Wiwynn SV7-G2 Gaudi 2 8-OAM Server

The Wiwynn SV7-G2 is a purpose-built 2U server platform engineered around Intel's Gaudi 2 AI accelerator architecture, hosting eight OAM (OCP Accelerator Module) form-factor Gaudi 2 processors. Designed for high-density AI training and inference at datacenter scale, the platform leverages Intel's Gaudi 2 third-generation deep learning silicon, which integrates 24 fully programmable Tensor Processor Cores (TPCs) and a Matrix Multiplication Engine (MME) per accelerator, delivering substantial throughput for large-scale neural network workloads. The eight OAM modules are interconnected via a high-bandwidth, low-latency all-to-all fabric using 21 x 100 GbE RoCE RDMA ports per Gaudi 2 device, enabling tightly coupled collective communication operations critical to distributed training jobs without reliance on a separate NVLink-equivalent switch tier.

Each Gaudi 2 OAM in the SV7-G2 platform includes 96 GB of HBM2e memory per device, aggregating to 768 GB of accelerator memory across the full eight-accelerator configuration. This memory capacity supports the residency of large language models, recommendation systems, and computer vision networks directly on-device, minimizing host-to-accelerator data movement. The server is built on an OCP-compliant chassis architecture with a high-efficiency power delivery subsystem and thermal management solution dimensioned for continuous full-load datacenter operation in standard 2U rack enclosures, making it well-suited for hyperscale and enterprise AI infrastructure deployments.

The SV7-G2 platform is optimized for use with Intel's open-source SynapseAI software stack, which provides framework integration with PyTorch and TensorFlow through Intel's Habana deep learning compiler. This enables enterprises to port and run standard AI workloads without requiring proprietary closed-stack toolchains. The server targets organizations building out on-premises AI training clusters, large-scale inference serving infrastructure, and generative AI development environments where dense compute, high interconnect bandwidth, and significant on-accelerator memory capacity are primary system design requirements.

Ideal for

• Large language model (LLM) pre-training and fine-tuning at scale, leveraging the eight-way Gaudi 2 all-to-all interconnect fabric for efficient distributed gradient synchronization across nodes
• High-throughput AI inference serving for generative AI applications, benefiting from 768 GB aggregate HBM2e memory to hold large model parameter sets on-accelerator
• Deep learning recommendation model (DLRM) training for enterprise e-commerce, digital advertising, and content personalization platforms requiring high embedding table capacity
• Computer vision and multimodal AI training pipelines for manufacturing quality inspection, medical imaging analysis, and autonomous systems development in on-premises datacenter environments
• Hyperscale AI research and MLOps cluster buildout, where OCP-compliant form factor and open SynapseAI software stack reduce infrastructure lock-in for enterprise IT teams
• Generative AI and foundation model development environments requiring dense accelerator memory and scalable inter-node RDMA fabric bandwidth within standard 2U rack footprints

Technical specifications

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