Xtreme MGX AI-A2U High-Performance GPU Server
The Xtreme MGX AI-A2U is an enterprise-grade 2U NVIDIA MGX™ modular GPU server engineered for artificial intelligence, deep learning, machine learning, and high-performance computing (HPC) workloads. Powered by dual AMD EPYC™ 9005 / 9004 Series processors and designed to support up to 4 dual-slot PCIe Gen5 graphics accelerators (including NVIDIA H200 NVL with NVLink bridging and RTX PRO 6000 Blackwell), this platform delivers high-density GPU acceleration in a modular, scalable form factor.
Featuring advanced multi-channel DDR5 memory support, extensive high-speed PCIe Gen5 interconnects, versatile Gen5 storage bays, dual 10GbE networking, and high-capacity redundant Titanium power supplies, the Xtreme MGX AI-A2U provides a resilient, high-throughput foundation for enterprise AI clusters, large language model training, and advanced data analytics.
NVIDIA MGX Modular Design: Built on the NVIDIA MGX™ modular server architecture for flexible, high-density GPU configurations optimized for AI and HPC.
4-GPU Acceleration: Supports up to 4 dual-slot PCIe Gen5 GPU accelerators, including NVIDIA H200 NVL PCIe GPUs with 2-way NVLink bridges and RTX PRO 6000 Blackwell Server Edition.
Dual-Socket Compute Architecture: Powered by dual AMD EPYC™ 9005 / 9004 Series Processors (Socket SP5) with cTDP up to 400W (up to 500W at 30°C ambient) for ultra-fast multi-threaded processing.
High-Speed DDR5 Memory: Supports high-capacity multi-channel DDR5 RDIMM memory architecture (24 DIMM slots) delivering maximum data throughput.
Advanced Gen5 Interconnects: Features 4 x FHFL PCIe Gen5 x16 slots dedicated for GPUs plus 3 additional FHFL PCIe Gen5 x16 expansion slots.
Flexible Gen5 Storage: Equipped with 6 x 2.5" Gen5 NVMe / SATA / SAS-4 hot-swap drive bays and 3 x M.2 slots for rapid OS boot and local storage.
High-Speed Networking: Equipped with dual 10GbE LAN ports (Intel® X550-AT2) and dedicated management LAN for secure remote administration.
Redundant Power Delivery: Equipped with 3+1 2000W 80 PLUS Titanium redundant power supplies designed to sustain multi-GPU power requirements.
Integrated BMC Remote Management: Integrated ASPEED AST2600 Baseboard Management Controller with dedicated LAN port for out-of-band monitoring and remote administration.
Form Factor
• 2U Rackmount NVIDIA MGX™ Modular GPU Server
Processor
• Dual AMD EPYC™ 9005 / 9004 Series Processors
• Socket SP5 (LGA 6096)
• cTDP up to 400W (up to 500W at 30°C ambient)
GPU Expansion
• Supports up to 4 x dual-slot PCIe Gen5 graphics cards / GPU accelerators
• Optimized for NVIDIA H200 NVL with 2-way NVLink bridges
Memory
• 24 x DDR5 RDIMM slots
• 12-channel architecture per processor
ExpansionSlots
• 4 x FHFL PCIe Gen5 x16 slots dedicated for full-height dual-slot GPUs
• 3 x additional FHFL PCIe Gen5 x16 expansion slots
Storage
• 6 x 2.5" Hot-Swap Gen5 NVMe / SATA / SAS-4 drive bays
• 3 x M.2 slots (PCIe Gen3)
Networking & I/O
• 2 x 10GbE LAN ports (Intel® X550-AT2)
• 1 x Dedicated 1GbE Management LAN port
• Video output via ASPEED AST2600 controller
Power Supply
• 3+1 2000W 80 PLUS Titanium Redundant Power Supplies
The Xtreme MGX AI-A2U is engineered for enterprise data centers, AI labs, and high-performance computing environments requiring modular GPU density for:
• Artificial Intelligence & Deep Learning model training and inference
• Large Language Model (LLM) development and deployment
• High-Performance Computing (HPC) & complex scientific simulations
• Visual computing, 3D rendering, and GPU cloud infrastructure
• NVIDIA OVX and enterprise AI clusters
Combining dual AMD EPYC™ 9005/9004 processing, NVIDIA MGX modular design with up to 4 x PCIe Gen5 GPU acceleration, high-density DDR5 memory, flexible Gen5 storage, dual 10GbE networking, and redundant 2000W Titanium power supplies, the Xtreme MGX AI-A2U delivers a robust, scalable compute foundation tailored for modern AI and enterprise computing environments.
