Lenovo ThinkSystem SR670 V3 with NVIDIA H100 PCIe 80GB GPU

The Lenovo ThinkSystem SR670 V3 (7D9DA00XNA) is a 2U dual-socket rack server purpose-engineered for high-density AI inference, deep learning training, and GPU-accelerated enterprise workloads. The system is designed to house up to eight double-wide GPUs or ten single-wide GPUs, and in this configuration is equipped with the NVIDIA H100 PCIe 80GB GPU, which is built on the NVIDIA Hopper architecture. The H100 PCIe GPU delivers up to 51 TFLOPS of FP64, 51 TFLOPS of FP64 Tensor Core, 378 TFLOPS of FP16 Tensor Core, and 3,958 TOPS of INT8 Tensor Core performance, making it one of the most capable commercially available accelerators for scientific computing and large-scale AI model development.

The SR670 V3 platform supports 4th Generation Intel Xeon Scalable processors (Sapphire Rapids) in a dual-socket configuration, paired with up to 8TB of DDR5 system memory across 32 DIMM slots, providing the memory bandwidth and CPU compute headroom required to feed multiple high-performance GPUs without bottlenecks. PCIe Gen 5 connectivity across the platform ensures maximum throughput between the host processors and the installed accelerators. The server also supports NVMe storage directly in the system, reducing I/O latency for data-intensive AI pipelines.

Designed for enterprise datacenters operating in demanding environments including UAE, GCC, EMEA, and APAC regions, the ThinkSystem SR670 V3 integrates Lenovo XClarity management for streamlined lifecycle operations, monitoring, and remote administration at scale. The system supports ASHRAE A2 and A3 thermal specifications and front-to-rear airflow, making it compatible with a broad range of datacenter cooling configurations. Its combination of Hopper-generation GPU acceleration, DDR5 memory, and PCIe Gen 5 interconnects positions the SR670 V3 as a premier platform for organizations building AI infrastructure, HPC clusters, and GPU-accelerated cloud services.

Ideal for

• Large-scale deep learning model training for natural language processing, computer vision, and generative AI using frameworks such as PyTorch and TensorFlow
• High-throughput AI inference serving for real-time enterprise applications including recommendation engines, fraud detection, and autonomous systems
• Scientific and technical high-performance computing (HPC) simulations in sectors such as energy, life sciences, and engineering
• GPU-accelerated data analytics and preprocessing pipelines requiring high memory bandwidth and low-latency storage access
• Multi-tenant GPU cloud infrastructure for enterprises deploying internal AI-as-a-service platforms across business units
• Rendering, simulation, and digital twin workloads in manufacturing, architecture, and media production environments

Technical specifications

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