TBird HGX-4U High-Performance 8-GPU AI and HPC Server

The TBird HGX-4U is an enterprise-grade, high-density 4U rackmount server engineered for demanding large-scale artificial intelligence (AI), deep learning model training, massive LLM inference, and high-performance computing (HPC) workloads. Powered by dual AMD EPYC™ 9005 or 9004 series processors, this system provides massive compute capacity, ultra-fast memory bandwidth, and extensive PCIe Gen5 capabilities required for cutting-edge AI infrastructure.

At the core of the platform is integration with the NVIDIA HGX™ H200 or H100 8-GPU system board using high-speed NVLink interconnects. Combined with 3-phase high-efficiency redundant power supplies, flexible drive configurations, and comprehensive remote management options, the TBird HGX-4U provides an optimal foundation for enterprise AI deployments and high-density GPU computing clusters.

• 8-GPU Acceleration: Designed to support NVIDIA HGX™ H200 / H100 8-GPU baseboards via high-speed NVLink interconnects for massive parallel computing.
• Dual-Socket Processor Performance: Powered by dual AMD EPYC™ 9005 / 9004 Series processors with up to 128 cores per socket.
• High-Capacity DDR5 Memory: Features 24 x DIMM slots supporting 12-channel DDR5 RDIMM memory architecture per processor.
• PCIe Gen5 Expansion: Includes high-speed PCIe 5.0 x16 slots for networking, fabric adapters, and an OCP 3.0 mezzanine slot.
• Versatile Storage Configuration: Supports up to 8 x 2.5-inch Gen5 NVMe/SATA/SAS hot-swappable drive bays for fast data throughput.
• Redundant Power Design: Equipped with 4 x 3000W 80 PLUS Titanium redundant power supplies in 3+1 or 2+2 configurations.
• Advanced Remote Management: ASPEED AST2600 BMC with dedicated management port and hardware-level monitoring.

• 4U Rackmount Server Chassis (448W x 175.8H x 880D mm)

• Dual-socket SP5 platform
• Supports AMD EPYC™ 9005 and 9004 Series Processors (TDP up to 400W per socket)

• NVIDIA HGX™ H200 / H100 8-GPU baseboard
• High-speed NVIDIA NVLink interconnect support

• 12-channel DDR5 RDIMM memory architecture per processor
• 24 x DIMM slots (Supports up to 4800/6000 MT/s RDIMM dependent on CPU generation)

• Up to 8 x FHHL PCIe Gen5 x16 expansion slots for networking/fabric cards
• 1 x OCP 3.0 Gen5 x16 slot

• 8 x 2.5-inch Gen5 NVMe / SATA / SAS hot-swappable drive bays
• 2 x M.2 slots (PCIe Gen4 x4)

• 1 x 10/100/1000 Mbps Management LAN port
• Front and rear USB 3.2 Gen1 ports
• Mini-DP video output (integrated via ASPEED AST2600)

• Dual ROM architecture for firmware redundancy
• ASPEED AST2600 Baseboard Management Controller with GIGABYTE Management Console

• 4 x 3000W 80 PLUS Titanium redundant power supplies (3+1 / 2+2 redundant)

The TBird HGX-4U is engineered for enterprise datacenters, research institutions, and hyperscale environments running intensive workloads including:

• Large Language Model (LLM) training and fine-tuning
• Real-time conversational AI and enterprise AI inference
• Generative AI and deep learning neural network training
• High-performance scientific and molecular modeling
• Complex engineering simulations and physics computing

Combining 8-GPU HGX acceleration, dual AMD EPYC™ processors, high-capacity DDR5 memory, high-speed PCIe Gen5 fabric expansion, and high-efficiency Titanium power supplies, the TBird HGX-4U delivers a scalable and robust foundation for next-generation data centers. Its architecture provides high-density compute power capable of meeting the performance demands of modern AI and HPC applications.