UCS C845A M8 (MGX)
A flexible enterprise AI server built on the NVIDIA MGX modular reference design.
A server whose NIC, GPU, and CPU configuration can be customized modularly via the NVIDIA MGX architecture. It supports 2–8 PCIe GPUs and handles diverse enterprise AI workloads such as inference, fine-tuning, and RAG.
What Is the UCS C845A M8 (MGX)
NVIDIA MGX is an open reference design that combines GPUs, CPUs, NICs, and storage as standard modules. The UCS C845A M8 adopts this MGX architecture as a flexible enterprise AI server, configurable with 2–8 PCIe GPUs to fit the workload. Rather than large-scale training, it addresses diverse enterprise AI tasks — inference, fine-tuning, RAG — on a single standard platform.
Key Capabilities
MGX modular design
Customize NIC, GPU, and CPU configuration modularly to match the workload
Flexible GPU expansion
2–8× NVIDIA/AMD PCIe GPUs cover everything from inference to fine-tuning
Blackwell generation support
Supports NVIDIA RTX PRO 4500 / 6000 Blackwell Server Edition
AI software integration
Integrates the NVIDIA AI Enterprise stack to standardize development and deployment
Key Specifications
| Item | Specification |
|---|---|
| Architecture | NVIDIA MGX open modular reference design |
| GPU | 2–8× NVIDIA/AMD PCIe GPUs |
| Supported GPUs (examples) | RTX PRO 4500 / 6000 Blackwell Server Edition |
| Software | NVIDIA AI Enterprise integration |
| Positioning | Flexible enterprise AI (inference · fine-tuning · RAG) |
What It's Used For
Enterprise inference services
Serve internal LLM and vision model inference with flexible GPU configurations
Fine-tuning & RAG
Mid-scale fine-tuning and internal knowledge RAG pipelines
Mixed workloads
Accommodate diverse departmental AI tasks on one standard server