Here it comes! China has directly launched a new chip, with computing power exceeding top GPUs by a thousand times

On October 15, it was reported that the research team led by Professor Sun Zhong from the Institute of Artificial Intelligence at Peking University, in collaboration with the research team from the School of Integrated Circuits, successfully developed a high-precision, scalable analog matrix computing chip based on resistive memory.

This chip first achieved precision comparable to digital computing. When solving key tasks such as large-scale MIMO signal detection, its computing throughput and energy efficiency are hundreds to thousands of times better than existing top GPUs, representing a historic breakthrough in China's computing architecture.

This is not a single technological achievement, but a strategic counterattack by China in the chip sector, where it had been constrained. It shows the world for the first time that China's chips are not just catching up, but can also change the track and lead.

This breakthrough breaks the fundamental logic of Western technological monopoly.

Over the past few decades, global computing power has been tightly controlled by a few American companies. GPUs, CPUs, and AI accelerators have almost all been in their hands.

Even if China made efforts in packaging and process technology, it was always limited to the framework of manufacturing.

But now things are different. The full analog matrix computing chip developed by the Peking University team does not rely on Western processes or imported core technologies. Instead, it uses self-developed devices, self-designed architectures, and self-created algorithms to redefine computing methods from the physical level.

This means that China has finally touched the bottom switch of the computing system - no longer locked by GPUs, but able to build new super brains on its own, gaining independent voice in the future fields of AI and high-efficiency computing.

More importantly, this is not just a toy in the laboratory, but an industrialized scalable path.

The team has already achieved 24-bit fixed-point inversion in a 16×16 matrix, with a relative error as low as 10⁻⁷; in a 128×128 scale calculation, the computing power directly crushes the single-core performance of high-end GPUs.

This chip can complete the work that traditional GPUs do in a day within one minute, and its energy consumption is only one percent of the latter's.

This breakthrough in energy efficiency makes green computing a reality, providing key support for China to build low-energy supercomputing centers and AI large model infrastructure.

It can be anticipated that behind this small chip lies a real structural rise in China's science and technology system. It demonstrates not only scientific and technological strength, but also the courage of route autonomy - daring to open new paths in the most closed fields, and daring to rewrite the rules in an unfair world.

Original: www.toutiao.com/article/1846030793006156/

Statement: This article represents the views of the author.