Samsung HBM3E 16-Hi 36GB (Shinebolt)

The Samsung HBM3E 16-Hi 36GB (Shinebolt), part number H5CG32AFBDX018, represents the most advanced High Bandwidth Memory available from Samsung, built on the HBM3E (High Bandwidth Memory Express) standard. This 16-layer stacked die configuration achieves a total capacity of 36GB per package, utilizing through-silicon via (TSV) interconnect technology to deliver extraordinary memory bandwidth far exceeding conventional GDDR or DDR memory solutions. The Shinebolt designation marks Samsung's second-generation HBM3E product line, engineered specifically for the most demanding compute-intensive environments where both capacity and bandwidth are critical constraints.

Operating within the HBM3E specification, the Shinebolt stack delivers bandwidth of up to 1.28 TB/s per package, enabled by a 1024-bit wide memory bus interface. The 16-Hi stack architecture pushes per-package capacity to 36GB, a significant leap over earlier 12-Hi HBM3E configurations, addressing the memory capacity requirements of large language model (LLM) inference and training workloads where model weights and activation states must reside in on-package memory. The device interfaces with host processors and accelerators via an interposer-based 2.5D packaging approach, making it the foundational memory component for next-generation AI accelerators and high-performance computing processors.

Designed for integration into enterprise AI accelerators, HPC processors, and datacenter-grade compute platforms, the Samsung HBM3E 16-Hi 36GB is procured as a component by system and platform integrators rather than as a standalone upgradeable module. Its adoption in accelerators deployed across hyperscale datacenters, AI supercomputing clusters, and scientific research facilities reflects its status as the current apex of production HBM technology. Omnixon Global supplies this component to enterprise, datacenter, and government buyers across the UAE, GCC, EMEA, and APAC regions.

Ideal for

• Large language model (LLM) training and inference on AI accelerators requiring maximum on-package memory bandwidth and capacity for billion-parameter model weights
• High-performance computing (HPC) and scientific simulation workloads such as computational fluid dynamics, molecular dynamics, and climate modeling that are bandwidth-bound
• Generative AI and multimodal AI inference deployments in hyperscale and enterprise datacenter GPU clusters where memory capacity directly limits batch size and throughput
• Deep learning recommendation model (DLRM) serving, where large embedding tables benefit from the high bandwidth and 36GB per-package capacity to reduce off-package memory accesses
• AI supercomputing node builds and accelerator card manufacturing requiring the highest-tier HBM3E stack for next-generation compute platforms
• Memory-intensive graph analytics, sparse linear algebra, and data analytics acceleration workloads where irregular memory access patterns demand maximum sustainable bandwidth

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.