AI Silicon

AI acceleratorsHBM4

Nvidia trims Rubin memory bandwidth without an announcement

The published specs for Nvidia's Vera Rubin fell by roughly 13 percent on memory bandwidth with no press release attached.

A Seoul inference chipmaker quadruples its worth in a single year

A Chinese driving-chip designer bets the benchmark was the wrong test

Defense hardwareFoundry Robotics

A California startup sells the assembly line instead of the robot

AI acceleratorsBenchmarks

A Shanghai designer takes aim at an Nvidia part it cannot buy

China semiconductorsCXMT

A Chinese DRAM maker starts HBM3E output with most wafers scrapped

InfineonMicrosoft

Three chipmakers put money behind Europe’s robot startups

Show More