Asus ESC8000-E12P 8x PCIe Gen5 AI Server with Gaudi 3

The Asus ESC8000-E12P is a high-density 4U rack server engineered specifically for large-scale AI training and inference workloads, accommodating up to eight Intel Gaudi 3 AI accelerators connected via PCIe Gen5 interfaces. Built on the Intel Xeon Scalable platform (Sapphire Rapids-SP or successor), the system delivers exceptional AI compute throughput by pairing the host CPU architecture with Gaudi 3's dedicated matrix multiplication engines and high-bandwidth memory, enabling organizations to deploy serious deep learning pipelines without the complexity of proprietary interconnect fabrics.

Each Intel Gaudi 3 accelerator integrated into the ESC8000-E12P features on-chip RDMA-capable Ethernet networking, allowing the eight accelerators to communicate directly over a standard 200GbE network fabric without requiring additional InfiniBand infrastructure. This design lowers total infrastructure complexity while maintaining the low-latency, high-bandwidth inter-accelerator communication essential for distributed training of large language models and other transformer-based architectures. The server supports enterprise-grade memory configurations and multiple NVMe storage options to keep accelerators fully fed with data.

Designed for demanding datacenter environments across AI research, cloud service provider infrastructure, and enterprise AI centers of excellence, the ESC8000-E12P is validated for deployment in high-density rack configurations with appropriate thermal and power provisioning. Its PCIe Gen5 backplane ensures that each of the eight accelerator slots operates at full bandwidth, eliminating the bottlenecks that can emerge when GPU or accelerator cards are constrained by legacy PCIe generations. Omnixon Global makes this platform available to enterprise and datacenter buyers across the UAE, GCC, EMEA, and APAC regions.

Ideal for

• Large language model (LLM) pre-training and fine-tuning at scale using distributed multi-node Gaudi 3 clusters interconnected over 200GbE RDMA fabric
• High-throughput generative AI inference serving for enterprise applications requiring sustained token generation across multiple concurrent sessions
• Computer vision model training for healthcare imaging, autonomous systems, and industrial quality inspection pipelines
• Financial services risk modeling and real-time fraud detection using deep neural networks requiring low-latency batch inference
• Government and research institution AI supercomputing nodes integrated into on-premises HPC clusters without dependency on proprietary networking
• Cloud service provider and colocation operators building GPU-as-a-service infrastructure optimized around open-standard Ethernet interconnects

Technical specifications

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