Nearly thirty years old, established in science and technology

Chapter 53: The process of use is the process of training. Isn’t it natural that the more you use it, the stronger you will get?

The two people moved their positions and sat in a small pavilion by the lake, but there was Zhang Jian opposite.

"You don't have anything to worry about, do you?" Ji Hong was a little puzzled. The Intelligent Computing Center was officially launched and the project has been settled. As soon as this model is completed, the subsequent development will be a matter of course.

At this time, what is Zhang Jian doing with this sad look on his face?
"Take a look." Zhang Jian forwarded a message to Ji Hong.

"Is Huang Renxun coming to China again?" What Zhang Jian sent was a photo. In the photo, Huang Renxun seemed to be the protagonist of a Northeast duo, wearing colorful clothes, which made people feel very down-to-earth at first sight.

"Not only did people come, but they also brought brand new products." Zhang Jian said:
“The hardware scale of the computing power card A900 is exactly the same as that of the A100/A800, and the high-speed interconnection bus bandwidth is only slightly lower than that of the A100.

"And it's different from the A800's lock. According to rumors, this bandwidth can be cracked privately. Once cracked, it will be exactly the same as the A100. If this is true, it has actually bypassed part of the ban."

Ji Hong understood Huang Renxun's operation. He is of Chinese descent. When he came to China and appeared in public in this outfit, he was expressing the feeling that we are compatriots.

Not only that, but we also bring products with great sincerity, which also expresses that we also want to sell high-end computing cards. The ban on sales is not Nvidia’s original intention, and we are also doing everything we can to solve the problem.

"Playing the emotional card?" Ji Hong shook his head: "It's useless. Don't say he is just a Chinese. Even if he is a real Chinese, Nvidia cannot become a Chinese company.

"Besides, even if he wants to, he can't make the decision on many things. He can't even unblock the game card, let alone other things.

"By circumventing the 'ban', this sword is always there, always hanging over the heads of those companies that use NVIDIA solutions.

"A sword that is always hanging over your head and will be cut off at any time. If it were you, would you be able to sleep? Would you be indifferent?

"The ancients all knew that 'before it rains in the sky, the mulberry soil will penetrate the earth, so you can prepare for your home.' No matter how short-sighted you are, the dark clouds are so dark that you can't see the sky. At this time, you should know to run home."

"That's what I say, but the computing power of our computing power cards is really far behind." Zhang Jian lamented:

“Even with the blessing of CTCA, the S4000 still can’t beat NVIDIA’s gaming graphics card RTX4090, let alone their AI card.

"We have evaluated that even the next-generation products currently being designed are almost half the level of A900/A100."

"You are just a fan of the authorities, Mr. Zhang." Ji Hong said with a smile:
"If it is really like what you said, your old leader Mr. Huang will not come to the country again in a hurry. Didn't he just come here in February or March? It has only been a few days and he has come again. It seems that he has been here again. I haven’t changed my clothes, it’s still the same outfit.

"And he also hinted that he could bypass the ban. Is it true? What's the A900? Are you sure it's not the A800 with another name?" Ji Hong's suspicion is also reasonable. Nvidia has a record of doing this.

"Huh?" Zhang Jian was also a little confused. Normally, if he could do these things, he should do them secretly. Otherwise, wouldn't he be afraid of leaking the information and being banned from selling?

Now this is not known to all potential customers, it is almost semi-public, which is indeed unreasonable.

"Not to mention what Nvidia is like, just think of it as real. Let me give you some reassurance. Let's put it this way, you underestimate CTCA. In other words, you should look at the thinking-like architecture from a development perspective."

Ji Hong said with a smile: "Today's performance level and data of S4000 are indeed not comparable to their strongest game card RTX4090, but wait, wait and see again after a round of training. It won't take long, only about a week." . If you are not busy, just stay in Hezhou for a week, and you will know by then."

"Will it grow?" Zhang Jian's mouth opened in shock. He was a professional. In one sentence, he understood what Ji Hong meant, but this... "What's your expression? Are you surprised? This is a quasi-thinking structure. The more you use it, the stronger it will become!" Ji Hong smiled and shook his head up and down twice:

"This is our biggest advantage. You should have some understanding of the CTCA team. They have taken classes twice in the past, and I have conveyed my ideas. They didn't report it to you?"

"I..." Zhang Jian was speechless. The team leader seemed to report to him, saying that parameters could be optimized and performance improved after going online. At the time, he thought it was a routine optimization. Emotions are such an optimization method—— It is equivalent to feedback reinforcement learning of AI.

"Yes! Feedback reinforcement learning, our architecture itself is a thinking-like architecture, which is based on AI. Long-term use is equivalent to the training process, and it will grow. Isn't it natural that the more you use it, the stronger it becomes?" Ji Hong emphasized:

"And not only the computing architecture, but also the ultra-large-scale inter-chip interconnection system. You also know that the larger the cluster, the greater the loss caused by communication.

"In a thousand-card intelligent computing center like this, the actual performance of a thousand-card cluster cannot be a thousand times that of one card. In fact, it is a big hurdle for a traditional cluster to reach 30% of the total performance.

"And we use this method similar to feedback reinforcement learning in AI training. Not to mention high, if we can optimize to 60%, do you know what it means?"

Zhang Jian twitched unconsciously. He wanted to say something but didn't know what to say - 60%. You are going to double the world's most advanced level. What else can I say?
Ji Hong didn't care what Zhang Jian was thinking, but continued:

“Increasing it to 60% means that in a kilocalorie-scale intelligent computing center, our AI computing power can be equivalent to that of the RTX4090 of the same scale, and we will not lag behind at all!
"If, as you said, your next-generation AI card can be half the size of the A100, with the advantages of CTCA and the support of the inter-chip interconnection system, we can directly surpass them at the same scale!

"Now tell me, what are you afraid of?"

"When you say that, I suddenly feel confident!" Zhang Jian, who was still a little worried, instantly came back to life with full health and said angrily: "It seems we must speed up the development of the next generation computing card!"

Ji Hong nodded. The scale of computing power is hard power, which is still very important.

Both of them know that although whispers and personal assistants are used to endorse it during publicity, the intelligent computing center will ultimately rely on the sum of its computing power to speak for itself.

Although the class thinking model is advanced, it is currently a top-secret technology of Juaner Intelligent Technology and does not meet the conditions for open source.

That means that most companies still need to use existing models to train their own artificial intelligence.

Or it can be jointly developed through cooperation, but the relevant training must be through Juaner intelligent technology, just like the Hezhou University Growth Pilot Model.

However, even so, the theory of quasi-thinking models quickly set off a hurricane around the world - in other words, even if it is not used, its existence itself is a huge deterrent.

Just like this moment, Huang Renxun had just returned to the hotel where he was staying. He didn't even think about the time difference and immediately summoned technical staff for an online meeting.

"What do you think of the publicly disclosed information in the patent application of Juan Er Intelligent Technology, and the principles of the quasi-thinking model mentioned by their founder Ji Hong today?"

As the inventor of the GPU and the founder of Nvidia, Huang Renxun is a technology leader himself. He actually already has the answer to this question, but in his heart he is eagerly hoping that someone can make a voice different from his own.

(End of this chapter)