AI model growth versus hardware capacity
A user argues that leading AI models are outgrowing current infrastructure, pointing to a claimed mismatch in how quickly model parameters and hardware capacities double.
TLDR
A user claims AI model parameters are doubling every four months, while system capacities—network speeds, PCIe and high-bandwidth memory—are doubling every 24 months. Their argument is that enthusiasm for ever-larger models is colliding with hardware limits.
Combined views
—
1 Source, first seen ago
