← Back to live feed1 story
Cloud service providers and AI laboratories have shifted to 8-high as the minimum stack height for HBM in low-end inference products, prioritizing bandwidth over cost efficiency, with Nvidia modifying Rubin chip specs and AMD's MI450 projected to use 12 HBM stacks.
Sign in to suggest
Key sources
- SOURCE@jukan05“the upcoming MI450, which is expected to increase the number of HBM stacks within the CoWoS package to 12, delivering 288GB of HBM4”x.com
- SUPPORT@vikramskr“I remember some lengthy discussions on why it doesn't make sense to go to higher stacks -- something about how there is not enough bandwidth to extract the full capacity from a high stack or something.”x.com
- SOURCEhuggingnewshuggingnews.com