Intel Gaudi 2 HL-225H Mezzanine GPU Accelerator for HPE ProLiant DL380 Gen11

The Intel Gaudi 2 HL-225H Mezzanine GPU Accelerator for HPE ProLiant DL380 Gen11 (P55522-B21) is a purpose-built AI training and inference accelerator engineered for deep integration into HPE's flagship two-socket rack server platform. Based on Intel's second-generation Gaudi architecture, the HL-225H is designed in a mezzanine form factor that enables direct, high-bandwidth connectivity within the DL380 Gen11 chassis without consuming standard PCIe expansion slots. The Gaudi 2 architecture incorporates a matrix of dedicated Tensor Processing Cores (TPCs) and Matrix Multiplication Engines (MMEs) optimized for the dense mixed-precision compute demands of large-scale deep learning models, supporting BF16, FP32, and INT8 data types natively.

The accelerator features 96 GB of High Bandwidth Memory (HBM2e) delivering substantial on-device memory capacity to accommodate large model parameters during training runs, reducing the need for frequent off-chip memory access and improving throughput for memory-bound workloads. Integrated RDMA-capable 24-port 100 GbE connectivity — built directly into the Gaudi 2 die — enables low-latency, high-bandwidth scale-out communication between multiple accelerator nodes without requiring a separate host fabric adapter, simplifying cluster architecture for distributed AI training environments.

As an HPE-qualified component carrying part number P55522-B21, the HL-225H is validated for deployment in enterprise datacenter environments running the HPE ProLiant DL380 Gen11, ensuring compatibility with HPE iLO management, HPE OneView infrastructure management, and standard enterprise operational workflows. It is well-suited for organizations in the UAE, GCC, EMEA, and APAC regions seeking to deploy scalable on-premises AI infrastructure for generative AI, computer vision, natural language processing, and high-performance data analytics workloads.

Ideal for

• Large-scale deep learning model training including transformer-based generative AI and large language models (LLMs) requiring high on-device HBM capacity
• High-throughput AI inference serving for natural language processing, recommendation systems, and computer vision pipelines in enterprise production environments
• Distributed multi-node AI training clusters leveraging the integrated 100 GbE RDMA fabric for efficient all-reduce communication without external networking hardware
• On-premises AI research and development platforms for data science teams requiring reproducible, low-latency access to dedicated accelerator compute resources
• Healthcare, financial services, and government AI workloads with strict data sovereignty requirements that preclude public cloud deployment of sensitive model training data
• Datacenter AI infrastructure consolidation within HPE ProLiant DL380 Gen11 deployments, enabling unified server and accelerator management through HPE iLO and OneView

Technical specifications

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