Lenovo ThinkSystem SR675 V3 AI Server

High-Density 3U AI Server — Lenovo ThinkSystem SR675 V3

Lenovo ThinkSystem SR675 V3 is a specialized high-density enterprise-class AI server in a 3U form factor, engineered for maximum GPU acceleration. The platform targets large-scale deep learning, large language model (LLM) inference, metaverse construction, 3D rendering (Omniverse), and high-performance computing (HPC).

The series’ key feature is its revolutionary modular bay architecture. The server is split into a base compute subsystem and swappable front modules (components), allowing the system to be flexibly adapted to different GPU types without being tied to hard-soldered PCIe slots. This architecture protects investments: when upgrading to a new accelerator generation, only the front module needs to be replaced — not the entire server.

Use Cases

The Lenovo ThinkSystem SR675 V3 is the optimal solution for organizations requiring:

• Deep neural network and large language model (LLM) training
• Industrial-scale generative AI model inference
• High-performance computing (HPC) and scientific simulation
• 3D rendering, digital twins, and the NVIDIA Omniverse platform
• GPU cluster and AI factory construction with high accelerator density
• Real-time computer vision and video analytics

GPU Architecture and Key Configurations

The server supports enterprise-grade GPUs through specialized front chassis (SR675 V3 and SR675i V3 variants):

• SXM configuration (high performance). Supports integrated NVIDIA HGX H100 / H200 platforms with 4 GPUs and NVLink inter-chip interconnect, complemented by Lenovo Neptune hybrid liquid cooling or air heatsinks. This configuration delivers maximum GPU-to-GPU bandwidth and is optimal for training large models.
• PCIe Gen 5.0 slot configuration. Supports up to 8 double-wide full-height, full-length GPUs (e.g., NVIDIA H100 NVL / L40S) or up to 8 single-wide cards (e.g., NVIDIA L4). The flexibility of the PCIe configuration allows the server to be adapted to a wide range of workloads — from inference to GPU-accelerated VDI.

Compute Platform

Processors. The server supports one or two 4th Gen (9004) or 5th Gen (9005) AMD EPYC processors with TDP up to 400 W per socket. Up to 256 compute cores are available per server. AMD EPYC processors deliver high memory bandwidth and a large number of PCIe Gen 5.0 lanes — critical for feeding GPU accelerators with data.

Memory. The server provides 24 DIMM slots (12 per processor) for DDR5 ECC RDIMM memory. Maximum capacity reaches 3 TB at speeds up to 4800 MT/s (4th Gen processors) or up to 6 TB at 6400 MT/s (with 5th Gen processors and 3DS RDIMM modules). The large memory capacity allows loading large models and datasets into RAM, reducing training time.

Storage Subsystem. Configuration depends on the front module. The server supports up to 8 × 2.5″ hot-swap drives (NVMe/SAS/SATA) or up to 6 ultra-compact E1.S NVMe drives. For OS installation, an internal dual-drive M.2 module with hardware RAID 1 is provided, separating the boot volume from the data array and improving fault tolerance.

Interfaces and Networking. The server offers up to 14 high-bandwidth PCIe slots (including PCIe Gen 5.0 lanes). The design allows high-speed InfiniBand or Ethernet (up to 400 Gbps) network adapters to be routed to either the rear or front panel for convenient cabling in AI clusters. This simplifies cabling in dense GPU racks and reduces cable length between nodes.

Power, Cooling, and Management

• Power. The fault-tolerant power subsystem includes up to 4 redundant power supplies rated at up to 2600 W each with hot-swap support and the highest Titanium (96%) efficiency rating. N+N redundancy ensures continuous server operation if one or more power supplies fail.
• Cooling. A dense array of powerful hot-swappable fans (Hot-Swap N+1) provides front-to-back airflow. When using SXM5 platforms, the combined Lenovo Neptune liquid cooling system is engaged, efficiently dissipating heat from GPUs with TDP of 700 W and above. This is critical for stable operation of GPU clusters under sustained load.
• Remote Management. Monitoring and management are handled by the standard XClarity Controller 2 (XCC2) chip, supporting deployment via Redfish REST API, TPM 2.0-based Root of Trust, and firmware protection. Integration with Lenovo XClarity Administrator and the XClarity One cloud platform simplifies management of GPU server fleets.

Why GHI SERVERS

• Direct supply from China. We work with CNY and RUB. Turnkey logistics of server equipment from China, including customs clearance and delivery to our Moscow warehouse.
• Custom configuration. We help select the front module (SXM or PCIe), GPU type (NVIDIA HGX H100/H200, L40S, L4), processor count, and memory capacity for your task and budget.
• Moscow warehouse. Prompt replacement of defective units and warranty service without long downtime.
• 20+ years of expertise. Authorized Lenovo partner since 2005. Pre-sales engineering, configuration assistance, and post-sales support.

Service and Support

• 24/7 technical support
• Pre-sales engineering and configuration assistance
• OEM warranty and full provenance documentation
• Installation, setup, and deployment of server equipment
• Component replacement and warranty service