we did this now, I think we would be gone 9 months.
A huge amount of time was spent on special tracing of the chip. There is a more rapid process razrabotki, placement and routing, which reduces the development time of 3 months, but produces chips, which are 2-5 times slower custom. We think that if we used the methodology of placing and tracing, we would have given a finished product in 6 months.
We think that Bitmain has spent about 5 months on creating a miner A3, and Halong — 9 months for a miner B52. Both companies, in our view, used a method of locating and tracing, given the relatively weak performance of their miners.
This is the time frame to create the chip from scratch. If the goal is to chase hardforum, the time frame will be much narrower. If you know in advance that you will need to redesign the chip, there are many short paths that reduce the overall time needed to enter the market. Change the design to match the changes will take less time than working from scratch, and a good team with a well-planned basic architecture is done in two weeks. The new chips will appear in 40 days. Then they will need to pack it another week, and send it to the manufacturer for Assembly. Then you will receive your miners and start mine.
If all the seals, parts and everything else is ready in advance to upgrade the chip to adapt it for hard forks, will take about 70 days, at least in theory. In practice, Bitmain will need 3-4 months to adapt the existing chip to the hard forks, and if the print will not be reserved, need 4-5 months. Any company that is not associated with Bitmain, can add another 30-60 days.
Economy of scale
Some people already understand the situation with the economy of scale. The more money you spend, the more effective each dollar. This effect is supported at every level of scale, including the scale of billions or tens of billions of dollars.
The simplest manifestation of such an economy, the volume of orders. If you order a hundred thousand of radiators, the price will be one. If a million — the price will be better. The larger the scale, the lower the price. This effect affects almost all parts of the hardware industry and is born because manufacturers reach a point where they can buy and allocate the equipment for your order, and then maintain 100 percent efficiency of this equipment. When scaling you get better customization and specialization in addition to cost savings, which means your products become more efficient and cheaper.
At a certain point makes sense just to buy all the power the manufacturer. A large percentage of production costs accounted for production equipment. Equipment that is idle 50% of the time, will give a two times lower efficiency of costing detail than equipment that works 100% of the time. Increasing volume of orders, you load the equipment high, which again reduces cost.
Obviously, someone needs this equipment to produce. If you massturbating until you have a certain hardware manufacturer starts to produce conveyors specifically for you and loads your equipment 100%, and thus it becomes cheaper.
And this is only the beginning. At each stage, every operator, manufacturer, and so forth, all take a margin, throwing 30% depending on volume of orders. If you have enough money, you build vertical integration or buying them and cutting off the margins, or creating your own.
Hardware goes through many stages. There is a stage of acquisition of raw materials such as iron and oil, processing of these materials, the subsequent production of base parts, which are sold for common products. These basic parts are often ready for 6 months or more, and that means providers always have large stocks, which allow you to supply parts to customers. Each step usually introduces the mediator and inefficient, particularly because the total path run on the track specific products. If you have a product that has sufficient scope and scale to justify a dedicated supply chain, you’ll bump costs, improve product quality and production efficiency, and will also be a step ahead of the others.
To numbers every time, imagine that every time you ten times exceed a certain amount, you can save 30% on each part. For example, if you spend $ 100 million on miners, you can get them for $ 500. Spending $ 1 billion on miners you can reduce the price to $ 350 per miner, just by increasing the amount. But if you get to $ 10 billion, the cost per miner may fall to $ 245. While your miners are not just cheaper, but also more productive.
When we started with the Obelisk, we get various sources that warned that Bitmain’s playing dirty and that if we try to produce in China, we will stop.
With this in mind, we picked up the ears of all with whom he worked, and carefully contacted the American manufacturer that had facilities in China. It was attractive, because the prices were half what we would pay, producing in America, and the production promised to be the largest item of our expenses.
We have made every effort to hide the relationship of the manufacturer with Obelisk, removed the manufacturer’s name with any public data, behaved very cautiously. The orders were made through a separate entity.
Then, one Friday night, our producer contacted us and without much warning or explanation said that can’t produce parts for us. As we said, our attempt to produce in China ended in failure. I believe this failure has cost us about $ 2 million.
