Researchers from China recently developed the AI chip “Taichi”, which clearly overshadows previous systems. Not only does it consume less energy, but it is also said to be up to 100 times more efficient than previous chips.
The performance of common computer chips has increased significantly in recent years. While Intel, Qualcomm and AMD were once the largest manufacturers, Google and Apple now also build their own components. A Chinese research team recently went one step further.
Researchers at Tsinghua University developed an AI chip that is based on photonics instead of traditional electronic transistors. The development could become the basis for photonic computing operations and efficient AI applications in the real world.
In this type of computing operation, a system processes information not through electrons, as usual, but through light particles. This allows even faster and more energy-efficient computing systems. The researchers named their chip “Taichi”. This could be a significant step in the evolution of artificial intelligence (AI).
AI chip: “Taichi” solves intellectual tasks at a high level
Optical technology was already showing the potential for superior processing speeds and high energy efficiency. This makes them a promising candidate for the next generation of computing. However, previous optical computing approaches were limited to simple AI tasks, such as digit classification or small pattern recognition.
The new “Taichi” system, on the other hand, handles advanced AI tasks with high computing power and energy efficiency. Unlike traditional electronic data processing that goes deep, the Taichi architecture expands its throughput and scale capacity broadly. This enables significantly more extensive computing operations that a system can carry out in parallel.
Using the AI chipset, the network scale could be effectively increased to the billion-neuron level, enabling support for a variety of advanced AGI tasks. AGI is a form of artificial intelligence that is said to be able to handle intellectual tasks at a level comparable to the human mind.
New AI chip significantly reduces energy consumption.
Taichi achieved comparatively high scores in complex classification tasks such as the 100-category ImageNet and the 1,623-category Omniglot database. The system is also capable of performing complex tasks such as composing music and creating stylized paintings.
Furthermore, the energy efficiency is 160 tera operations per second per watt. This is a significant improvement over current photonic integrated circuits and a huge advance over traditional AI chips.
According to Fang Lu, the study's author, Taichi could accelerate the development of powerful optical solutions and serve as critical support for the fundamental model and a new era of AGI. On this basis, artificial intelligence could become even more effective and cost-effective.
Also interesting:
Source: https://www.basicthinking.de/blog/2024/04/21/ki-chip-taichi/
