Intel Gaudi 3 OAM 128GB HBM2E (8-module system card)

The Intel Gaudi 3 OAM 128GB HBM2E (8-module system card), offered by Supermicro under manufacturer part number 1-300-100-006, is a high-density AI accelerator system card integrating eight Intel Gaudi 3 OAM (Open Accelerator Module) processors onto a single board. Built on Intel's third-generation Gaudi architecture, each OAM die features dedicated matrix multiplication engines and tensor cores optimized for large-scale deep learning training and inference, with the eight-module configuration delivering a combined 128GB of HBM2E memory with high aggregate memory bandwidth across the full card. The Gaudi 3 architecture provides significant generational improvements in FP8, BF16, and FP16 compute throughput compared to its predecessors, making it well suited for the demanding matrix operations at the heart of large language model (LLM) training and transformer-based inference workloads.

The OAM form factor adheres to the OAM Universal Baseboard (UBB) open standard, enabling standardized integration into compatible OAM-ready server platforms including Supermicro's purpose-built AI training servers. Each Gaudi 3 OAM module incorporates onboard 24 x 200GbE RoCEv2 high-speed interconnects (across the full eight-module card), enabling direct, low-latency scale-out networking between nodes without requiring a separate, proprietary switching fabric. This design reduces infrastructure complexity for multi-node AI clusters and allows the card to participate in large distributed training runs across hundreds of accelerators using standard Ethernet-based RDMA protocols.

As a Supermicro-integrated system card, the 8-module Gaudi 3 configuration is intended for enterprise datacenter deployments where maximum accelerator density, memory capacity, and scale-out networking are critical. It is supported by Intel's open software stack, including the Habana SynapseAI SDK, which provides compatibility with PyTorch and TensorFlow frameworks, enabling enterprise teams to migrate and optimize existing AI workflows without dependency on proprietary toolchains. This accelerator is available through Omnixon Global for enterprise buyers across the UAE, GCC, EMEA, and APAC regions.

Ideal for

• Large language model (LLM) pre-training and fine-tuning at scale across multi-node Gaudi 3 clusters using RoCEv2 scale-out networking
• High-throughput generative AI inference serving for transformer-based models requiring large HBM memory capacity per node
• Enterprise computer vision and multimodal model training demanding high BF16/FP8 compute density within a constrained datacenter footprint
• Distributed deep learning research requiring an open, standards-based accelerator ecosystem with PyTorch and TensorFlow framework support
• AI-optimized private cloud infrastructure buildouts in regulated industries seeking alternatives to proprietary accelerator and networking stacks
• Sovereign AI datacenter deployments across EMEA and APAC requiring high-density, OAM-standard accelerator hardware with worldwide logistics support

Technical specifications

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