HPE ProLiant DL384 Gen12 (NVIDIA GH200) Server

The HPE ProLiant DL384 Gen12 is a purpose-built AI infrastructure server engineered around the NVIDIA GH200 Grace Hopper Superchip, which integrates an NVIDIA Hopper GPU with an ARM-based NVIDIA Grace CPU via NVLink-C2C interconnect delivering 900 GB/s of chip-to-chip bandwidth. This architecture eliminates the traditional PCIe bottleneck between CPU and GPU, enabling unified memory access across up to 624 GB of combined HBM3 and LPDDR5X memory per GH200 module. The server is optimized for workloads that demand extreme memory capacity and bandwidth, particularly large language model inference and generative AI training where model weights and activation tensors exceed the addressable memory of conventional GPU accelerator nodes.

Designed within the HPE ProLiant Gen12 server family, the DL384 incorporates HPE's silicon root of trust, iLO 7 baseboard management controller, and support for HPE GreenLake cloud management, enabling consistent lifecycle, firmware, and security management across on-premises and hybrid deployments. The platform supports HPE's Compute Ops Management software for fleet-scale operations and integrates into existing enterprise data center environments with standard rack form factors and high-density power delivery.

The DL384 Gen12 targets enterprise AI teams, research institutions, and cloud service providers deploying foundation models, multimodal AI, high-performance computing simulation, and large-scale data analytics. Its GH200-centric design makes it particularly well suited for inference serving of models with tens to hundreds of billions of parameters, where low-latency token generation and high throughput are critical operational requirements.

Ideal for

• Large language model inference serving for enterprise generative AI applications requiring high-throughput, low-latency token generation with models exceeding 70 billion parameters
• Foundation model fine-tuning and continued pre-training where unified CPU-GPU memory capacity reduces the need for model parallelism across multiple nodes
• Multimodal AI workloads combining vision, language, and structured data that benefit from the GH200's high-bandwidth unified memory architecture
• High-performance computing and scientific simulation workloads that require tight CPU-GPU coupling and large addressable memory pools
• AI-driven drug discovery, genomics, and molecular dynamics research demanding sustained FP64 and FP8 compute throughput with massive dataset residency
• Enterprise AI platform consolidation where a smaller number of high-capacity GH200 nodes replaces larger clusters of conventional GPU servers to reduce data center footprint and operational complexity

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.