We have absolutely no evidence that there was Bitmain. We had relationships with other companies that confirmed that they had experienced similar, but had no evidence that this involved Bitmain. I don’t even know whether it was necessary to include this part in the story, because we have nothing except a few warnings, which resulted in the form of a valid notification.
But the industry is well know that Bitmain playing dirty. This is known to all. We knew that they have made and will make attempts to influence our supply chain so that we did not succeed. And the same will be with all their competitors.
Mining farm is perhaps the one area where economy of scale does not work. Good electricity prices are given on small amounts, are distributed worldwide and have unique conditions. Large companies are finding it difficult to build a system that can collect cheap electricity around the world. However, the cheap electricity and data centers keep small companies that individually do not have large amounts of no electricity, no can see.
From what I could dig up, the average professional mining farm pays somewhere in the 4-6 cents for electricity, and then another 3-6 cents for management and maintenance. The average mining farm cost $ 50 per kilowatt per month. With the improvement of technology and the growth of the industry, we can expect that this number will be closer to 35 dollars per kilowatt per month (including maintenance, land taxes, etc.) to 2019-2020 year. Hardly anyone who pays more than $ 80 per kilowatt per month, will remain competitive unless the cryptocurrency will not suddenly rise in price next year.
20% of miners have, apparently, paid less than $ 35 per kilowatt and another 5% — less than $ 20. According to my estimates, if the price of Bitcoin will fall significantly, these miners will remain in the case, and everyone who pays 50 dollars and above, they will have to leave.
It is difficult to understand where is Bitmain, but based on what we saw, it is possible to estimate the mark was somewhere around $ 30 per kilowatt per month.
The chip is not important
Most mining startups seem incredible attention is paid to the chip. But as far as we could understand, the chip is not even half the story. Chip important, of course, but if all you have is the best in the world chip, that alone will not make you competitive as a producer.
The miner what purpose? Make as many hashes, and will pay for it as little as possible. A faster chip means that you need to spend less money on the chips to get the hash rate. And more energy-efficient chip means that you need to spend less money on electricity to get the hash rate. But you spend money on chips and electricity. You spend them on power supplies, controllers, ports like Ethernet, housing, sockets, fans, shelves and more.
The chip is only part of the equation in successful mining. If you do not see the whole picture, you just lose money. Thus died the Butterfly Labs (and others) — they made high-performance chip that produces hundreds of watts of heat. For comparison, the Bitmain chips give six watts each.
Bitmain is impressive
People underestimate Bitmain. Yes, they have plenty of money, and Yes, they dominate the economy of scale. But they also dominate because the market first. They dominate because they have the best developers chip in the world of cryptocurrency. They dominate because innovate in dozens of places, to cut costs and improve efficiency, where people do not even suspect. They hire the best and pay them well. And so they are on top.
What can I say? I think people underestimate Bitmain or suggest that they play dirty, because that might otherwise. But it’s not. They play dirty, because this is another way to optimize their business and because they know what they want. Everything else is already optimized. If we want to understand mining, we must accept that the entity that controls most of the mining today is powerful, impressive and highly qualified, for an entity.
The main conclusion that can be drawn from all this is that mining is for big players. The more money you spend, the more your advantage, and change this equation is not so simple. Sooner or later there will be a great agent who produces and controls the vast majority of hash rate yet. But I don’t think that in the next few decades will appear a lot of manufacturers that produce relatively competitive miners. The production somehow leads to centralization.
And although the news is discouraged, it’s not the end of the world for Bitcoin or other cryptocurrencies on the basis of the Protocol Proof of Work. Decentralization can see is good, but there are many other mechanisms that retain a monopoly among manufacturers. For example, the situation with Bitcoin/Segwit2x. More than 80% can see openly supported the activation Segwit2x, but the movement as a whole has withered.
There are many other tools available to developers and communities, cryptocurrencies, allowing to deal with a hostile hasraton, including hardforce and splits of communities. The owners can see and know about it so try not to do anything that could prevent a healthy flow of income. And now, when we inevitably move to a centralized Hasrat, we as developers and inventors are forced to play by the new rules and new conditions. We are all in one pot, even if you remain sparse.
Subscribe to our channel in the Telegram to be aware of. Is not yet know.