UCS C885A M8 (HGX)
An ultra-dense, 8-GPU HGX-based server purpose-built for large-scale LLM training.
A dense AI server packing 8× NVIDIA HGX H100/H200 (or AMD MI300X/MI350X) GPUs into a single 8U chassis. It delivers massive accelerated compute for LLM training, fine-tuning, and large-scale inference.
What Is the UCS C885A M8 (HGX)
NVIDIA HGX is an ultra-dense GPU baseboard that unites 8 SXM GPUs with NVLink and NVSwitch. The UCS C885A M8 packs this HGX into a single 8U chassis as a server dedicated to large-scale training, equipped with ultra-fast GPU-to-GPU communication and multiple high-speed NICs to deliver the intensive compute needed for LLM training, fine-tuning, and large-scale inference. Its performance is validated by MLPerf training benchmarks.
Key Capabilities
8-GPU HGX density
8× NVIDIA HGX H100/H200 (or AMD MI300X/MI350X) OAM GPUs in a single chassis
High-performance CPU & memory
Dual AMD EPYC (up to 96 cores) · 24× DDR5 RDIMM
High-speed east-west NICs
Up to 8× ConnectX-7 / BlueField-3 SuperNICs for lossless GPU-to-GPU communication
Proven performance
8U form factor · validated by MLPerf training benchmarks
Key Specifications
| Item | Specification |
|---|---|
| GPU | 8× HGX H100/H200 or AMD MI300X/MI350X (OAM) |
| CPU | Dual AMD EPYC (up to 96 cores) |
| Memory | 24× DDR5 RDIMM |
| Networking | Up to 8× ConnectX-7 / BlueField-3 SuperNICs |
| Form factor | 8U · MLPerf training validated |
What It's Used For
Large-scale LLM training
Pre-training and fine-tuning models from billions to hundreds of billions of parameters
Dense inference factories
Inference services handling massive concurrent request volumes
HPC & scientific computing
Large-scale accelerated compute for simulation and research