HGX H200 Optimized X13 8U 8GPU

Description

The HGX H200 Optimized X13 8U 8GPU is a state-of-the-art, high-performance server engineered for advanced artificial intelligence workloads, deep learning training, and large-scale language model inference. Designed with enterprise-grade components, cutting-edge GPUs, and optimized interconnects, this server delivers unparalleled computational power, memory bandwidth, and storage throughput, making it ideal for AI research labs, data centers, and high-performance computing (HPC) environments.

Chassis and General Specifications

• Form Factor: 8U rack-mountable server

• GPU Support: 8 NVIDIA H200 SXM5 Gen5 GPUs

• RoHS Compliant: Yes

• System Models: X13DGH-T-P.1560TS-R4K12B, RoHS SYS-821GE-TNHR and EWCSC

• Warranty: 3 years labor, 3 years parts, 1 year CRS under limited warranty

This server is engineered for maximum reliability and sustained performance under heavy AI and HPC workloads, ensuring long-term operational stability.

Processor (CPU)

• Model: Intel Xeon Platinum 8592+

• Configuration: Dual Processor (2P)

• Total Cores: 64 cores per CPU (128 cores total)

• Base Clock: 2.0 GHz

• TDP: 350W per processor

• Cache: 320MB L3

• SKU: P4X-EMR8530N-SRN6V-MCC (x2)

These processors provide exceptional parallelism and computational throughput, enabling the system to efficiently handle large-scale AI models and intensive scientific computations.

Memory (RAM)

• Module: MEM-DR564MC-ER64

• Type: DDR5-5600 ECC RDIMM 2Rx4

• Capacity per Module: 64GB

• Number of Modules: 16

• Total Memory: 1TB (1024GB)

With 1TB of ultra-fast DDR5 memory, the server can manage massive datasets and extremely large AI models while maintaining high efficiency and system integrity thanks to ECC support.

Storage

• Drive Model: HDS-25N4-003T8-E1-TXD-NON-008

• Type: 2.5″ NVMe PCIe 4.0 SSDs

• Capacity per Drive: 3.84TB

• Number of Drives: 8

• Enterprise Grade: 1 DWPD (Drive Writes Per Day), TLC

• Form Factor: 7mm

The server offers 30.72TB of high-speed NVMe storage, delivering low-latency, high-throughput data access critical for AI training, HPC workloads, and large-scale database operations.

Graphics Processing Units (GPUs)

• Model: NVIDIA H200 SXM5 Gen5

• Quantity: 8 GPUs

• Interconnect Bandwidth: 900GB/s via NVLink 5.0 / NVSwitch

• AI Optimization: Supports vLLM, Megatron-DeepSpeed, HuggingFace LLMOps

The GPUs provide extreme acceleration for AI workloads, enabling training and inference of massive language models with exceptional performance and efficiency.

Networking and Connectivity

• Network Card: AOC-MH25G-m2S

• Type: Dual-Port 25GbE SFP28 NIC

• Quantity: 2

• RoHS Compliant: Yes

This networking setup ensures ultra-fast data transfer and cluster communication, supporting distributed AI pipelines and supercomputing environments.

Additional Features

• TPM Module: AOM-TPM-9670V-P, SPI TPM 2.0 using SLB9670, RoHS

• RAID Support: Intel VROC Premium, RAID 0, 1, 5, 10 via AOC-VROCPREMOD

• Software License: SFT-DCMS-SINGLE, 1-year Soika Enterprise subscription

• Super Cluster Interconnection: Local cluster support for high-performance AI workloads

Key Advantages

• Optimized for AI training, inference, and very large language models (vLLM)

• Fully compatible with Megatron-DeepSpeed and HuggingFace LLMOps frameworks

• Enterprise-class reliability and durability, suitable for continuous heavy workloads

• RoHS compliant, environmentally responsible

• Ideal for AI research labs, data centers, and high-performance computing clusters

• Scalable for future AI model growth and large-scale deployment

Summary

The HGX H200 Optimized X13 8U 8GPU is a high-end AI server solution that combines dual Intel Xeon Platinum 8592+ processors, 1TB DDR5-5600 ECC memory, 8 NVIDIA H200 SXM5 GPUs, and enterprise-grade NVMe storage to deliver unmatched performance for AI and HPC workloads. With advanced networking, software optimizations, and cluster interconnect support, this system is designed to handle massive AI models, high-throughput training tasks, and next-generation computational workloads efficiently and reliably.