Dell PowerEdge XE9680 8x Gaudi 3 PCIe Server

The Dell PowerEdge XE9680 configured with 8x Intel Gaudi 3 PCIe accelerators is a purpose-built AI training and inference server engineered for large-scale deep learning workloads in enterprise data centers. Built on the Intel Gaudi 3 architecture, each accelerator delivers high-throughput matrix multiplication and tensor operations optimized for transformer-based models, large language models, and generative AI pipelines. The system integrates tightly with Intel's open software ecosystem via the SynapseAI SDK, enabling compatibility with PyTorch and TensorFlow frameworks without requiring proprietary middleware lock-in.

The PowerEdge XE9680 chassis provides a high-density, dual-socket platform supporting Intel Xeon Scalable processors alongside the eight Gaudi 3 PCIe accelerators interconnected via Intel's scale-up fabric, enabling high-bandwidth peer-to-peer communication between accelerators within the node. Each Gaudi 3 device features 128 GB of HBM2e memory per card and dedicated RDMA-capable 200 Gb Ethernet ports built directly onto the accelerator, supporting scale-out cluster configurations without dependency on external InfiniBand switches. The server's thermal and power infrastructure is designed to sustain full accelerator utilization continuously under data center operating conditions.

Targeted at enterprise AI infrastructure teams, cloud service providers, and research institutions across UAE, GCC, EMEA, and APAC regions, the XE9680 8x Gaudi 3 PCIe represents a viable alternative to proprietary GPU ecosystems for organizations prioritizing open standards, software portability, and total infrastructure flexibility. Its rack-optimized form factor integrates with standard Dell OpenManage and iDRAC management tooling, enabling unified lifecycle management alongside existing PowerEdge fleet deployments.

Ideal for

• Large language model (LLM) pre-training and fine-tuning for enterprise generative AI platforms requiring multi-accelerator parallelism
• High-throughput AI inference serving for real-time NLP, computer vision, and recommendation engine workloads at production scale
• Multi-node distributed deep learning training clusters leveraging Gaudi 3's built-in RDMA Ethernet for scale-out without additional networking ASICs
• Enterprise MLOps pipeline acceleration supporting continuous model training, evaluation, and redeployment cycles in regulated industries
• Scientific computing and HPC simulation workloads in energy, life sciences, and financial modeling that benefit from high HBM2e memory bandwidth
• Sovereign AI infrastructure deployments in government and defense environments requiring on-premises AI compute with open-stack software control

Technical specifications

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