Lenovo ThinkSystem SR670 V2 Intel Gaudi 2 GPU Solution

The Lenovo ThinkSystem SR670 V2 Intel Gaudi 2 GPU Solution is a purpose-built AI training and inference platform combining the 2U/4U ThinkSystem SR670 V2 server architecture with Intel's Habana Gaudi 2 deep learning accelerators. The Gaudi 2 accelerator is fabricated on TSMC's 7nm process and integrates 24 fully programmable Tensor Processor Cores (TPCs) alongside Matrix Multiplication Engines (MMEs), delivering high throughput for large-scale deep learning model training. Each Gaudi 2 accelerator includes 96 GB of HBM2E memory with approximately 2.45 TB/s of memory bandwidth, enabling enterprises to train and run inference on large language models and computer vision workloads with high memory capacity per accelerator.

The SR670 V2 chassis is engineered to support multiple Gaudi 2 accelerators in a dense, thermally optimized 2U form factor, with high-speed on-board RoCE v2-capable Ethernet interconnects integrated directly into the Gaudi 2 die — 24 x 100 GbE ports per accelerator — eliminating the need for external InfiniBand networking fabrics for scale-out configurations. This tight integration of compute and networking fabric simplifies datacenter cabling, reduces latency between accelerators in multi-node training jobs, and lowers overall infrastructure complexity. The solution supports Intel's SynapseAI software framework, which provides compiler, runtime, and performance libraries optimized for Gaudi 2, along with compatibility with PyTorch and TensorFlow through Habana's bridge libraries.

Designed for enterprise datacenters and high-performance AI compute clusters, the ThinkSystem SR670 V2 Intel Gaudi 2 GPU Solution carries Lenovo's enterprise-grade build quality, Lenovo XClarity management integration for lifecycle and system management, and a configurable topology (CTO — Configure To Order, part number 7D2VCTO1WW) allowing organizations to tailor CPU, memory, storage, and accelerator count to specific workload requirements. The solution is qualified and available through Omnixon Global for enterprise buyers across UAE, GCC, EMEA, and APAC regions.

Ideal for

• Large language model (LLM) pre-training and fine-tuning at scale across multi-node Gaudi 2 accelerator clusters using RoCE v2 fabric
• Enterprise AI inference serving for NLP, recommendation systems, and generative AI models requiring high HBM2E memory capacity per accelerator
• Computer vision model training for manufacturing quality inspection, medical imaging analysis, and autonomous systems development
• Financial services deep learning workloads including risk modeling, fraud detection, and time-series forecasting at datacenter scale
• Research and HPC environments requiring dense, programmable AI compute with open software stack compatibility (PyTorch, TensorFlow via SynapseAI)
• Hyperscale and enterprise datacenter AI infrastructure buildouts where simplified scale-out networking without external InfiniBand switches reduces complexity

Technical specifications

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