Wuzhen, Babbitt CTO Jin Lei: Open up the power to stop the lake and promote the industrial blockchain to accelerate

On the morning of November 9th, at the "2019 World Blockchain Conference·Wuzhen" Sub-forum, hosted by Babbitt, "The Blockchain Event: New Hotspots and Explorers", Babbitt CTO banned the publication of the fusion calculation Power, release value" keynote speech.

In recent decades, the computing power of the whole society has experienced explosive growth, which has brought many subversive changes to human life, science and technology. However, the current computing power in the blockchain industry is like a dammed lake. The computing power is not open to other industries, and the value is not released. In response to this pain point, Babbitt launched the moment pool cloud platform, using computing power as a bridge, using intelligent scheduling algorithms, distributed storage and hybrid cloud deployment and other innovative technologies to provide computing services for Web users and enterprise users, and then open the block. The gap between the chain world and other industries has accelerated the landing of the industrial blockchain.

Jinlei

The following is the full text of Jin Lei's speech, Babbitt finishing:

I am very happy to have this opportunity to share with you here. I am Babel CTO Jin Lei.

I have been doing technology development in the Internet industry before, and I started to contact the blockchain industry after joining Babbitt last year. During the year, I can feel the rapid development of the blockchain industry.

I want to talk about the blockchain from different angles and how to implement the application in specific industries.

Picture 7

I want to use these two pictures to lead the next topic. Looking at the picture on the left, the friend in the room may be the first reaction: this is a hydroelectric power station, there is cheap electricity, you will think of mining, and then think about the benefits. But today I don't want to talk about mining in a narrow way, because mining is nothing more than hardware, computing power, income, etc. I want to talk about the power from a broader perspective.

Everyone said that 2020 is the first year of the blockchain landing. Under this background, how does the computing power promote the landing of this industry? How to bring more value to society? This is a question we are going to answer now.

Let's go back to these two pictures. The one on the left is the reservoir dam. The picture on the right may not be known to many friends. In fact, it is a dammed lake. What is the difference between a reservoir and a barrier lake? The reservoir is an artificially enclosed lake. It can be manually adjusted to open the floodgate during the dry season to relieve the downstream drought, and to close the reservoir during the flood season, while avoiding flooding in the downstream. Therefore, through the regulation of the reservoir, it has a great positive effect on downstream production and life.

However, the barrier lake is a lake formed by the collapse of earth and stone and is not controlled by human will. As the water level rises or the looseness of earth and stone is likely to collapse at any time, there is a great threat to downstream production and life.

In my opinion, today's blockchain industry's computing power is actually more like a barrier lake . These computing powers are surrounded by the barrier lake, which has a high water level, and these calculations are between other industries. There is very little circulation, and it is also plagued by the outside world. POW mining consumes a lot of computing power, but the income of mining today is very low.

Adam Smith has a very famous theory of free trade, which means that both parties involved in free trade can actually benefit from trade. The same theory can be applied to the flow of power between the blockchain and other industries. If we can get the calculations, it can actually bring more value to the whole system .

When it comes to computing power, it must be inseparable from Moore's Law. Moore's Law says that every 18 months to 24 months, the number of transistors in our integrated circuits will double. As early as 1971, Intel released the 4004 chip with only 2,300 transistors. The chip manufacturing process is 10 micrometers, which is equivalent to the thickness of the current hair. Today, the chips in the smartphones in your hands already contain 10-20 billion transistors, and the manufacturing process is reduced to 9 nanometers or even 7 nanometers. The number of transistors in the chip can be seen as an approximate indicator of power. We can see that in the last few decades, the entire computing power is an explosive growth. At the same time, due to the explosive growth of computing power, it has also brought many subversive changes to our lives, including society, science and technology.

Many examples around us, such as smart phones and even 4G and 5G applications, have brought great convenience to our lives. Let me talk about a few examples that people may not be so familiar with, but have a profound impact on human society.

Case 1: Gene sequencing. There are four kinds of bases in human DNA, and the thing that genetic sequencing does is to determine the order of the four bases in DNA. By sequencing the individual's genes, we can discover the genes of the disease, so that the disease can be treated and prevented early. The first complete human genetic sequencing was done in 2003, and it took 13 years to spend a total of 13 years before and after. Today, some friends may have seen some commercialization of gene sequencing products, and we only need to do a gene sequencing today in less than 2 days, and the cost is also less than a thousand dollars. This is the explosive growth of computing power in the past 20 years, including the fact that storage becomes cheaper and cheaper, so that new algorithms can be applied, greatly improving the efficiency of computing.

Picture 8

Case 2: Drug development. The three scientists on the screen won the 2013 Nobel Prize in Chemistry, and the reason for giving them the Nobel Prize was that they proposed a molecular computing model in the 1970s. What is a molecular computing model? It is by using this model that we can simulate the process of chemical reactions by computer calculation. The inherently complex chemical reactions are very time consuming. We have to do thousands of experiments in drug development, and we can imagine the time cost. Through molecular computational models, we can simulate the entire process very simply in a computer and then predict the results of chemical reactions inside. I can imagine that if we invest a lot of computing power, we can quickly complete many experiments that were very time consuming before to get the results. However, in the 1980s and 1990s, molecular computing models were not widely adopted in drug development. Why? Because the computing power at the time was too expensive.

For example: In 1985, the fastest supercomputer in the world was Cray-2. This machine had to sell 16 million dollars, and it was inflationary. The current price could be hundreds of millions. The dollar is gone. But today, the processor power of an outdated iPhone 4 is twice that of the Cray 2, and we can buy an iPhone for a few hundred dollars. It is the explosive growth of computing power, accompanied by the rise of big data, which makes the molecular computing model applied in drug development, greatly reducing the cost of research and development. Everyone can imagine that with the introduction of new drugs, the original terminal illness has become controllable, bringing great fortune to human society.

