Lenovo ThinkSystem SR680a V3 Intel Gaudi 3 GPU Solution

The Lenovo ThinkSystem SR680a V3 Intel Gaudi 3 GPU Solution is a purpose-built, high-density AI accelerator server designed for large-scale deep learning training and inference workloads. At its core, the system integrates Intel Gaudi 3 AI accelerators, which are built on a 5nm process node and deliver substantial improvements in compute throughput and memory bandwidth compared to the previous generation. Each Gaudi 3 accelerator features 64 tensor processor cores (TPC), dedicated matrix multiplication engines (MME), and 128 GB of HBM2e memory per accelerator, enabling the system to handle massive model sizes and complex neural network architectures with high efficiency.

The SR680a V3 chassis is engineered to house up to eight Intel Gaudi 3 accelerators in a dense 8U form factor, interconnected via a high-bandwidth RoCE-based fabric using 24 integrated 200 Gbps Ethernet ports per accelerator. This tightly coupled interconnect architecture supports efficient all-reduce and collective communication operations critical for distributed AI training across multi-node clusters. The system also supports dual Intel Xeon Scalable processors (5th generation, codenamed Emerald Rapids) and high-capacity DDR5 memory, providing the CPU-side compute headroom needed to feed AI workloads without bottlenecks.

Positioned for enterprise datacenters, hyperscale AI infrastructure, and sovereign AI deployments, the ThinkSystem SR680a V3 is compatible with industry-standard software frameworks including PyTorch and TensorFlow through Intel's SynapseAI SDK. The platform supports scale-out configurations via standard Ethernet networking, avoiding proprietary fabric dependencies and simplifying integration into existing datacenter network topologies. With its combination of open ecosystem compatibility, dense accelerator packaging, and enterprise-grade Lenovo XClarity management integration, the SR680a V3 represents a comprehensive solution for organizations deploying generative AI, large language model training, and high-performance AI inference at scale.

Ideal for

• Large language model (LLM) pre-training and fine-tuning at scale, including models with hundreds of billions of parameters distributed across multi-node GPU clusters
• Generative AI inference serving for enterprise applications requiring high throughput and low latency, such as real-time content generation and AI-assisted enterprise search
• Computer vision and multimodal AI model training for industries including healthcare imaging, autonomous systems, and industrial quality inspection
• High-performance scientific computing and simulation workloads that benefit from tensor-accelerated numerical methods in research and engineering environments
• Sovereign AI datacenter deployments requiring open-standard networking and full stack transparency for regulated industries such as government, finance, and defense
• Enterprise MLOps pipelines requiring scalable, manageable accelerator infrastructure with centralized lifecycle management via Lenovo XClarity Administrator

Technical specifications

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