NVIDIA NVLink Fusion Expands With NVHBM Custom High-Bandwidth Memory

NVIDIA is expanding its NVLink Fusion technology with a new custom high-bandwidth memory solution called NVHBM. This innovation aims to meet the growing demands of AI infrastructure, which requires optimal performance from compute, memory, storage, networking, and software systems. NVHBM offers higher memory performance and efficiency, delivering up to 30% greater memory bandwidth and 15% lower HBM power consumption compared to traditional HBM architectures.

By integrating NVIDIA’s custom memory controller into the HBM base die, NVHBM frees up valuable silicon area on XPU compute dies, resulting in up to 25% more area available for compute. NVIDIA has established a standard NVHBM implementation, available from multiple memory providers, reducing engineering efforts required for integration and qualification across suppliers. This approach enables faster deployment of custom AI chips for hyperscalers and AI innovators.

Amazon’s Annapurna Labs is the first to work on NVHBM technology as part of its collaboration with NVIDIA on NVLink Fusion. This partnership will enhance performance and efficiency for AI workloads, leveraging AWS’s support for NVLink Fusion. Annapurna Labs will integrate NVHBM into its next-generation Trainium chips starting with Trainium4, allowing Amazon chips and NVIDIA GPUs to work together seamlessly on common rack-scale architecture.

NVIDIA's NVLink Fusion platform enables partners to connect custom XPUs and CPUs to NVIDIA’s rack-scale platform, providing access to NVIDIA NVLink chiplets, NVLink-C2C, NVLink Switches, and NVIDIA MGX systems and racks. This collaboration allows hyperscalers and AI-native companies to focus on XPU innovation while utilizing a proven technology stack for scale-up and scale-out networking, rack-scale systems, and software. This creates a faster and lower-risk path to deploying semi-custom AI infrastructure.