Case 3: Artificial Intelligence. This is a very hot topic in recent years. Let's briefly review the history of the entire artificial intelligence development. In the 17th and 18th centuries, Bayesian theory was proposed. Bayesian theory is the theoretical foundation of machine learning and artificial intelligence. But before the 1960s, the development of artificial intelligence was almost always based on the principle of statistics. We would set an equation to generate a fitted equation from past data and use the results of the fitted equation to predict the likelihood of future events. . Between the 1960s and the 1990s, the development of the entire artificial intelligence was almost stagnant, there was not much innovation, and there was not much exciting results. It was not until the 1990s and 2000 that scientists began to turn to computer-driven training in data-driven methods, which is what we now call deep learning. At the same time, with the advent of the GPU, it provides cheap computing power, making large-scale parallel computing possible, which promotes the rapid development of machine learning or deep learning.

Through these three examples, we can truly see that the growth of computing power has brought tremendous changes to human society and science and technology.

Although we have seen tremendous growth in computing power over the past few decades, there is no end to the human demand for computing power , or the computing power we have now is still very small. Why do you say that? We still use a few data to speak.

Picture 9

The familiar 1880 graphics card, its computing power is 11Tera Flops, Flops represents the graphics card calculated per second floating point. A 1080 state graphics card calculation is 10 to the 12th power. Some people have estimated the integration of global computing power. It is at the 21st power of 10, and the global computing power is about 11 images. Scientists have also estimated our human brain. This is an estimated value. It is in 10,000 Yotta Flops, like more than 10 28 powers. What is the concept? The power of the human brain is 100 billion times the sum of the world's computing power. The ultimate goal of doing machine learning or artificial intelligence research today is to achieve AGI (General Artificial Intelligence). The machine intelligence that you see today or the deep learning artificial intelligence can be said to be a pseudo artificial intelligence, because they can only solve a small problem in a specific scope and very specific scene. However, general artificial intelligence actually wants Let the machine think like a human being, and be able to solve different problems in various complicated environments. You can see from these figures that if we want to achieve AGI, our computing power must be at least a hundred million times more .

With the continuous optimization of the manufacturing process and the continuous iteration of the computing architecture, I believe that in the near future, our computing power will grow at an exponential level . On the other hand, our stock computing power also has a very large optimization space. It is estimated that the current high-end graphics cards in the world are in the order of 10 million sheets, and millions of graphics cards are bought and mined by miners every year. Coupled with the ASIC chip, in fact, the blockchain plays a very important role in driving the growth of computing power, including chip research and development.

In these days of meetings, you may hear a lot of words "industry blockchain", what is "industry blockchain"? The industrial blockchain is to let the blockchain be integrated with specific industries, but one reality is that friends in other industries may not know much about the blockchain, and sometimes even biased. I just wonder if we can plan to open up, let the blockchain promote the development of computing power, promote the development of other industries, and integrate with other industries from the side, let them understand the blockchain and promote the application of the blockchain.

Babbitt has invested a lot of research and development strength in this vision. Our goal is to open up barriers to the flow of computing power and bring more social value . You can watch this product at matpool.com. I am here to give two figures, so that everyone has a more intuitive feeling: today if you have a 1080-state graphics card to mine, your daily income is about 5.6 yuan. But as a scientist or engineer, you want to train the model. If you go to the mainstream cloud computing provider to rent the GPU server, you have to pay 10-20 yuan / hour, and 5.6 yuan / day. The income, the other side is the cost of 10-20 yuan / hour, we simply calculate the difference between the income and expenses between the 10-20 times. If we can circulate the power, it should be something that everyone can see.

Picture 10

In order to achieve this goal, we have made a lot of efforts, we have done a lot of technical innovation , we have developed our own intelligent scheduling algorithm. Because they are different from traditional cloud computing providers, their servers are deployed in IDC, which has a very good and stable network environment and very low latency. But the computing power on our side is connected through the public network. In many cases, the network environment is uncontrollable. Many times, these machines are connected to our platform through the intranet and through routers. These machines do not have their own independent IP. Without independent public IP, external users can't access them directly, and we have developed our own unique network penetration technology, allowing our users to access these hardware with very low latency. In the previous examples, we can feel that computing power and storage are not separated. We are pre-open-source technologies and develop our own unique distributed storage with cache.

There is also a big problem. A large part of the mining industry's computing power comes from AMD's graphics cards. But the problem is that all the mainstream deep learning computing frameworks in the world are based on NVIDIA's CUDA (Compute Unified Device Architecture) framework. To solve this problem, we also tried to get the tensorflow deep learning framework to run on the OpenCL framework. In addition to the public cloud platform just mentioned, our entire technology architecture also supports the deployment of private and hybrid clouds to accommodate a wide variety of needs.

Picture 11

There are two forms of our computing power lease : one is the C-end user, and everyone can rent the computing power through the form of the Web. In the form of a machine, users can access these computing powers through SSH and other forms. For our enterprise users, we also provide APIs at the PaaS level, including the Python SDK. Users can submit computing tasks to our nodes in the form of APIs, and do a unified scheduling with our nodes.

Finally, share some data, the moment pool cloud platform was launched in October this year, and now less than a month, there are already 1000 registered users. When they ran nearly 10,000 cards on our platform, the PaaS task submitted more than 10,000.

In the end, I want to say that the calculation of the power, the circulation between various industries, this goal is actually very big, there are many difficult problems, in this process we need to do it step by step. In the process, I also hope that all of you who have like-minded friends can join together as much as possible. Regardless of whether you have the power in your hand or want to rent the calculations to calculate, you can work with us to jointly promote the application of the blockchain and promote the development of this industry.