Lenovo ThinkSystem SR675 V3 with NVIDIA L40S 48GB PCIe

The Lenovo ThinkSystem SR675 V3 (7D9QCTO3WW) is a high-density 2U dual-socket AMD EPYC-based rack server engineered to support demanding AI inference, professional visualization, and GPU-accelerated compute workloads at enterprise and datacenter scale. Configured with NVIDIA L40S 48GB PCIe GPUs, the SR675 V3 delivers exceptional parallel processing throughput, combining Ada Lovelace GPU architecture with high-capacity GDDR6 memory to handle large model inference, rendering pipelines, and data-intensive analytics simultaneously.

The NVIDIA L40S 48GB PCIe GPU integrated into this configuration is purpose-built for universal GPU workloads, offering a balance of AI training, inference, and professional graphics in a single-slot PCIe form factor. With 142 TFLOPS of FP32 performance, 362 TOPS of INT8 throughput, and 91.6 TFLOPS of TF32 Tensor Core performance, the L40S accelerates generative AI, large language model inference, 3D rendering, and simulation workloads. The 48GB of GDDR6 memory with ECC support enables the loading of large AI models and complex scene data without memory bottlenecks.

The ThinkSystem SR675 V3 chassis is optimized for GPU density, supporting up to eight PCIe Gen 5 GPUs and providing the thermal and power infrastructure necessary to sustain continuous high-performance GPU operation in production datacenter environments. It is compatible with AMD EPYC 9004 series processors, PCIe Gen 5 connectivity, and Lenovo's XClarity management ecosystem, making it a capable platform for organizations across UAE, GCC, EMEA, and APAC regions deploying AI infrastructure, virtual workstation fleets, or HPC clusters.

Ideal for

• Large language model (LLM) inference serving for enterprise AI applications requiring high throughput and low latency with large model sizes up to and beyond 48GB VRAM per GPU
• GPU-accelerated 3D rendering and visualization for media, engineering, and design workflows leveraging the L40S Ada Lovelace architecture and RT Cores
• AI-assisted medical imaging, drug discovery simulation, and genomics workloads requiring high-precision floating-point compute and large memory capacity
• Virtual workstation delivery at scale using NVIDIA vWS and vPC virtualization, enabling remote professional-grade graphics for distributed enterprise teams
• Computational fluid dynamics (CFD), finite element analysis (FEA), and scientific simulation workloads that benefit from high FP32 throughput and large GPU memory
• Generative AI model fine-tuning and multi-model parallel inference in cloud-adjacent on-premises datacenter deployments requiring dense GPU-per-rack configurations

Technical specifications

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