Lenovo ThinkSystem SR780a V3 8x Gaudi 3 OAM Server

The Lenovo ThinkSystem SR780a V3 is a purpose-built AI training and inference server designed to house eight Intel Gaudi 3 OAM (Open Accelerator Module) accelerators in a dense 8U rack form factor. Built around Intel's Gaudi 3 architecture, each accelerator delivers high-bandwidth memory via HBM2e and features 24 100GbE RoCE network ports per accelerator for low-latency, high-throughput scale-out networking without reliance on proprietary interconnects. The server is engineered to meet the rigorous demands of large-scale generative AI model training, LLM fine-tuning, and deep learning inference at enterprise and hyperscale datacenter deployments.

The SR780a V3 platform supports dual 4th or 5th Generation Intel Xeon Scalable processors, providing substantial CPU resources to manage data pipelines, preprocessing, and orchestration workloads alongside the Gaudi 3 accelerators. The system supports high-capacity DDR5 system memory and multiple PCIe Gen 5 expansion slots, enabling flexible storage and networking configurations. Lenovo's ThinkSystem engineering delivers enterprise-grade thermal management, hot-swap component support, and integration with XClarity Administrator for centralized lifecycle and telemetry management across large AI infrastructure deployments.

Designed specifically for buyers across UAE, GCC, EMEA, and APAC regions, the SR780a V3 with Gaudi 3 OAM is positioned as a compelling alternative for organizations building or scaling AI compute clusters on open-standard interconnects. The server is compatible with industry-standard orchestration frameworks including Kubernetes and supports Intel's Gaudi software stack, which includes optimized libraries for PyTorch and TensorFlow workloads, enabling enterprise AI teams to deploy and manage large language models and multimodal AI pipelines with production-grade reliability.

Ideal for

• Large language model (LLM) pre-training and fine-tuning at scale using open-standard RoCE-based AI compute clusters
• Generative AI inference serving for enterprise applications requiring high throughput and low latency across concurrent requests
• Computer vision and multimodal AI model training for industrial inspection, autonomous systems, and media analytics
• High-performance deep learning research and experimentation in academic, government, and enterprise R&D environments
• AI-as-a-Service infrastructure build-out for cloud service providers and managed AI platform operators across EMEA and APAC
• Federated and distributed training workloads requiring scalable, standards-based networking without proprietary fabric lock-in

Technical specifications

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