NVIDIA GB300 (Lenovo) Rack

NVIDIA GB300, Lenovo AI Infrastructure

Lenovo NVIDIA GB300 solutions deliver enterprise-scale AI computing built around the NVIDIA GB300 Grace Blackwell platform, combining powerful GPU acceleration, high-bandwidth memory and advanced networking to support demanding AI workloads.

Designed for organisations moving beyond AI experimentation into production, Lenovo GB300 systems provide a high-performance foundation for generative AI, agentic AI, large language models, inference, AI training, digital twins, scientific computing and accelerated enterprise applications. PDF Data sheet can be found here.

Lenovo NVIDIA GB300 NVL72 Overview, Key Features

• 72 × NVIDIA Blackwell Ultra GPUs, providing approximately 20 TB of GPU HBM3E memory across the rack.
• 36 × NVIDIA Grace CPUs, providing 2,592 Arm Neoverse V2 CPU cores.
• 130 TB/s of NVLink bandwidth, allowing the 72 GPUs to communicate as a unified compute fabric.
• Up to 576 TB/s GPU memory bandwidth.
• 37 TB of total fast memory, combining GPU HBM3E and CPU memory.
• Integrated NVIDIA ConnectX-8 networking, supporting up to 800 Gb/s connectivity per GPU.
• 18 compute trays, each containing four Blackwell GPUs and two Grace CPUs.
• 9 NVLink switch trays provide the high-speed GPU fabric.
• 6 or 8 power shelves, depending on configuration.

Contact Multimedia Technology AI Group to discuss Lenovo NVIDIA GB300 solutions and identify the right AI infrastructure for your workload and deployment requirements.