Lenovo ThinkSystem SR675 V3 GPU Server

The Lenovo ThinkSystem SR675 V3 (7D9QA01CNA) is a 4U rack server engineered specifically for GPU-accelerated AI training, inference, and high-performance computing workloads. Built on the AMD EPYC 9004 series (Genoa) processor platform, the SR675 V3 supports dual AMD EPYC processors and accommodates up to eight NVIDIA H100 or H200 SXM5 GPUs, making it one of the highest GPU-density platforms in the ThinkSystem portfolio. The architecture leverages NVLink and NVSwitch interconnects across the GPU complex, enabling high-bandwidth peer-to-peer GPU communication essential for large-scale generative AI model training.

The server is designed to address the demanding thermal and power requirements of next-generation GPU workloads, incorporating a direct liquid cooling (DLC) architecture alongside air cooling options to sustain full GPU TDP operation at scale. PCIe Gen 5 connectivity throughout the platform ensures that CPU-to-GPU and storage-to-GPU data paths are not bottlenecked, while support for high-capacity DDR5 memory provides the host memory bandwidth required to feed multiple concurrent GPU workloads. The system also supports high-speed networking via OCP 3.0 and PCIe expansion slots, enabling 400GbE InfiniBand or Ethernet fabric integration for multi-node AI cluster deployments.

Targeted at enterprise AI infrastructure teams, research institutions, cloud service providers, and national laboratories, the ThinkSystem SR675 V3 is optimized for large language model (LLM) training, diffusion model inference, molecular dynamics simulation, climate modeling, and other compute-intensive HPC applications. The platform integrates with Lenovo XClarity Administrator for unified systems management and supports standard IPMI/Redfish interfaces, enabling seamless incorporation into existing enterprise data center automation and orchestration workflows.

Ideal for

• Large language model (LLM) pre-training and fine-tuning using multi-GPU distributed training frameworks such as Megatron-LM and DeepSpeed
• Generative AI inference serving for enterprise applications requiring high-throughput, low-latency GPU compute at scale
• High-performance computing simulations including computational fluid dynamics, molecular dynamics, and climate/weather modeling
• Computer vision and multimodal AI model development requiring sustained FP8/FP16/BF16 tensor compute across multiple GPUs
• AI research cluster node deployment where dense GPU packing per rack unit reduces data center footprint and fabric hop count
• Enterprise MLOps pipelines requiring a standardized, manageable GPU server platform integrated with data center orchestration tools

Technical specifications

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