ASRock Rack 6U8X-EGS2 NVIDIA HGX H100 AI Server

The ASRock Rack 6U8X-EGS2 is a 6U rackmount AI server engineered around the NVIDIA HGX H100 8-GPU baseboard, delivering the full NVLink-interconnected H100 SXM5 GPU configuration for demanding large-scale generative AI, large language model training, and high-performance computing workloads. The system pairs eight NVIDIA H100 SXM5 80GB GPUs — connected via fourth-generation NVLink at up to 900 GB/s bidirectional GPU-to-GPU bandwidth — with dual Intel Xeon Scalable (Sapphire Rapids) processors and high-capacity DDR5 ECC registered memory, forming a balanced compute platform capable of sustaining the extreme memory bandwidth and floating-point throughput required by transformer-based model training at scale.

The 6U8X-EGS2 incorporates high-bandwidth network infrastructure to match the GPU compute density, supporting NVIDIA ConnectX-7 or equivalent 400GbE/NDR InfiniBand host channel adapters across multiple PCIe Gen 5 slots, enabling low-latency, high-throughput GPU-direct RDMA communications essential for multi-node distributed training clusters. Storage connectivity includes NVMe U.2 drive bays for fast local dataset staging, complemented by redundant hot-swap power supply units rated for the substantial power envelope of a fully loaded HGX H100 configuration. The thermal design employs a high-airflow chassis with optimized fan zones to sustain continuous full-load GPU operation in standard data-center rack environments.

Targeted at hyperscale operators, national laboratories, financial services firms, and enterprise AI infrastructure teams deploying LLM training or large-scale inference clusters, the 6U8X-EGS2 is designed for dense rack integration — typically deployed in conjunction with high-radix InfiniBand or 400GbE fabrics. Its 6U form factor accommodates the HGX H100 baseboard's mechanical and thermal requirements while maintaining reasonable rack-unit efficiency in environments where per-rack GPU density and inter-node bandwidth are primary design constraints.

Ideal for

• Large language model (LLM) pre-training and fine-tuning at scale, leveraging NVLink-connected H100 SXM5 GPUs for sustained FP8/BF16 tensor core throughput across multi-node clusters
• Generative AI inference serving for production deployments requiring high aggregate GPU memory capacity (640 GB HBM3 per node) to host very large model parameter sets without sharding overhead
• High-performance scientific computing and simulation workloads, including molecular dynamics, climate modeling, and computational fluid dynamics, benefiting from H100 FP64 double-precision performance
• Multi-node distributed deep learning research infrastructure within academic institutions or enterprise AI centers of excellence requiring dense, NVLink-optimized compute nodes
• Computer vision and multimodal foundation model training pipelines that demand high memory bandwidth and fast all-reduce collectives across eight tightly coupled GPUs
• Financial services risk analytics and quantitative modeling workloads requiring both high FP64 throughput and the ability to run mixed-precision inference pipelines on a single consolidated platform

Technical specifications

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