AI Rack-scale

Rack-scale systems

NVIDIA GB200 NVL72 by HPE is designed for training, fine-tuning, and inferencing very large AI models, including models with more than one trillion parameters. Its 72 NVIDIA Blackwell GPUs and 36 NVIDIA Grace CPUs are connected through NVIDIA NVLink to provide a single, high-bandwidth 72-GPU domain. The system is well suited for AI service providers and ambitious model builders that need extreme performance and HPE expertise for deployment and ongoing operations.

NVIDIA GB300 NVL72 by HPE advances the NVIDIA NVL72 platform with 72 NVIDIA Blackwell Ultra GPUs, 36 NVIDIA Grace CPUs, high-speed NVIDIA NVLink connectivity, and increased GPU memory. It is optimized for massive real-time AI training and inference for models exceeding one trillion parameters. Customers gain a fully integrated, liquid-cooled rack-scale solution supported by HPE deployment, performance engineering, and global services expertise.

NVIDIA GB200 NVL4 by HPE is optimized for converged HPC and AI workloads that require both GPU acceleration and double-precision performance. Its dense, modular design supports scientific simulation and AI model training, with up to 136 NVIDIA Blackwell GPUs per rack. It is well suited to research labs, universities, and sovereign organizations seeking an Arm-based environment, direct liquid cooling, and HPE Performance Cluster Manager for streamlined cluster deployment and lifecycle management.

AMD Helios AI Rack by HPE is a high-density system for trillion-parameter model training and high-volume inference, built in an open-standards architecture. It integrates 72 AMD Instinct™ MI455X GPUs with AMD EPYC™ "Venice" CPUs, AMD Pensando™ networking and AMD ROCm™ software in a unified design optimized for power, cooling, and serviceability, while supporting open, interoperable rack-scale fabrics such as UALink™ over Ethernet (UALoE). HPE adds a purpose-built HPE Juniper Networking scale-up Ethernet switch, scale-out switches and extensive liquid-cooling and deployment expertise, giving cloud service providers and neoclouds a flexible alternative designed to reduce proprietary lock-in.

HPE differentiation extends beyond supplying infrastructure. HPE brings decades of experience designing, integrating, deploying, and supporting the world's largest supercomputing and direct-liquid-cooled environments. Customers can work with one partner for data center design, system integration, installation, validation, performance tuning, resident expertise, global support, and financing. This end-to-end approach helps organizations reduce facility and execution risk while keeping complex AI environments operating predictably throughout their lifecycle.