QCT QuantaGrid G3-8OAM Gaudi 3 Server System

The QCT QuantaGrid G3-8OAM is a purpose-built AI accelerator server system designed around Intel's Gaudi 3 architecture, integrating eight Intel Gaudi 3 OAM (Open Accelerator Module) mezzanine accelerators into a high-density 5U chassis. The Gaudi 3 architecture delivers a significant generational leap in AI compute throughput, featuring 128 GB of HBM2e memory per OAM module, 64 matrix multiplication engines (MMEs), and 8 TPC (Tensor Processing Core) clusters per accelerator. The platform leverages Intel's 5 nm process node and provides high-bandwidth, low-latency scale-out connectivity through integrated 24-port 200 GbE RoCE networking per accelerator, enabling direct server-to-server communication without requiring a separate networking switch fabric for smaller clusters.

The G3-8OAM system is engineered for demanding generative AI training and large-scale inference workloads, delivering competitive FP8 and BF16 compute performance across the full OAM array. The server chassis accommodates dual Intel Xeon Scalable host CPUs, providing substantial system memory capacity and PCIe Gen 5 host connectivity to the accelerator subsystem. The OAM form factor and modular architecture align with OCP (Open Compute Project) standards, supporting ecosystem-wide tooling and operational consistency. The system ships with support for Intel Gaudi software stack, including the SynapseAI SDK, enabling compatibility with PyTorch and other major ML frameworks through a hardware abstraction layer.

Targeted at enterprise AI infrastructure, cloud service providers, and large-scale datacenter deployments across EMEA, GCC, UAE, and APAC regions, the QuantaGrid G3-8OAM is positioned as a full-system solution for organizations requiring scalable, standards-aligned AI compute. QCT's system integration brings together validated thermal management, power delivery rated for high sustained accelerator TDP, and datacenter-ready manageability interfaces including IPMI and Redfish-compliant BMC. Omnixon Global provides access to this platform for enterprise buyers across the UAE, GCC, EMEA, and APAC territories.

Ideal for

• Large language model (LLM) pre-training and fine-tuning at scale, leveraging the combined HBM2e memory capacity and high-bandwidth inter-accelerator fabric across all eight Gaudi 3 OAM modules
• Generative AI inference serving for production enterprise deployments requiring high-throughput FP8 and BF16 tensor operations with deterministic latency
• Deep learning recommendation model (DLRM) training for e-commerce, media, and enterprise personalization platforms requiring dense embedding table operations
• Computer vision and multimodal AI model training pipelines in healthcare imaging, autonomous systems, and industrial quality inspection
• High-performance AI research and experimentation environments within enterprise datacenter or HPC clusters, utilizing RoCE-based scale-out networking for multi-node training jobs
• Sovereign AI infrastructure buildouts for government, financial services, and regulated industries in UAE, GCC, and EMEA requiring locally operated, enterprise-grade AI compute

Technical specifications

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