China's Taichi-II Chip: First Optical AI Processor Outperforms Nvidia H100

Beijing researchers have introduced the Taichi-II chip, the world's first fully optical AI processor, which significantly outperforms NVIDIA's H100 in energy efficiency. This innovative chip operates entirely on light, leading to a remarkable six orders of magnitude improvement in energy efficiency, particularly in low-light imaging scenarios. Additionally, Taichi-II demonstrates a 40% increase in accuracy for classification tasks compared to the H100. The chip employs a novel training method called Fully Forward Mode (FFM) learning, which allows for parallel processing directly on the optical chip, enhancing its performance. Developed by Professors Fang Lu and Dai Qionghai from Tsinghua University, the Taichi-II chip represents a substantial advancement from its predecessor, the original Taichi chip, which had already set records for energy efficiency. The introduction of Taichi-II is strategically significant, especially in light of U.S. restrictions on China's access to advanced GPUs for AI training, providing a viable alternative for continued progress in AI technology. The chip's capabilities could play a crucial role in China's technological advancements and defense capabilities, marking a pivotal moment in the evolution of AI and optical computing.

- Taichi-II is the first fully optical AI processor, surpassing NVIDIA's H100 in energy efficiency.

- It achieves a six orders of magnitude improvement in energy efficiency in low-light scenarios.

- The chip features a 40% accuracy boost in classification tasks compared to the H100.

- Fully Forward Mode (FFM) learning enables efficient parallel processing on the optical chip.

- The development of Taichi-II offers a strategic alternative amid U.S. restrictions on advanced AI hardware for China.