NVIDIA Developing NVHBM for Feynman GPUs

NVIDIA partners with Samsung, Micron, and SK hynix to produce custom NVHBM using TSMC's 3nm node for its upcoming Feynman GPU architecture.

is developing custom High Bandwidth Memory, dubbed NVHBM, for its upcoming Feynman GPUs. The company has requested that , , and manufacture the base die for this memory using 's 3nm node to ensure consistent performance.

Samsung has accepted the request to use TSMC's process, departing from its traditional turnkey approach. The NVHBM solution utilizes a 1c sixth-generation DRAM process for the core die, while Samsung evaluates its own foundry as a production backup.

NVIDIA claims NVHBM offers 30 percent more bandwidth and 15 percent lower power draw than HBM4E. This shift comes as Samsung faces foundry losses, estimated at 6.74 trillion won in 2025, though losses are projected to decrease by 41 percent by 2026.


Related news

Samsung Unveils Enhanced 8LPU+ Chip Process

Samsung Develops New AI-Specific NAND Storage

NVIDIA to Adopt Intel Foveros 3D Packaging

Samsung Q3 Profit Surges 782% to Record High

Samsung Completes HBM4E Testing for NVIDIA