Why is it said that FPGA or ZK universal server has zero chance in the Aleo project?
As we all know, the Aleo project is a unique existence in the field of ZK projects, mainly due to its incentive design of PoSW and Coinbase puzzle mechanisms, which is unprecedented and very innovative.
It is precisely because of this design that the workers who are transforming Ethereum see hope. Of course, their algorithms are completely different, but machines with graphics cards can also be used. Since the transformation of Ethereum, we have not seen projects with such large financing amounts like Aleo.
In fact, Aleo’s PoSW seems to us to be more similar to a computing power calculation of AI. It is a calculation of ZK proof. Although it has some limitations, it can still play a certain role in the production of ZK proof. It is not the pure WK in the past. project.
As the Aleo mainnet approaches, we see Bitmain officially announcing plans to launch a dedicated mining machine for the Aleo project. This is actually a good thing for the Aleo project. Alex, the CEO of Aleo, also forwarded the news very happily. Of course, Bitmain is definitely not the only manufacturer that makes ASICs. It’s just that they have officially announced it and are an established leading manufacturer, which will have a certain role in promoting the market.
There is actually another machine model in the ZK field, which is similar to a server. I guess many people may not know it, and it can also be called ASIC . According to our past knowledge, ASIC machines cannot change the algorithm, like Big Pie or Ether, but this kind of server can change the algorithm, but the “change algorithm” mentioned here is only limited to the ZK algorithm, which means that it can only change the algorithm. It can adapt to ZK projects, but it cannot be as flexible as a graphics card or FPGA, so it can be called a “dedicated” ZK server or ASIC.
We know very well which manufacturers make ZK servers, both domestic and foreign. At this point, some people started to have questions, so what kind of things does Bitmain do?
The answer is non- ZK servers , which are Aleo-specific ASICs .
Why? We will do the analysis later.
Prior to this, a manufacturer, SuperScalar, officially announced the launch of FPGA machines K10 and K11, which support the Aleo algorithm. The parameter data are as follows:
We will not comment on this data for now, I believe knowledgeable friends will make their own judgment.
In the field of FPGA, we know that there are still several manufacturers doing it, both domestic and foreign, such as Ingonyama, the technology provider behind 1to.
The purpose of launching FPGA is nothing more than to create a time gap before the dedicated ASIC after entering the main network, because everyone also knows that the ultimate must be a dedicated ASIC, and wants to compete with the graphics card first. I think it has an advantage compared with the graphics card, but Whether it can compete with graphics cards is a question mark.
Let me first talk about our expectation that the WK competition after the Aleo mainnet will be divided into two stages:
- Graphics card and FPGA or ZK server
- between dedicated ASICs
Graphics cards and non-dedicated ASICs are expected to gradually withdraw from the market in half a year, unless the price of Aleo is very strong, but it is definitely not cost-effective in terms of input-output ratio and payback cycle. We know that in the end, the competition is nothing more than two factors: power consumption and electricity price . Of course, the final level of the computing power of the entire network will be determined by price. Ensuring profits is a major prerequisite.
It is estimated that there are more than 9 million N cards surviving in the Ethernet era. Of course, with the postponement of Aleo for almost a year, many graphics cards were sold, but in fact they were recovered very quickly. These graphics cards did not disappear.
So once it is on the mainnet, we expect that the computing power of the entire network will soon reach 50 billion, yes, 50 billion! You heard that right, and maybe more! We know that the testnet phase 3–2 has a maximum computing power of 800 million, which felt quite high at the time. In fact, most graphics cards did not enter the market, mainly because many factors were uncertain during the testnet period, and the startup costs of major graphics card manufacturers were too high. Although the test period is only 2 months, the electricity cost will be terrifyingly high!
