Chapter 387: World-leading Supercomputer Performance (4/4)

On August 10th, the press conference of Qinglong Technology Company was held as scheduled, and the venue was located in the auditorium specially used for product release at the headquarters of Phoenix Software Company.

Although Qinglong Technology Company has made a lot of money from communication equipment, it is not in a hurry to build its own headquarters, and Ren Zhengfei has never mentioned this matter.

Generally, such a major event as the construction of the headquarters needs to be reported to Ye Zishu for approval. Since Ren Zhengfei has no such plan, and Phoenix Software still has some space, he did not remind.

The number of people in the headquarters of Qinglong Technology Company is not very large at present, and the total number is less than 5,000. Most of them are scientific research personnel, and a small part are marketing personnel.

However, the number of sales channels established before adds up to a lot. At the same time, in order to further expand the sales market of communication equipment, the number of international sales staff is also rapidly increasing.

Although their product launch conference was held at the Phoenix Software Company, Ye Zishu did not go to watch it. Although the personal computer business was very important to them, it was only a small part of Ye Zishu's business.

In view of the fact that many things will be related to artificial intelligence in the future, the construction of intelligent supercomputers has reached the point of no delay, and Ye Zishu basically stays with the chip project team for the past two days.

At present, the R\u0026D personnel of the chip project team are mainly engaged in the research and development of server-related chips, as well as the research and development of smart chips that Ye Zishu wants.

In addition, there are many other chips that need to be developed, mainly the research and development of automotive chips, but these chips are not in a hurry, they are just behind.

He will still hand over the research and development of automotive chips to Huanyu Group in the future, but currently they have no chip research and development personnel, and the limited talents in China are basically on the side of Qinglong Technology Company.

When the number of talents here increases, some of them will be distributed to Huanyu Group, because the chips used in cars are not small.

Traditional fuel vehicles want to reflect a sense of high technology. The number of chips in a car is as high as 500 to 600, and there are more than 40 types of chips.

As for the number of chips used in electric vehicles, there are even more chips, often more than a thousand, and the demand for chips in cars is also very large.

Just when Qinglong Technology Company held a new product launch conference, the chip project team was doing the final simulation test on the smart chip.

Ye Zishu also stayed here specially, waiting for the chip test results. When the end of get off work was approaching, the simulation test results finally came out, and the program did not report any logic errors.

Although whether the chip is really OK,

It will not be known until after the tape-out, but according to the advanced level of the chip design software they use, it is almost announced that the chip design has been successful.

They have also developed a large number of chips, and up to now, no chip has passed the simulation test, and the tape-out has failed.

After learning that the simulation test passed, Ye Zishu immediately called Wu Chaoqiang and asked them to make arrangements for the production line and give priority to tape-out of smart chips.

If only one smart supercomputer is built, the cost of the chip is a bit high. In order to reduce the cost of smart chips, he plans to build 15 smart supercomputers at a time.

The 11 intelligent supercomputers are placed in the headquarters of Phoenix Technology's 11 subsidiaries. Since these headquarters are located all over the country, the services of artificial intelligence can easily cover the whole country.

The remaining 4 intelligent supercomputers are used as spares and placed in the headquarters of Phoenix Software Company, and they will usually be turned into supercomputers dedicated to Leaf Book.

Of course, in order to further exert the power of artificial intelligence, these 15 intelligent supercomputers will form a cluster. When a supercomputer's computing power is exhausted, it will promptly call a supercomputer with sufficient computing power as a supplement.

In order to avoid malicious resource encroachment, when the overall computing power consumption reaches the 70% warning line, the most resource-consuming calculations will be collectively kicked out.

Generally speaking, it is impossible for calculations to consume such a large amount of calculations, unless it is a super mathematical problem, but it will also control the use of resources and extend the calculation time to solve the problem.

For intelligent supercomputers, Yezishu adopts an open strategy. Any unit in the country buys it, and Qinglong Technology Company will sell it.

