HPE NVIDIA B200 SXM6 192GB GPU for HPE Cray XD

The HPE NVIDIA B200 SXM6 192GB GPU for HPE Cray XD (P68120-B21) is built on NVIDIA's Blackwell architecture, representing a significant generational leap in GPU compute density and memory capacity for large-scale AI and HPC infrastructure. The B200 SXM6 form factor integrates directly into HPE Cray XD system blades via the SXM6 high-bandwidth interconnect, enabling tightly coupled multi-GPU configurations with NVLink 5 providing GPU-to-GPU bandwidth that far exceeds PCIe-attached alternatives. With 192GB of HBM3e memory, the accelerator supports exceptionally large model parameters and datasets entirely in GPU memory, reducing data movement overhead in memory-bound workloads.

Designed for hyperscale AI training and inference at enterprise and national-laboratory scale, the B200 delivers substantial improvements in FP8, FP16, BF16, and TF32 tensor compute throughput compared to its predecessors, enabling faster iteration cycles for foundation model development and scientific simulation. The second-generation Transformer Engine and support for FP4 and FP6 precision formats further extend efficiency for inference serving of large language models and multimodal AI systems. HPE's integration of this GPU within the Cray XD portfolio ensures the component is validated and optimized for the system's liquid-cooled chassis, high-density power delivery, and NVLink Switch fabric.

For enterprise IT and datacenter operators deploying AI infrastructure at scale, the HPE NVIDIA B200 SXM6 192GB GPU (P68120-B21) represents a purpose-built,-qualified accelerator for HPE Cray XD platforms. Its combination of Blackwell compute capabilities, high-capacity HBM3e memory, and SXM6 mechanical and electrical integration makes it a foundational component for organizations building next-generation AI supercomputing clusters, scientific research environments, and large-scale generative AI production systems across the UAE, GCC, EMEA, and APAC regions.

Ideal for

• Large language model (LLM) pre-training and fine-tuning requiring high memory capacity and sustained tensor throughput across multi-GPU nodes
• Generative AI inference serving for enterprise-scale deployments of multimodal and transformer-based models with strict latency requirements
• High-performance scientific computing and simulation workloads in climate modeling, molecular dynamics, and computational fluid dynamics
• Deep learning recommendation model (DLRM) training with large embedding tables that benefit from 192GB HBM3e capacity
• Sovereign AI and national research infrastructure buildouts requiring-integrated, datacenter-validated GPU accelerators
• Mixed-precision AI workloads leveraging FP8 and BF16 tensor formats for energy-efficient training and inference in liquid-cooled HPE Cray XD environments

Technical specifications

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