NVIDIA Expands NVLink Fusion with NVHBM Custom High-Bandwidth Memory

The mainstreaming of AI agents and trillion-parameter workloads is presenting new requirements for infrastructure. NVIDIA has expanded NVLink Fusion with NVHBM (NVIDIA NVHBM) technology to enable hyperscalers and AI innovators to build next-generation semi-custom AI infrastructure.

Existing HBM architectures placed memory controllers on the XPU die, consuming silicon area that could otherwise be used for computation. NVHBM addresses this by integrating NVIDIA's custom memory controllers into the HBM base die. This delivers performance benefits including 30% higher memory bandwidth, 15% lower power consumption, and 25% more XPU compute die area compared to standard HBM4E.

NVIDIA has established a standard NVHBM implementation available from multiple memory suppliers, reducing engineering burdens for customers and accelerating the time-to-market for custom AI chips. Amazon's Annapurna Labs is leading the development of NVHBM technology as part of the NVLink Fusion collaboration, with support planned for the next-generation Trainium4 chip. This enables Amazon chips and NVIDIA GPUs to operate together within a common rack-scale architecture.