Because intelligent supercomputers are not very special, except for intelligent chips and special architecture design, there is nothing special about other things.

If some artificial intelligence programs are run on intelligent supercomputers, the operating efficiency is much faster than that of ordinary supercomputers.

With the business expansion of Phoenix Technology Company, it is likely to deploy intelligent systems to government units in the future, using intelligent supercomputers, and the effect will be better.

For example, if the smart Skynet system under development wants to achieve the effect he mentioned, it is best to deploy the background running program on a smart supercomputer.

It's just that the performance evaluation indicators of intelligent supercomputers are different from ordinary supercomputers. If the same standard is used to evaluate, intelligent supercomputers may not necessarily have an advantage.

Although they also have their own set of evaluation standards, they are only used by themselves and have not been widely recognized by the outside world.

Therefore, in addition to intelligent supercomputers, Qinglong Technology Company will also develop ordinary supercomputers. At that time, it will be left to the customers how to choose.

Anyway, ordinary supercomputers can also process intelligent programs, but the running effect is not as good as that of intelligent supercomputers.

In order to avoid being unable to keep up with demand in the future, it is necessary to build a new supercomputer, so the supercomputer architecture designed by Leaf Book is scalable.

In other words, if the performance cannot meet the requirements in the future, the number of chips and modules can be added to make the speed faster. Of course, the demand in the early stage is not so large, and it can be built smaller.

In this way, when selling to the outside world, according to the actual needs of customers, supercomputer construction can be carried out to meet the current and future needs.

The architecture he designed can stack the number of chips to 1 million. According to the evaluation of ordinary supercomputers, the peak calculation can reach 12 billion billion calculations.

It is more than twice as fast as the Fuyue supercomputer in Japan in 2020. The theoretical floating-point calculation peak of the Fuyue supercomputer is only 5.3 trillion times per second.

However, if you want to pile up to this level, the power consumption will be very high. Although he used a lot of advanced architecture designs when designing the chip, but the manufacturing process is here, even if the energy consumption is reduced, there is a limit.

Therefore, although the theoretical peak value can be so high, it will not be done in practice, unless it is used for special purposes, regardless of the cost.

Anyway, the company under him is a commercial enterprise, and it pays attention to the best cost performance. It is impossible to pile up the number of chips to the level of 1 million, and only 100,000 can basically meet the demand.

Even if the capacity needs to be expanded in the future, it will reach the top when it is piled up to about 500,000 at most, and the cost performance will drop sharply if it is piled up. It is not as cost-effective to use new chips to build a supercomputer.

And the price is also very impressive. To achieve the effect of 12 billion exascale, the price of each ordinary supercomputer exceeds 10 billion yuan, which is absolutely super expensive in this era.

This is not a big deal, and the cost of use is also not affordable for ordinary enterprises. If you want to obtain performance that surpasses the times, this is the price.

The cost of ordinary supercomputers is similar to that of intelligent supercomputers, but in terms of selling price, intelligent supercomputers are more expensive. After all, intellectual property has a price.

Under the same parameters and performance, the price of intelligent supercomputers will be 1.5 times that of ordinary supercomputers. He decided to sell this price because he felt that it was too high to sell.

If it is used to run ordinary programs, it is not cost-effective to buy an intelligent supercomputer, but it is too cost-effective to run artificial intelligence programs.

The artificial intelligence mentioned here refers to the intelligent program built by the front-end artificial intelligence technology provided by him. The back-end artificial intelligence will not be open to the outside world, let alone deployed on other people's supercomputers.

After passing the smart chip test here, he immediately sent the design data to Xuanwu Technology Company in an encrypted manner, and after 20 days, he could get the tape-out results.

During this period, preparations for other parts of the supercomputer also need to start. The production speed of chips is very fast. It only takes a day or two for them to produce a chip for a supercomputer.

Qinglong Technology Company has established a supercomputer department. Ye Zishu checked their work and found that all aspects are progressing smoothly, which made him feel relieved.