Lenovo ThinkSystem SR665 V3 with AMD Instinct MI210 64GB PCIe

The Lenovo ThinkSystem SR665 V3 with AMD Instinct MI210 64GB PCIe is a high-density, dual-socket server platform engineered for the demands of modern enterprise AI inference, HPC simulation, and data-intensive analytics. Built on the AMD EPYC 9004 series processor architecture and paired with the AMD Instinct MI210 accelerator, the system leverages AMD's CDNA2 GPU architecture to deliver high-throughput compute across double-precision, single-precision, and half-precision floating-point workloads. The MI210 accelerator connects via a PCIe 4.0 x16 interface, providing broad compatibility within existing datacenter infrastructure without requiring specialized interconnect fabrics.

The AMD Instinct MI210 accelerator integrated into this configuration offers 64 GB of HBM2e memory with 1.6 TB/s of peak memory bandwidth, enabling the processing of large model parameters and datasets that exceed the capacity of conventional GPU solutions. With support for AMD ROCm open software platform, the system integrates into established machine learning and HPC software ecosystems including PyTorch, TensorFlow, and OpenMP/MPI-based simulation frameworks. The SR665 V3 chassis supports up to two AMD EPYC processors and substantial DDR5 memory capacity, ensuring the host CPU and memory subsystem are not a bottleneck for GPU-accelerated pipelines.

Designed for enterprise datacenter deployment, the ThinkSystem SR665 V3 is a 2U rack server that supports Lenovo's XClarity management ecosystem, enabling administrators to provision, monitor, and maintain systems at scale across UAE, GCC, EMEA, and APAC datacenters. The platform is suited for organizations requiring reliable, high-performance compute infrastructure for AI model training support, scientific simulation, financial risk modeling, and large-scale data transformation workloads.

Ideal for

• AI inference serving for large language models and computer vision pipelines in enterprise datacenter environments
• High-performance computing (HPC) simulations in energy, engineering, and life sciences requiring sustained FP64 throughput
• Financial services risk analysis and Monte Carlo simulations demanding high memory bandwidth and double-precision accuracy
• Large-scale data analytics and ETL acceleration where GPU-parallel processing reduces time-to-insight for enterprise data teams
• Machine learning model training support tasks including feature engineering, preprocessing, and gradient computation offloading
• Government and research institutions running climate modeling, molecular dynamics, or computational fluid dynamics workloads

Technical specifications

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