Okay, based on the above background, let’s return to the topic of our article: Why is it said that the chance of FPGA or ZK server in the Aleo project is zero? Let’s analyze it from two perspectives:
1. Cost-effectiveness
Let’s talk about FPGA first. Theoretically speaking, the cost or the cost per unit of computing power must be higher than that of graphics cards (mid- to low-end). This is not calculated in detail. Different hardware technologies themselves determine the cost difference, which can be manipulated. If you look at history and look at other projects, you will know why FPGA has never become popular, but has always been dominated by graphics cards. Of course, this is talking about a newly purchased machine. If it is retained from the previous Ethernet era, it still has certain advantages. After all, the power consumption is definitely lower than that of a graphics card, but it also requires the cooperation and optimization of the algorithm. After all, the underlying technical architecture is completely different, otherwise the computing power advantage cannot be exerted, which means that algorithm optimization needs to be done separately.
Let’s talk about the ZK server. We already have some data and will not make it public for the time being. Let’s first analyze it from the perspective of the universality of the principle. First of all, although it is a server, its core is still an ASIC chip, which means that it needs to be re-architected and designed, and it needs to find a chip foundry to tape out, and then it needs to be packaged and tested. Processes all mean costs. Let’s look at another example in life. Anything with multiple functions and suitable for multiple scenarios will be more expensive. This is basic common sense. The same principle applies to ZK servers, so why are their final forms and graphics card machines Similarly, in order to achieve versatility, like the graphics card, some other common components will also be used to complete the whole machine.
At this time, you may say that graphics cards are also universal, yes, but graphics cards can indeed adapt to the needs of many scenarios. The premise is that there is demand, so that the quantity can be large. Only when the quantity is large can the cost be further reduced and the price can be lowered. This is the most basic, but is there really a huge demand for ZK servers? We don’t know yet, but we will analyze it later from the perspective of ZK proving market demand, and everyone will know.
Overall, we believe that ZK servers are similar to FPGAs. In theory, the cost or cost per unit of computing power must be higher than that of graphics cards (mid- to low-end), but the power consumption should be lower than that of graphics cards. Another one, like FPGA, also requires algorithm optimization. The underlying technical architecture is completely different. Otherwise, the computing power advantage cannot be exerted, which means that algorithm optimization needs to be done separately.
Of course, if it is a high-end graphics card, such as 4090 or above, the cost or the cost per unit of computing power may be similar, but the power consumption is lower than that of a graphics card with the same computing power. In terms of algorithm computing power, each hardware architecture needs to be optimized separately. You need to consider whether this is necessary, because it is determined by market demand.
In fact, many people don’t know that the computing power of the graphics card has been optimized to a very high level. It is not as low as everyone imagined in the 3–2 stage of the test network. Therefore, after entering the main network, it will be the first stage of the competition as we mentioned earlier. At the first stage, we think it is competing with graphics cards, but we don’t know whether it can compete with graphics cards or how competitive it is. We think it may not be as good as graphics cards (mid- to low-end).
And from another perspective, in terms of residual value or liquidity, graphics card > FPGA > ZK server, which means that ZK server is the weakest, after all, it can only support ZK projects. Unless they (FPGA and ZK servers) can open a big gap with graphics cards in terms of computing power, power consumption or price, judging from the current situation, it may not be possible.
2. ZK proves market demand
Whether it is FPGA or ZK server, one of the external selling points is that it can “dig more” because the algorithm can be changed. But let’s take a look at what other good WK projects are currently on the market that allow you to dig more? FPGA says it can mine KAS, but isn’t KAS’s ASIC machine good? Isn’t it nice to use a graphics card?
The ZK server said that other ZK projects can be mined, such as Scroll or Zks. Let’s not talk about when other ZK projects will launch decentralized proofs, and what their incentives and economic models are or whether they have them. Let’s first analyze the market demand for ZK proofs on L2 and get Zks Example.
I believe everyone familiar with L2 should know the following picture, a ranking of TVL:
We can see that Zks is №1 in the ZK Rollup technology project. Scroll was launched not long ago, and the current amount is not very large.
Many people think that general-purpose machines can mine other ZK projects, but don’t forget, this is ZK Rollup. If you don’t understand what ZK Rollup is, you can go to Baidu on your own. In other words, the ZK Rollup project on L2 is a Batch and a ZK proof, unlike Aleo’s default privacy situation where each transaction requires a ZK proof. This is not of the same order of magnitude in terms of the number of ZK proofs. Of course, the premise is that Aleo’s ecosystem can develop and grow, which we will leave for later.
