Samsung HBM4E Passes Testing for NVIDIA's "Rubin Ultra" GPU

What happened

Samsung's upcoming HBM4E memory has reportedly passed testing for NVIDIA's next-generation "Rubin Ultra" AI GPU accelerator family. This is a reported test-pass result and does not confirm volume production or supply. As background, Samsung's previous HBM3E did not meet NVIDIA's requirements, leading NVIDIA to choose Samsung for GDDR7 only instead of HBM3E. On specifications, Samsung's HBM4E designs are currently running at 14 Gbps per pin and 2,048 pins per stack, achieving a maximum bandwidth of 3.6 TB/s per stack; at 16 Gbps, the maximum memory bandwidth will increase to 4 TB/s for a single stack, with about a dozen memory stacks per AI accelerator or GPU. Samsung currently offers this memory in a 12-layer (12-high) configuration with a memory density of 48 GB per stack, while future customer requirements will also include 8-layer 32 GB and 16-layer 64 GB stacks.

Impact and signal

Affected components and suppliers

Source and time

Source: TechPowerUp · 10/7/2026 · View original source