Riding the wind of rebirth

Mai Xiaomiao didn't care about Zhou Zhi's idea: "First of all, there is the problem of Xiaozhi's current computing power. Previously, we only considered the scale of computing power, but did not consider the continuous iteration of computing power. It is equivalent to the current computing power system changing from two-dimensional to three-dimensional. In theory, it is the superposition of countless computing power layers. From the architecture system to the software system, corresponding changes must be made."

"These are your problems," Zhou Zhi said with a wry smile, "I am more concerned about my own problems. We have invested nearly 100 million yuan in Xiaozhi. According to your plan, Xiaomiao, in the next three years, no, in the next year, half a year, or half a year, to what extent do we need to increase the computing power?"

"I think that in order to perfectly complete the system development, function development, and experimental tasks..." Mai Xiaomiao stretched out a palm, with five delicate fingers fully opened: "Five times!"

"No!" Zhou Zhi immediately denied: "Moore's Law is too scary. The computing power of information equipment is now increasing by an order of magnitude every two years. This is especially scary for a system like ours that is small but capable of supporting a large system. It's scary to eliminate 100 million in two years. Now we have to waste 500 million every two years. Who can bear it?"

Gu Congjun, Yao Shijie, and Zhang Jie kept their mouths tightly shut, not daring to say a word. Such devotion was truly terrifying.

"But our costs have come down. Haven't Academician Li and Mr. Masuoka already developed the computing power card? Hasn't Grandpa Hu already produced computing power chips for us?"

"The investment over there is also a gold-devouring beast." Zhou Zhi held his forehead with his hand and still felt his head was heavy. Although Clover was already profitable enough, if the upgrade and iteration capabilities of computing chips and computing cards could not keep up with Moore's Law, then it would be ruthlessly eliminated by the world.

Currently, Clover has teamed up with several large electronics companies in Shanghai to design computing chips and computing cards. These two important electronic devices are designed based on the Clover Group's operating system with independent intellectual property rights. Although the operating interface is very unfriendly and is still command-controlled, its computing power and networking capabilities exceed all current designs in the world.

In fact, no one is arguing about this. This is the world's first neural network computing system. With the insistence of Zhou Zhi and Mai Xiaomiao, it is required to have multiple advanced capabilities such as flexible expansion, free integration, dynamic allocation, network composition and device virtualization. Such powerful functions conceal its many disadvantages, such as its very unfriendly interface, too high requirements for operators, high management requirements, and high requirements for operation and maintenance monitoring.

From the project design to the production of the first computing power card, Academician Li Guangnan's team overcame more than 86 technical difficulties and obtained more than product patents. The first generation of computing power cards used X series chips, which were not as smooth as Clover's own system in terms of computing adaptability. This also contributed to Hu Changfeng's determination to develop computing power chips.

In Lao Hu's words, if we don't do it now, we may not even have the courage to do it in the future.

However, this road is destined to be difficult. The domestic Silicon Valley base in Shanghai undoubtedly does not have the capabilities to challenge the world's giant companies. Fortunately, Zhou Zhi's computing power solution provided Hu Changfeng with a very good opportunity.

As an industrial computer platform, the computing power card developed based on MINICOS has excellent compatibility and expansion performance, and can be customized, which means it is very convenient to expand and upgrade computing power. This will undoubtedly attract many rigid application scenarios in China.

For example, in the formation and management of digital assets in major government agencies and enterprises, the word "scalable" directly hits the hearts of managers.

The main cost of the systems of large units now lies in the hardware, and the mainframes, mid-range computers and minicomputers purchased from overseas have fixed storage space and do not provide customers with the function of arbitrary expansion.

The solution for major companies is generally to estimate their own space needs based on their annual data volume, and then multiply it by two, or at most three, to allocate the required immediate storage space to ensure their own production activities.

But the problem is that if the data increases dramatically after two or three years, or even if the business explosive growth is underestimated, the three-year storage space provisioning often cannot even last for one year. The general solution is to add tapes, copy historical data to tapes for storage, and then clear the data in the hard disk library to free up system space.

However, the annoying thing about tapes is that they are too slow to read, and multiple reads may even damage them, resulting in data loss. The general way to read tape data is to first determine which tape the historical data is on, then take the tape out, restore it to the hard disk system, and then read it from the hard disk system.

The timeliness of this is very limited.

Apart from this, the only remaining method is to change the equipment.

Now Zhou Zhi and Hu Changfeng have brought another solution to everyone. We have a system called computing power cabinet. Each cabinet is equipped with a communication switch. The switch can control the data interaction between computing power cards and storage cards, which is equivalent to a powerful server.

Compared with other servers, the special capability of this computing power cabinet is that it can flexibly configure data communication equipment, computing power equipment, and storage equipment, and can adapt to the company's own "computing power elasticity" and "storage elasticity" to the greatest extent.

In the application scenario where it is used as a "data warehouse", the system can be expanded. For systems that cannot be expanded, it can also be used as a data warehouse to independently store historical data for the production system to read at high speed through an interface. The larger the scale, the better the performance.

This actually solves the pain points in data management for countless industries that have special needs for the preservation of historical data, such as finance, insurance, and postal services, as well as industrial and commercial, tax, and traffic management units in government agencies.

Of course, this is just one application scenario that can be realized by the computing power cabinet, but even this one scenario already has a broad application scenario and meets the rigid needs of many corporate units. To put it simply, there is a market.

However, the one who made money was the Shanghai Silicon Valley led by Hu Changfeng, while Clover was in a miserable situation. The money was invested directly into iterative research and development and Mai Xiaomiao's laboratory. Not only was it not enough, but it even lost money.

But these already fall under the scope of business management, and I can’t explain them in such detail to Mai Xiaomiao.

"Let's build a temple inside the snail shell." Zhou Zhi thought about it and said helplessly: "We will add some computing power, and then Xiao Miao, you can divide one computing power layer into two layers, so there will be four layers, right? It can almost meet your requirements, right?"

Mai Xiaomiao frowned and typed on the computer: "If we stop the Go training, this will be enough."

"That's it! He can't finish his homework and still wants to play? Stop his Go extracurricular class first!" Zhou Zhi acted like a parent making decisions for his own child: "Wait half a year until the family is well-off, then make up for it!" (End of this chapter)