NVIDIA has expanded its NVLink Fusion program with NVHBM, a new high-bandwidth memory technology that moves the memory controller off the XPU die and into the HBM base die itself. Compared with standard HBM4E, NVHBM offers up to 30% greater memory bandwidth, 15% lower power consumption, and frees up to 25% more compute die area. Amazon's Annapurna Labs is the first partner to collaborate on NVHBM, building on its existing NVLink Fusion support, with integration planned starting with Trainium4 chips to enable Amazon and NVIDIA hardware to share a common rack-scale architecture.
Table of contents
AWS and NVIDIA Continue NVLink Fusion CollaborationVertically Integrated and Horizontally OpenQuestions this post answers
What is NVIDIA NVHBM and how is it different from standard HBM4E memory?
NVHBM is a new high-bandwidth memory technology from NVIDIA that integrates the memory controller into the HBM base die instead of the XPU die, unlike traditional HBM architectures. This design delivers up to 30% greater memory bandwidth, 15% lower HBM power consumption, and frees up to 25% more area on the XPU compute die compared with standard HBM4E, and is offered through NVLink Fusion to hyperscaler partners. Engineers evaluating custom AI chip memory architectures can follow NVHBM developments on daily.dev.
Which company is the first to collaborate with NVIDIA on NVHBM technology?
Amazon's Annapurna Labs is the first partner to work with NVIDIA on NVHBM as part of its broader NVLink Fusion collaboration. Annapurna Labs will also integrate NVLink scale-up architecture, with support planned starting with its next-generation Trainium4 chips, enabling Amazon chips and NVIDIA GPUs to share a common rack-scale architecture. Teams tracking AWS and NVIDIA infrastructure partnerships can stay current on daily.dev.