Lenovo ThinkSystem NVIDIA GB200 NVL72 Rack-Scale GPU

The Lenovo ThinkSystem NVIDIA GB200 NVL72 Rack-Scale GPU (7D9SCTO1WW) represents a landmark in accelerated computing infrastructure, built on NVIDIA's Blackwell architecture. The NVL72 configuration integrates 36 Grace CPU modules and 72 Blackwell B200 GPUs interconnected via fifth-generation NVLink, forming a single unified rack-scale system that operates as one coherent GPU. This architecture delivers up to 1.4 exaflops of AI compute at FP4 precision and 720 petaflops at FP8, purpose-engineered for the most demanding large-scale AI training, inference, and high-performance computing workloads in enterprise and hyperscale datacenters.

The system leverages NVLink Switch technology to provide full GPU-to-GPU bandwidth across all 72 GPUs within the rack, eliminating traditional PCIe bottlenecks and enabling tightly coupled model parallelism at unprecedented scale. Each B200 GPU is paired directly with a Grace CPU via chip-to-chip NVLink-C2C interconnect, yielding a unified memory architecture with high-bandwidth, low-latency access between CPU and GPU memory pools. The rack integrates direct liquid cooling as a foundational design element, supporting the extreme thermal density generated by this level of compute concentration.

Deployed as a Lenovo ThinkSystem platform, the GB200 NVL72 benefits from Lenovo's enterprise systems integration, including XClarity management, rigorous rack-level validation, and compatibility with Lenovo's broader ThinkSystem datacenter portfolio. It is positioned for organizations building sovereign AI infrastructure, frontier model training clusters, and high-throughput inference farms serving real-time enterprise AI services across sectors including financial services, life sciences, energy, and government-grade research institutions.

Ideal for

• Large-scale generative AI model training, including trillion-parameter foundation models requiring dense GPU-to-GPU communication and unified memory at rack scale
• High-throughput enterprise AI inference serving, enabling real-time responses for large language models and multimodal AI applications with reduced latency per query
• Computational drug discovery and genomics workloads requiring sustained FP64 and FP8 mixed-precision throughput across tightly coupled GPU pools
• High-performance computing simulations in energy, climate modeling, and engineering, leveraging the Grace CPU's memory bandwidth alongside Blackwell GPU compute
• Sovereign AI datacenter buildouts for government and regulated industries requiring on-premises, rack-scale AI infrastructure with enterprise lifecycle management
• Advanced recommender systems and ranking pipelines at hyperscale, exploiting the NVL72's unified memory and NVLink fabric to accelerate embedding lookups and sparse model inference

Technical specifications

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