Supermicro AS-8125GS-TNMR2 Intel Gaudi 2 Server

The Supermicro AS-8125GS-TNMR2 (system model SYS-821GS-TNMR2) is a purpose-built AI training and inference server integrating Intel Gaudi 2 deep learning accelerators into a dense 8U rackmount chassis. The system is designed around Intel's Gaudi 2 HL-225H mezzanine compute cards, which are interconnected via a high-bandwidth on-board fabric using 21 x 100GbE RDMA-capable ports per accelerator — 24 internal ports for all-to-all server-level communication and 3 external ports for scale-out networking — eliminating the need for a separate high-speed switch fabric for many training cluster configurations.

Each Intel Gaudi 2 accelerator delivers substantial matrix multiplication throughput for mixed-precision AI workloads, supporting BF16 and FP32 natively, and is equipped with 96 GB of HBM2e memory per accelerator with up to 2.45 TB/s of aggregate memory bandwidth across the full complement of 8 accelerators in the system. The platform is validated for use with Intel's Habana SynapseAI software stack, providing support for popular deep learning frameworks including PyTorch and TensorFlow through a mature, production-grade software ecosystem.

Targeted at enterprise datacenters, cloud service providers, and AI research institutions, the AS-8125GS-TNMR2 supports dual 4th Gen Intel Xeon Scalable (Sapphire Rapids) processors to provide ample CPU-side compute for data preprocessing, orchestration, and host-side inference tasks. The system's architecture is optimized for large language model (LLM) training, generative AI model development, and high-throughput inference at scale, making it a competitive platform for organizations pursuing serious AI infrastructure buildouts across the UAE, GCC, EMEA, and APAC regions.

Ideal for

• Large language model (LLM) pre-training and fine-tuning at scale using distributed data-parallel and model-parallel strategies across Gaudi 2 accelerator clusters
• Generative AI application development and production inference serving for text, image, and multimodal foundation models
• High-throughput deep learning research requiring dense HBM2e memory capacity for training transformer-based architectures with large batch sizes
• Enterprise AI platform consolidation, enabling multiple simultaneous training jobs within a single high-density 8U chassis to maximize datacenter floor space efficiency
• Scale-out AI cluster deployments leveraging Gaudi 2's native 100GbE RDMA fabric for low-latency, high-bandwidth inter-node communication without additional proprietary networking hardware
• Regulated-industry AI workloads in financial services, healthcare, and government requiring on-premises accelerated compute with full data sovereignty

Technical specifications

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