GIGABYTE G593-ZX1 GPU Server (NVIDIA HGX H200)

The GIGABYTE G593-ZX1 is a 5U AMD EPYC-based GPU server engineered around the NVIDIA HGX H200 SXM platform, integrating eight NVIDIA H200 SXM5 GPUs interconnected via NVLink and NVSwitch fabric. Built on AMD's 4th Generation EPYC (Genoa) processor architecture, the system delivers exceptional parallel compute density suited to the most demanding generative AI, large language model training, and high-performance inference workloads at scale.

Each H200 SXM5 GPU features 141 GB of HBM3e memory with a 4.8 TB/s memory bandwidth per GPU, representing a substantial uplift over its predecessor for memory-intensive transformer model operations. The HGX H200 baseboard enables full NVLink interconnect bandwidth between all eight GPUs, allowing the server to function as a unified 1,128 GB HBM3e memory pool for exceptionally large model parameter sets. High-speed PCIe Gen 5 host connectivity and support for multiple OCP NIC 3.0 mezzanine adapters ensure the GPU complex can be fed data at rates commensurate with its compute throughput.

The G593-ZX1 targets enterprise data center operators, cloud service providers, and AI research institutions requiring a dense, rack-optimised platform for multi-node LLM training clusters and high-throughput inference serving. Its 5U chassis accommodates dual AMD EPYC processors alongside the HGX H200 module, NVMe storage expansion, and redundant power infrastructure, making it a complete, self-contained AI compute node ready for integration into InfiniBand or high-speed Ethernet scale-out fabrics.

Ideal for

• Large language model pre-training and fine-tuning requiring unified high-bandwidth GPU memory pools exceeding 1 TB HBM3e
• High-throughput generative AI inference serving for enterprise applications demanding low-latency, high-concurrency token generation
• Multimodal foundation model development combining vision, language, and structured data across distributed multi-node GPU clusters
• Scientific simulation and computational fluid dynamics workloads leveraging FP64 and FP8 mixed-precision capabilities of H200 Tensor Cores
• Recommender system training at hyperscale, exploiting the high memory bandwidth of HBM3e to accelerate embedding table operations
• AI-driven drug discovery and molecular dynamics research requiring sustained double-precision throughput across large parameter spaces

Technical specifications

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