Inventec Intel Gaudi 3 8-OAM Server P-Series

The Inventec Intel Gaudi 3 8-OAM Server P-Series (P-G3-8OAM) is a purpose-built AI accelerator platform integrating eight Intel Gaudi 3 OAM (Open Accelerator Module) processors within a dense server chassis. Built on Intel's third-generation Gaudi architecture, each Gaudi 3 accelerator delivers 128 MB of on-chip SRAM and 96 GB of HBM2e memory per OAM module, providing the high-bandwidth memory capacity essential for training and inference of large-scale AI models. The platform leverages Intel's integrated 24-port 200 Gb/s Ethernet fabric via RoCEv2 for high-speed all-to-all communication between accelerators, eliminating the need for a separate networking switch fabric within the node.

The Gaudi 3 architecture features 64 Tensor Processor Cores (TPCs) and a Matrix Multiplication Engine (MME) per OAM, delivering substantial throughput for both FP8 and BF16 precision workloads. The P-Series server form factor from Inventec is engineered for hyperscale and enterprise datacenter density, supporting PCIe Gen 5 host connectivity and designed for compatibility with the OCP OAM mechanical specification. The eight-OAM configuration enables fully interconnected all-reduce operations across all accelerators within the node using the native Gaudi 3 HCCL (Habana Collective Communications Library) stack, which is optimized for scale-out distributed training across multi-node clusters.

Targeted at enterprise AI infrastructure, HPC, and large language model (LLM) deployments, the P-G3-8OAM platform is supported by Intel's Gaudi software ecosystem including the Intel Gaudi Software Suite, SynapseAI SDK, and compatibility with PyTorch and TensorFlow frameworks. The system is positioned for organizations in datacenter environments across the UAE, GCC, EMEA, and APAC regions seeking to build or expand GPU-accelerated AI infrastructure with an Intel-native solution.

Ideal for

• Large language model (LLM) pre-training and fine-tuning at scale using distributed multi-node Gaudi 3 clusters
• Enterprise generative AI inference serving for transformer-based models including LLaMA, Mistral, and similar architectures
• High-performance deep learning research requiring high-bandwidth memory and fast all-to-all accelerator interconnect
• Computer vision and multimodal AI model training for autonomous systems, medical imaging, and industrial inspection workloads
• HPC and scientific computing workloads benefiting from FP8 and BF16 mixed-precision matrix operations
• Datacenter AI infrastructure build-out for cloud service providers and enterprise AI platforms in EMEA and APAC regions

Technical specifications

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