The picture above is the official browser of Zks, and the screenshot is the data of Batch. One Batch has about 600 transactions, and one Batch has one ZK certificate. Interested friends can calculate it by themselves. We roughly calculated that it is about more than 70 per year. Ten thousand proofs, calculated from the current data.
From the TVL ranking picture above, we can see that the gap between Zks and the first Arb is about 14 times. Assuming that Zks is currently the first, then the total number of ZK certificates in a year is more than 9.8 million, which is 10 million. Okay, everyone has no idea about this number. Let’s compare the data of Aleo test network 3–2 and look at the picture below:
Let’s not talk about other hardware incentive plans. Currently, only Aleo, the ZK project, has a clear incentive plan and economic model. Other ZK projects have never been mentioned at all. Based on this quantity, do you think it is necessary to have a decentralized proof? To put it bluntly, a few 6-card 3080 machines may be all it takes. What else is needed for decentralization? Of course, their proof algorithm may be more complicated than Aleo’s, but it can be done with a few server-level ones. This is assuming that Zks is the first place, not to mention that the difference is still more than 10 times.
This is still the case of Zks. ZK is the №1 project in Rollup technology, let alone other projects. Although Scroll has always claimed that it wants to develop decentralized proofs, first of all we think it cannot reach the height of Zks. We must first solve the market demand. Let’s talk about the issue, otherwise it will all be empty talk.
So we have seen other ZK projects say they have ideas for decentralized proofs, but based on the current data, it is not necessary at all, and it will also reduce efficiency. In the long run, as the ZK ecosystem on L2 continues to develop and more and more technologies are implemented, decentralization will prove to be necessary, but it will take time.
Some people may say that ZK technology can be used in more other scenarios. In fact, we have analyzed this direction before, and indeed there are, such as ZKML, but these are still in the early stages and will take time. After all, ZK technology has not been implemented for many years. In the past year, although we have seen various ZK hardware acceleration activities and competitions, they are more academic-oriented. From an application perspective, it still takes time to promote and popularize, at least on Ethereum. There are not many applications yet, which is one of the reasons why Aleo first used PoSW incentives to incentivize its monopoly on hardware computing power for 10 years.
To sum up, we should analyze the data rationally and do not blindly over-YY or listen to what others say. The idea of ”dig more” is simply untenable at the moment.
Summarize
After the above analysis, everyone should be able to understand that in the current PoSW track of Aleo, we believe that only dedicated ASIC is the end game!
This is why Bitmain does not make ZK servers, but makes Aleo dedicated ASICs.
So does Bitmain have an advantage on Aleo? Some people think that they have accumulated a lot of experience in previous chip experience and are also the leading ASIC manufacturers, thinking that they will definitely have an advantage on Aleo.
These perceptions are entirely the thoughts of people who don’t know much about the ASIC industry and the Aleo project. Anyone in this circle should know that Bitmain’s past advantages were only based on pie-like algorithms, such as KAS. They Have you ever achieved impressive results on other non- big pie algorithms? No! Therefore, we will not analyze it in detail. The conclusion is that we believe that Bitmain does not have any advantages based on the current situation we have!
There is no need to compare graphics cards and dedicated ASICs. There is no doubt that just look at past projects. One of the biggest advantages of graphics cards is good liquidity, but when the price drops below 5U, the profit is gone, so what is the liquidity? When the price is higher than 5U, the investment return ratio and payback period are definitely better than dedicated ASIC.
As for FPGA and ZK servers, this is like a triathlon and individual competitions. Although they have flexibility, they will definitely lose in individual events, whether in terms of power consumption, computing power, price, or cost. The advantage is incomparable.
Some people may worry about changing the algorithm later. Looking at other PoW projects, unless they say they do not support ASIC at the beginning, who will change the algorithm casually once they are on the mainnet? What’s the point of the change? Unless it has a very significant impact! Aleo’s attitude towards ASIC has been very clear. Aleo’s CEO forwarded the news of Bitmain’s official announcement, as well as the founder and investor of the ZPrize competition. Now that Bitmain has made the official announcement, it is also a signal, which shows that they have been We are all supporters of ZK hardware acceleration, including ASIC of course!
