Rewrite the Technological Landscape

Meng Qian gave Alphahim a time for another meeting, and then went to prepare for the Gale Global Developers Conference on the third day.

On the third day of the conference, everyone saw a lot of relatively small products, which are things that most users don't usually pay attention to.

Such as ultrasonic flowmeters, hard disk burners, some automotive hardware, factory equipment, and some weird-looking robots and so on.

There are a lot of products on display, but to be honest, except for robots, which are still a little popular, most of the products are not too popular among ordinary users.

However, people in the industry are interested in talking about these products on display today and have started various research and analysis.

Because today's focus is not on the products themselves, but on the chips behind them.

Behind all the products on display today are dedicated chips.

The theme of the third day of the conference is simple and crude special-purpose chips.

After the morning's exhibition, Meng Qian appeared on the stage again in the afternoon. It can be seen that Meng Qian attaches great importance to this year's conference, which is the conference with the most frequent appearances.

"The debate between general-purpose chips and special-purpose chips has been a topic since 2012. Now this controversy is not very hot, but I want to talk about this topic this year, because I never like to wait for things to happen completely. To take it seriously, I am used to seeing trends when I look at events.

In 1969, Busicom of Neon Country selected Intel, which made the world's first general-purpose processor 4004 appear. By rewriting software, various application functions were realized, which greatly released productivity. Since then, The age of Intel had begun, and so had the age of general-purpose computing.

We can try to explore this issue from the perspective of economics. People in the field of economics know that productivity is an important factor in measuring economic growth and economic level. Since 1974 in the United States, one-third of productivity growth has been brought about by the computer.

And a new technology that gains benefits through commercial success will stimulate continuous investment in innovation, thereby further expanding production, and users will also have the opportunity to buy stronger computing power at a lower price.

Coupled with the fact that the growth in demand will continue to drive supply, such a virtuous circle has continuously promoted the progress of the general computing industry in the past few decades.

But we can look at some public data now. First of all, since 2008, the annual growth rate of the cost performance of chips has dropped from 48% to 11% today.

Looking at this data again, this is Intel's public financial statement last year. Intel's fixed costs have exceeded variable costs.

In fact, Gale Semiconductor is about to face this problem.

In other words, if we continue to develop at the previous speed, it is very likely that the more we develop, the more we will lose money. However, if we do not develop at the previous speed, the iteration speed of the industry will slow down.

But profit is the premise of business. If it loses money every year, how many years can it be paid by Intel or Dafeng Semiconductor?

How to solve this problem? "

Meng Qian pointed out this very real problem from the perspective of economics, and from the perspective of the development of the previous life, Intel’s choice is to slow down the iteration speed, which is clearly written in Intel’s financial statements in the next few years .

"I don't know what Intel thinks now. For us, under such a big background, one of the issues we have paid more and more attention to in the past two years is the development of dedicated chips.

Because special-purpose chips do not need to make any compromises in terms of versatility and compatibility, their cost performance and energy efficiency are much higher than those of general-purpose chips.

Moreover, special-purpose chips can do better in specific fields, and their performance can often reach dozens or even hundreds of times that of general-purpose chips.

But before, because the iteration speed of general-purpose chips was enough to adapt to the evolution of computers, few people paid attention to special-purpose chips.

The development of a dedicated chip requires the cooperation of ecology, otherwise the commercial road is too narrow and it will lose its commercial significance.

Coupled with the lack of flexibility of dedicated chips, new functions cannot be added after the chip design is completed, which has become a limitation.

However, the development of deep learning in the past few years has shown the market the infinite possibilities of dedicated chips.

What will be the future of special-purpose chips and general-purpose chips? It is impossible for us to consider this issue apart from the economy. The rapid development of general-purpose chips in the past 40 years is due to economic benefits.

Users can use general-purpose chips with more functions at the same price, users will naturally choose general-purpose chips, and companies will not spend this money on investing in dedicated chips.

But now, the cost of general-purpose chips has risen, and product tiering has emerged in the computer industry.

Smartphones are an obvious layered product, and the rapid development of smart phones has rapidly driven the development of AI chips.

Facts have proved that as long as there is a large enough market and high enough profits, people will not hesitate to engage in specialization.

Let's take a look at the AI ​​chip data. In 2013, the global AI chip market exceeded 8 billion soft sister coins. Judging from your data in the first half of 2014, this year is likely to exceed 12 billion soft sister coins. With mobile smart devices According to our prediction model, the market size of AI chips is likely to exceed 30 billion soft sister coins in 2016, which can basically reach one-fifth of the global server CPU scale.

By 2020, AI chips are expected to exceed 100 billion. At that time, the market size of AI chips can reach half of the server CPU market.

But how many years have the server been developed, and how many years have the smartphone been developed? This is the potential of dedicated chips, because in the professional field, the advantages of dedicated chips are too great.

Once the ecology is established, the chip will expand faster than the general-purpose chip because of its specialization, and it can bring greater benefits to the corresponding field.

If we use general-purpose chips to support the development of artificial intelligence, we may have to wait 20 years for the benefit feedback of artificial intelligence.

And all the products we exhibited at the conference today are actually raising a question to the market. The chips of these products can be specialized, so do they have enough commercial value? "

Meng Qian raised a question and started a lively discussion in the background. Meng Qian asked them to discuss for a while before continuing, "No matter who answers this question now, it is a guess, but a well-founded guess is worthy of attention, at least From our cognition, we think that the era of dedicated chips has come.

Of course, I don’t think general-purpose chips will die, just like I don’t think personal computers will die. The mature market established by general-purpose chips will not collapse so easily, but we must accept the fact that general-purpose chips will go through a period of time development is slow until a real technological revolution occurs, such as a materials revolution.

However, at the moment when it is difficult for general-purpose chips to continue to break through, no law can give us hope. On the contrary, the potential of special-purpose chips is full of possibilities. We have reason to believe that chips and data centers will gradually move towards a heterogeneous era. In the new era, more and more dedicated chips and dedicated ecology will appear.

This is the development trend of chips in our eyes. "

Meng Qian began to draw cakes for the market again, but this cake painting is still relatively responsible, whether it is the various research and development results of the company in this life, or the development momentum of dedicated chips that he saw when he passed away in 2019.

The revolution of chip materials cannot be seen in a short period of time, so general-purpose chips have to accept such a reality. Starting from the 10nm process, the development speed of special-purpose chips is getting faster and faster.

And all those who listened to Meng Qian's prediction would be able to see one thing as long as they were willing to spend a little more energy in studying the knowledge in this field.

The Hongmeng architecture is the most suitable for the development of dedicated chips.

Compared with X86 and power, in the field of dedicated chips, the absolute advantage of Hongmeng architecture will be demonstrated.

Meng Qian doesn't need to say these words on the spot, but it is self-evident that if the dedicated chip can really be developed, it will mean something to the Hongmeng architecture.

Soon the provocative media asked a question.

"The era of general-purpose chips is dominated by X86, but what about the era of dedicated chips?" "