Lenovo ThinkSystem SD650-N V3 with NVIDIA H100 SXM5 80GB (Neptune Liquid Cooling)

The Lenovo ThinkSystem SD650-N V3 is a high-density, liquid-cooled tray server engineered specifically for the most demanding AI training, inference, and high-performance computing workloads in enterprise datacenters. Built on the Neptune direct water cooling architecture, it integrates tightly with Lenovo's Neptune liquid cooling infrastructure to deliver exceptional thermal efficiency, enabling sustained full-throttle GPU performance without the thermal throttling constraints typical of air-cooled systems. Each tray houses dual Intel Xeon Scalable (4th Gen, Sapphire Rapids) processors alongside NVIDIA H100 SXM5 80GB Tensor Core GPUs, interconnected via NVIDIA NVLink and NVSwitch fabric for ultra-high GPU-to-GPU bandwidth.

The NVIDIA H100 SXM5 80GB GPU at the heart of this solution leverages the Hopper architecture, delivering up to 3,958 TFLOPS of FP8 sparse tensor performance and 80GB of HBM3 memory with 3.35 TB/s memory bandwidth per GPU. The SXM5 form factor enables fourth-generation NVLink connectivity with 900 GB/s total NVLink bandwidth, making multi-GPU scaling across tightly coupled nodes highly efficient. These capabilities make the SD650-N V3 ideally suited for large language model (LLM) training, generative AI development, molecular dynamics simulation, computational fluid dynamics, and other multi-node parallel workloads that demand both extreme compute density and memory capacity.

Designed for enterprise datacenter deployment, the ThinkSystem SD650-N V3 integrates into Lenovo's DW612S enclosure, supporting up to six SD650-N V3 trays per chassis for exceptional rack density. The Neptune liquid cooling system removes heat directly from CPUs and GPUs via cold plates connected to facility water, dramatically reducing cooling overhead and supporting Power Usage Effectiveness (PUE) targets below 1.1 in optimized datacenter environments. Lenovo XClarity management software enables centralized monitoring, firmware orchestration, and lifecycle management across large-scale deployments, meeting the operational demands of enterprise IT and cloud service organizations across EMEA and APAC regions.

Ideal for

• Large-scale generative AI and large language model (LLM) training across multi-node GPU clusters
• Enterprise AI inference serving for real-time natural language processing, computer vision, and recommendation systems at scale
• High-performance scientific computing including molecular dynamics, climate modeling, and computational fluid dynamics
• Deep learning research and development requiring high HBM3 memory capacity for oversized model parameters
• Financial services quantitative modeling and risk simulation demanding sustained FP64 and mixed-precision throughput
• Government and defense HPC workloads requiring dense GPU compute in space- and power-constrained datacenter environments

Technical specifications

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