News

Technology

Chinese Researchers Achieve 100x AI Inference Speed Boost with All-Optical Chip Interconnect Using One-Ninth the Computing Power

A Peking University research team has developed an all-optical interconnect system that accelerates distributed AI inference by more than 100 times while consuming only about one-ninth the computing resources of conventional GPU systems. The prototype features a 400 Gbps silicon photonic transceiver and a 16×16 optical switching chip, delivering a total switching bandwidth of 6.4 Tbps. The findings have been published in National Science Review.

AI自動駕駛技術 / 硬體

23 days ago

Source: Interesting Engineering

All Articles

DJI Mic 3 Two-Person Wireless Microphone Kit Drops to $219 in Its First-Ever Sale

Uber CPO Sachin Kansal on Hotels, Robotaxis, and AI Strategy: 'We're Not Trying to Do Everything'

Scientists Discover the Brain Makes Decisions Far Earlier Than Previously Thought

NVIDIA Ising Decoding Method Cuts Color Code Logical Error Rate by Over 300x

World's First Bird-Sized Aquatic-Aerial Robot Swims, Dives, and Autonomously Returns to Flight

U.S. Military's First Combat Deployment of Saronic Corsair USVs Strikes Iranian Submarine at Bandar Abbas Naval Base

Inspired by Tinker Bell and Mew, a Floating Robot Redefines Home Companionship

What Anthropic's Latest AI Research Found — And What It Can't Tell Us

US Navy Expands Flight III Destroyer Fleet with SPY-6 Radar, Enhanced Missile Defense

Airborne LiDAR Uncovers Over 230,000 Hidden Gold Rush Sites and Hazardous Terrain Across Victoria, Australia

Chinese Researchers Achieve 100x AI Inference Speed Boost with All-Optical Chip Interconnect Using One-Ninth the Computing Power

Single-Crystal Turbine Blade Technology Detected in Russian Kh-101 Missile Wreckage, Suggesting GE-Class Engine Capability