Samsung Showcases HBM4E For NVIDIA Rubin
Samsung announced at NVIDIA GTC 2026 that its sixth-generation HBM4 memory is in mass production and it is showcasing HBM4E with 16Gbps IO delivering up to 4.0TB/s per stack and 48GB per 16‑Hi stack. Samsung also highlighted SOCAMM2 server memory in mass production and PCIe 6.0 PM1763 SSDs for NVIDIA's Vera Rubin platform. Together these components target Rubin Ultra's multi‑chiplet design to increase capacity up to 384GB HBM and dramatically raise bandwidth for AI training and inference.
Key Points
- 1Announces HBM4E achieving 16Gbps IO, 4.0TB/s per stack, and 48GB per 16‑Hi stack
- 2Highlights SOCAMM2 mass production and PCIe 6.0 PM1763 SSDs for scalable NVIDIA AI infrastructure
- 3Enables Rubin Ultra to scale up to 384GB HBM and significantly increase bandwidth for AI workloads
Scoring Rationale
High novelty and industry-wide scope supported by official Samsung/GTC announcements, limited independent benchmarking evidence.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
