Newsletter Subscribe
Enter your email address below and subscribe to our newsletter
[forminator_form id="25163"]
All voices matter
Nvidia unveils NVHBM custom memory for AI chips
- Nvidia announced NVHBM, a custom memory design that claims 30% more bandwidth and 15% lower power than standard HBM4E by embedding the memory controller in the HBM base die.
blogs.nvidia+1 - Amazon's Annapurna Labs is the first partner, planning to integrate NVHBM with its next-generation Trainium4 chips via NVLink Fusion.
blogs.nvidia+1 - The technology frees up to 25% more compute die area for custom AI accelerators, though benefits apply to future designs rather than current Rubin systems.
tomshardware
Nvidia on Wednesday announced NVHBM, a custom high-bandwidth memory technology that promises 30% more bandwidth and 15% lower power consumption than standard HBM4E, expanding the company's NVLink Fusion platform for partners building custom AI chips.
A New Approach to Memory Architecture
The technology rethinks how memory controllers interact with AI accelerators. Traditional HBM designs place the memory controller on the main processor die, consuming silicon area that could be used for compute. NVHBM instead integrates Nvidia's custom memory controller directly into the HBM base die, freeing up to 25% more area on the compute chip for additional processing capabilities.blogs.nvidia+1
"By integrating the memory controller into the 3D HBM stack instead of the XPU, NVHBM delivers up to 30% greater memory bandwidth and 15% lower HBM power consumption," Nvidia said in its announcement. The design also reduces PHY and support area by up to 67% compared with the JEDEC HBM4E standard and simplifies interposer routing, providing up to 80% more usable silicon across the layout.wccftech+1
Nvidia said NVHBM is built on the same technology it will use for future GPUs and will be validated and offered by leading memory partners as a standardized implementation.blogs.nvidia
Amazon's Annapurna Labs First to Adopt
Amazon's Amazon.com, Inc. Annapurna Labs will be the first partner to work on NVHBM, building on its existing collaboration with Nvidia around NVLink Fusion. Annapurna's next-generation Trainium chips, starting with Trainium4, will use NVLink Fusion to connect Amazon chips and Nvidia GPUs under a common rack-scale architecture.tomshardware+1
"NVHBM represents a new architectural approach to advancing high-bandwidth memory performance and efficiency," said Nafea Bshara, vice president of Annapurna Labs at Amazon. "We look forward to this technology collaboration to benefit future AWS infrastructure designs."blogs.nvidia+1
Expanding the NVLink Fusion Ecosystem
NVLink Fusion, first announced in May 2025, enables partners to connect custom XPUs and CPUs to Nvidia's rack-scale platform, accessing NVLink chiplets, NVLink switches, and MGX systems. The addition of NVHBM gives partners another building block, though as Tom's Hardware noted, these benefits apply to future custom designs rather than Nvidia's own Rubin rack-scale systems already in production.tomshardware+1
Partners including MediaTek and Marvell have already adopted NVLink Fusion for semi-custom silicon efforts. WCCFTech reported that Nvidia's own Feynman GPUs, expected in 2028, could be the first Nvidia products to adopt NVHBM.blockchain+1
Sources (4)
- 1 NVIDIA NVLink Fusion Expands With NVHBM Custom High-Bandwidth Memory blogs.nvidia.com
- 2 Nvidia custom 'NVHBM' promises 30% higher bandwidth, 15% lower power than commodity HBM4e — custom base die and PHY will be available to NVLink Fusion partners www.tomshardware.com
- 3 NVIDIA Develops Custom 'NVHBM' Memory For AI, Claiming 30% More Bandwidth and 15% Lower Power Than HBM4E wccftech.com
- 4 NVIDIA Expands NVLink Fusion with NVHBM Memory Tech blockchain.news
