Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

Nvidia, beware! IBM has a new analog AI chip that could give the H100 a run for its money

Aug 27, 2023 - techradar.com
IBM has unveiled a prototype for an analog AI chip that is up to 14 times more energy efficient than current leading components. The chip, revealed in Nature, is designed to reduce the energy consumption of generative AI platforms, which are known for being power-hungry. The analog chip differs from digital chips as it can manipulate analog signals and understand gradations between 0 and 1, unlike digital chips that only work with binary signals. IBM's chip can model up to 17 million parameters and mimics the way a human brain operates, performing computations directly within memory.

The prototype chip could potentially displace Nvidia, which currently powers many of today's generative AI platforms, if IBM successfully iterates on the prototype for mass market. The IBM prototype was shown to be roughly 14 times more efficient per watt in experiments, with simulations suggesting such hardware could be between 40 and 140 times as energy-efficient as leading GPUs.

Key takeaways:

  • IBM has unveiled a prototype for an analog AI chip that is up to 14 times more energy efficient than current industry-leading components.
  • The chip aims to reduce the energy consumption of generative AI platforms, which are currently very power-hungry.
  • IBM's chip can encode 35 million phase-change memory devices per component and can model up to 17 million parameters.
  • The prototype was shown to be roughly 14 times more efficient per watt, and simulations suggest such hardware could be between 40 and 140 times as energy-efficient as today’s leading GPUs.
View Full Article

Comments (0)

Be the first to comment!