Carding 4 Carders
Professional
IBM is on track to build the most powerful AI chip yet.
IBM is developing an innovative computer chip that mimics the brain and promises to be the fastest in performing artificial intelligence tasks. The new chip, called NorthPole, is a digital architecture compatible with the silicon chips used in modern computers.
NorthPole differs from traditional chips in that it combines computing power with memory blocks that store information.
Researchers from IBM have demonstrated how NorthPole is able to efficiently perform AI-based image recognition tasks with high energy efficiency and speed, outperforming even the latest chips from the leading graphics processor manufacturer NVIDIA. In addition, it successfully handles speech recognition and natural language processing tasks. NorthPole also provides better performance per transistor — the tiny electronic switches that underpin computing technology. The chip accommodates 22 billion transistors in a space of just 800 square millimeters.
However, this specialized design of NorthPole limits its capabilities, as it is not able to perform artificial intelligence training and is not suitable for working with large models. Despite this, the Modha team plans to show how multiple NorthPole chips can be used to support large language models.
These developments are important for the future of artificial intelligence and computer technology. On the one hand, they confirm that IBM continues to strive to improve the efficiency of computing, inspired by biological principles. On the other hand, they demonstrate a variety of approaches to implementing the hardware level of artificial intelligence and the lack of a universal solution suitable for all tasks. Perhaps in the future we will see more hybrid systems combining different types of chips to better meet the diverse needs of artificial intelligence.
IBM is developing an innovative computer chip that mimics the brain and promises to be the fastest in performing artificial intelligence tasks. The new chip, called NorthPole, is a digital architecture compatible with the silicon chips used in modern computers.
NorthPole differs from traditional chips in that it combines computing power with memory blocks that store information.
Researchers from IBM have demonstrated how NorthPole is able to efficiently perform AI-based image recognition tasks with high energy efficiency and speed, outperforming even the latest chips from the leading graphics processor manufacturer NVIDIA. In addition, it successfully handles speech recognition and natural language processing tasks. NorthPole also provides better performance per transistor — the tiny electronic switches that underpin computing technology. The chip accommodates 22 billion transistors in a space of just 800 square millimeters.
However, this specialized design of NorthPole limits its capabilities, as it is not able to perform artificial intelligence training and is not suitable for working with large models. Despite this, the Modha team plans to show how multiple NorthPole chips can be used to support large language models.
These developments are important for the future of artificial intelligence and computer technology. On the one hand, they confirm that IBM continues to strive to improve the efficiency of computing, inspired by biological principles. On the other hand, they demonstrate a variety of approaches to implementing the hardware level of artificial intelligence and the lack of a universal solution suitable for all tasks. Perhaps in the future we will see more hybrid systems combining different types of chips to better meet the diverse needs of artificial intelligence.
