Penguin Computing Intel Gaudi 3 8-OAM Altus Server

The Penguin Computing Altus G3-8P is a high-density AI accelerator server built around eight Intel Gaudi 3 OAM (Open Accelerator Module) processors, designed to address the most demanding large-scale AI training and inference workloads in enterprise and hyperscale datacenter environments. Intel Gaudi 3 delivers a substantial generational leap over its predecessor, featuring a 64 MB on-chip SRAM, 128 GB HBM2e memory per OAM module (across the full 8-OAM configuration), and 24 Tensor Processor Cores per die alongside Matrix Multiplication Engines optimized for BF16 and FP8 precision. The architecture integrates 24 x 200 Gb/s RDMA-capable Ethernet ports per OAM for scale-out fabric connectivity, enabling high-bandwidth, low-latency communication across multi-node AI clusters without requiring proprietary interconnect hardware.

The Altus G3-8P platform from Penguin Computing is engineered for open-standards AI infrastructure, supporting the OCP Open Accelerator Infrastructure (OAI) form factor and offering deep integration with Intel's Gaudi software stack, including the Intel Gaudi PyTorch bridge and SynapseAI SDK. This allows enterprises to run leading AI frameworks—including PyTorch and TensorFlow—with optimized kernel libraries and model parallelism strategies across all eight accelerators. The server is positioned as a turnkey, rack-ready system that pairs the Gaudi 3 OAM modules with a validated host CPU platform, high-capacity system memory, and NVMe storage to deliver a complete, production-ready AI compute node.

Targeted at enterprise IT teams, national AI research institutions, and cloud service operators across the UAE, GCC, EMEA, and APAC regions, the Altus G3-8P represents a compelling alternative in the AI accelerator server segment. Its reliance on standard 200 GbE networking reduces fabric complexity, and the open-ecosystem software stack lowers long-term dependency risk. The system is suitable for generative AI model training, large language model (LLM) fine-tuning, computer vision pipelines, and high-throughput inference serving at scale.

Ideal for

• Large language model (LLM) pre-training and fine-tuning using distributed data and tensor parallelism across all eight Gaudi 3 OAM accelerators
• Generative AI inference serving for enterprise applications requiring high throughput and low latency at scale
• Computer vision and multimodal AI model training for manufacturing quality inspection, medical imaging, and autonomous systems
• High-performance scientific computing and simulation workloads that benefit from BF16/FP8 matrix acceleration
• Multi-node AI cluster deployment as a building block node, leveraging native 200 GbE RDMA scale-out without additional networking ASICs
• Sovereign AI and national research datacenter deployments in EMEA and APAC requiring open-standards, auditable AI infrastructure

Technical specifications

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