HPE NVIDIA H200 SXM5 141GB GPU for HPE Cray EX

The HPE NVIDIA H200 SXM5 141GB GPU for HPE Cray EX (P65894-B21) is built on NVIDIA's Hopper architecture and integrates 141GB of HBM3e memory, delivering exceptional memory bandwidth and capacity for the most demanding AI and high-performance computing workloads. Designed natively for the HPE Cray EX supercomputing platform, the GPU leverages the SXM5 form factor to enable high-speed NVLink interconnects between GPUs within a node, enabling tightly coupled multi-GPU parallelism that is critical for large-scale model training and scientific simulation.

With HBM3e memory providing substantially higher bandwidth compared to prior generations, the H200 SXM5 accelerates memory-bound workloads including large language model inference, generative AI, molecular dynamics, and climate modeling. The Hopper Transformer Engine with FP8 precision support dramatically increases throughput for transformer-based neural networks, while Confidential Computing capabilities at the GPU level address enterprise security and data governance requirements in shared or multi-tenant infrastructure environments.

As a factory-integrated HPE Cray EX component, P65894-B21 is qualified for HPE's liquid-cooled blade ecosystem, supporting the high power and thermal envelopes required by H200-class silicon. This tight hardware and firmware co-validation with HPE Cray EX infrastructure simplifies procurement, integration, and lifecycle management for enterprise datacenters and national laboratory environments targeting exascale-class AI and HPC convergence.

Ideal for

• Large language model (LLM) pre-training and fine-tuning at scale using distributed multi-GPU parallelism across HPE Cray EX nodes
• Generative AI inference serving for enterprise-grade foundation models requiring high memory capacity and sustained throughput
• Computational fluid dynamics, molecular dynamics, and climate simulation workloads in national lab and research datacenter environments
• High-performance data analytics and digital twin simulations that demand large in-GPU memory footprints and fast memory bandwidth
• Confidential AI and secure enclave workloads leveraging Hopper architecture hardware-level Confidential Computing for regulated industries
• Convergent HPC and AI pipelines within HPE Cray EX supercomputing fabric, enabling simultaneous simulation and machine learning workflows

Technical specifications

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