Recent.
A number of brokerage institutions issued research reports.
The release of the SORA model
It will further widen the computing power gap
The industry also believes.
SORA will rapidly promote the commercialization of applications.
Boost the scale of the industry.
And then it is transmitted to the upstream computing infrastructure.
The demand for computing power will explode again.
SORA widens the computing power gap
Computing power is one of the three core elements of AI.
The importance is self-evident.
Take Sora, for example.
It continues the previous GPT model.
source data -transformer-diffusion-emerge".Technology Path
It means that its development is mature.
It also requires massive data + large-scale parameters + large computing power as the base.
AndTrainingThe amount of data required.
Much larger than the amount of data for text training.
Simply estimated by CITIC**.
For a 60 frames ** (about 6 to 8 seconds).
Sora to generate at least.
Approximately 1.2 million tokens (the smallest unit of text processing).
That's a considerable amount of computation.
OpenAI's technical documentation was also revealed.
As the training data increases.
Sample qualityNoticeable improvement.
This also means that more computing power is needed to support it.
The shortage of computing power has become a "problem" in the AI era
The shortage of computing power is not recent.
Since the launch of ChatGPT.
AI computing power has always been a lingering shadow.
In April 2023.
ChatGPT's paid program was forced to be suspended.
This situation underscores an indisputable fact.
In the context of current technological developments.
How to efficiently and sustainably provide sufficient computing resources
It has become a common problem faced by global technology companies
Before the release of SORA.
OpenAI CEO Sam Altman said.
Negotiations are underway with investors.
It aims to increase global chip production capacity.
Expand its ability to support AI performance.
The project may need to be raised.
Up to $5 trillion to $7 trillion.
Computing power is scarce and expensive.
It has become a core factor restricting the development of AI.
It's not just that.
It's as simple as making the AI model run slower.
Also directly related.
The commercialization and industrialization of AI technology
There is not enough computing power to support it.
Cool AI apps.
It can only stay in the laboratory.
Unable to walk into ordinary people's homes.
Founder of Fuxi Think Tank
Director of the Center for Internet Governance at Tsinghua University.
Li Xiaodong once said.
Computing power increases and costs are reduced
It can be made of general artificial intelligence.
Come to the side of ordinary people.
How to fill the "gap" in computing power?
Present. The common demand for computing power in all fields of society.
Explosive growth.
Expected in the next five years.
The average annual growth rate of the total scale of global computing power.
will be more than 50%.
During the 2023 World Internet Conference Wuzhen Summit.
Yang Jie, chairman of China Mobile, said.
In the face of this trend.
It is necessary to speed up the industrial practice of computing power networks
Strengthen the overall utilization and large-scale supply of computing power such as general computing, intelligent computing, and supercomputing.
Promote computing power to become like hydropower.
Social-level services that can be accessed and used immediately.
He Zhengyu, chief technology officer of Ant Group, appealed.
The industry has strengthened software and hardware synergy and ecological construction
Together, we can improve our computing infrastructure.
Computing power scale, availability, and tolerance.
February 27th.
World Internet Conference.
It will be held during Mobile World Congress 2024.
"Computing Power Network" Thematic Forum
Invite the world's industry leaders to discuss.
"Computing Power Network: Intelligent Network Empowers a Smart World".
Just"The network helps provide high-quality computing power"
to cope with the rapid development of AI computing needs".
to create low-cost, green, and sustainable computing power".
The three major topics were discussed separately.
Jointly promote the digital and intelligent transformation in the era of change.
World Internet Conference.