Tech Talk: What's the role of storage capacity in AI development?

Editor's note: Generative AI tools like ChatGPT, Midjourney and Sora have been fueling our daily discussions recently for their jaw-dropping capabilities. However, this initial excitement also comes with anxiety: What exactly can we expect from AI? Are fears of human labor being replaced by AI warranted? Can we find ways to coexist? In this special series of Tech Talk, we invite scholars and industry experts to explore these questions and see if we can find any common ground. This episode focuses on the significance of storage capacity in the progression of AI technology.

Computing power and algorithms are familiar concepts in the current frenzy of artificial intelligence (AI), while storage capacity, a relatively less mentioned thing, also plays a vital role in supporting the needs of AI applications and the technological development.

The development of AI requires boosting storage capacity in tandem with computing power, Feng Dan, a deputy to the 14th National People's Congress and dean of the School of Computer Science and Technology, Huazhong University of Science and Technology, told CGTN reporter Zhao Chenchen in a recent interview.

"Data, computing power and algorithm are the three key elements of AI development," said Feng. "First and foremost, data must be stored. Only if it is properly stored can it be effectively used and its worth realized."

Feng mentioned that in their research, they always found emphasis more laid on computing power than storage power when constructing a computing center. However, such a center must have a balance of computing power as well as the storage capacity if it is to support the operation of AI algorithms, said Feng.

Besides, strong storage can accelerate computing power when the latter is insufficient.

"Advanced storage can store high-quality data for AI large models to conduct related training and reasoning, making them more accurate, and facilitate computing or reduce wait time, thus improving the overall performance," said Feng.

Storage efficiency and energy saving

Storage, like computing, also consumes a large quantity of energy, especially when data needs to be quickly accessed.

"Therefore, we need to make management accordingly," said Feng.

Based on its access frequency, data can be classified into hot, warm and cold data.

"Data that requires instant and frequent access is hot data, which may progressively change to warm data over time. The one that is only accessed on occasion, with longer retention time, is cold data," she said.

To save energy, hot data is always placed in online storage and cold data in offline storage, and for warm data, it can be placed in the in-between storage, which may be slower in speed but larger in capacity.

Storage facilitates AI technology, and vice versa. Based on AI prediction on data access frequency, data can be managed in different storage in advance, said Feng.

Two-pronged way to enhance storage capacity

The expert called for the enhancement of storage capacity so as to develop it into "new quality productive forces" – advanced productivity featuring high technology, high efficiency, and high quality that is highlighted in this year's government work report. 

She introduced improvement in two aspects.

One is to use existing storage media and devices to build large-scale storage systems to meet the needs of AI applications, which includes computer architecture, parallel scheduling technology, as well as technologies on hardware and software, controller, data management, and data migration and scheduling.

The other is to strengthen basic research. For example, look for the next generation of faster and more suitable storage media for AI applications, such as the memristor-based ones, which can realize efficient in-memory computing.

Feng also encouraged the cultivation of young talents in developing storage capacity, including those for basic research and for pragmatic application.

"The development of technology cannot be separated from contributions made by young people," said Feng.

Video editor: Chen Chen, Zhao Chenchen

Videographer: Zhang Hengrui

Cover image: Zhu Shangfan