The neural network solution designed by the three-member team of Sinton required huge computing resources. If CPU is used for computing, it may take several months.

Alex and Ilya took a unique approach to select GTX580 graphics card specially designed by NVIDIA for computing, and the calculation was completed in less than a week. The Hinton and the other two rely on this neural network.

ALEXNET won the IMAGENET Challenge, which led to the follow-up story of Qiandu Xingrui Google and Microsoft's home purchase.

Huang Renxun had heard of the story of the Hinton team from Li Rui before, but at this time he realized that Nvidia and this story had such a deep connection. He couldn't help but smile and said, "It can do such a thing for the development of artificial intelligence."

It is my honor to contribute!"

Hinton also said: "When we chose to use GPU for computing, we actually didn't understand whether it could work, but as the experiment progressed, we were surprised to find that the GPU architecture is very suitable for training nerves.

The network has a very fast training speed and can also do training beyond the previous scale, which greatly improves our research efficiency."

Ilya added: "The reason why ALEXNET is successful is that we have trained data that is beyond imagination. If we use CPU to operate, it is simply an impossible task, but if we use GPU to operate,

, It has become so simple and wonderful. I would like to know that Nvidia has not yet continued its plans for new GPU development, and perhaps your new products can give birth to the next stage of artificial intelligence!”

Huang Renxun smiled and said, "Has Li Rui not told you that Xingrui and Nvidia have reached a cooperation intention and are preparing to develop a new GPU specifically for artificial intelligence computing and neural network training, and provide corresponding services on our CUDA software platform.

development function.”

"This is really good news!" Irya was extremely excited.

Not only is it good news, but it is also a secret weapon.

When Huang Renxun appeared at DEEPMIND, he was warmly welcomed.

"Jason, I'm glad to see you. Nvidia's GPU has been a huge help in our research." Hassabis's opening remark was almost the same as Ilja. For those researchers, who can give them

Whoever is the great hero is provided by research.

Huang Renxun smiled bitterly and said, "To be honest, I feel sorry for your gratitude. When developing GPUs, I just thought that they would be used by players in the game to get a better experience, but I didn't expect that they would be used by artificial intelligence.

How helpful is the research? But now I realize that artificial intelligence will become the future trend, and machine learning will be an emerging computing model to solve a large number of responsible and important problems for humans, so Nvidia will definitely move towards this in the future

Go and work hard.”

"That's great. We are looking forward to Nvidia's latest products, which will definitely help our research more." Legge also said excitedly.

With Huang Renxun's joining, the direction of this negotiation changed slightly. When Huang Renxun read the research results of DEEPMIND, he was also very surprised.

"Artificial intelligence can already surpass humans in games, which I didn't expect before." Huang Renxun said with emotion, "Especially when I think that these are all trained through GPUs, I am honored."

Hassabis said: “If you can give me more GPUs, I think your sense of honor will be greatly enhanced!”

This harmless little joke caused everyone to laugh, but the "joke" was quickly moved into negotiations.

"If DEEPMIND accepts the acquisition of R incubator, we are willing to build a super data center shared with REDO. This data center will install more than 20,000 GPU chips developed by Nvidia specifically for machine learning and neural network research."

Ladao.

DEEPMIND Everyone was stunned.

A new model of GPU specially created?

Twenty thousand yuan!

It was obvious that Hasabis and Leg's faces changed and their eyes became dull.

The main researchers such as Graves and Kaguku Oglu were so excited that they looked like they wanted to pounce on him and force Li Rui to write a written promise immediately.

This plan was not whimsical, but was formulated by Li Rui after talking with Professor Hinton.

On the one hand, huge orders can urge Nvidia to carry out the design, development and production of new GPUs as soon as possible.

On the other hand, this super data center is of great help to the research and development of REDO, and is also a huge secret weapon to attract DEEPMIND.

Graves swallowed and said, "Dalula, can you explain in detail the performance of these GPUs?"

Dalula looked at Huang Renxun.

Huang Renxun knew that the time for his performance was coming, and he smiled and said, "NVIDIA's GPU was originally developed for games and can be used in machine learning. To describe it in a sentence of the ancient Chinese saying, it is just a careless planting of willow trees. Unexpectedly,

One day they will form a forest and bring shade to people. Since we know that artificial intelligence is the direction of future human technology development, Nvidia will specially design a GPU born for artificial intelligence, which has lower power, stronger computing power, and performance.

It's more concentrated. I can't disclose the specific parameters for the time being. I can only say that the computing power will be more than ten times higher than the current GTX580!"

Grayfos' eyes were straightened and he murmured: "Ten times...ten times...ten times..."

No one knows more about what this means than DEEPMIND, who is on the front line of artificial intelligence research.

If it weren't for the formal scene of the negotiation, overreaction would have led to a significant drop in negotiation chips. Li Rui estimated that they would immediately open a few cans of beer and cheer while drinking.

Even though it has tried its best to suppress it, DEEPMIND's subsequent negotiation attitude has also changed dramatically, and he no longer insists on the 800 million magnesium quotation.

DEEPMIND is priced at 800 million magnesium, with three main purposes.

First, it will gain enough exit benefits for venture capitalists including Thiel and Elon, which will cost about 30 million magnesium.

The second is to improve the salary level of researchers. Their salary is too low, not even one-third of the same type of company. Since they don’t know how long the company will make a profit, they must prepare for at least five years of salary.

reserve.

Compared to the third point, the first two purposes seem to be small.

In order to continue to develop more advanced artificial intelligence algorithms, DEEPMIND is preparing to build its own data center, with an estimated investment of 200 million magnesium, which is the big part of the expenditure.
Chapter completed!