HPE ProLiant Compute DL384 Gen12 Server

The HPE ProLiant Compute DL384 Gen12 is a purpose-built AI accelerated server engineered around the NVIDIA GH200 Grace Hopper Superchip, delivering a tightly integrated CPU-GPU architecture on a single unified memory fabric. The GH200 Superchip combines a 72-core NVIDIA Grace Arm-based CPU with an NVIDIA H200 Tensor Core GPU interconnected via NVLink-C2C, providing 900 GB/s of chip-to-chip bandwidth and a coherent shared memory space of up to 624 GB (480 GB HBM3e + 144 GB LPDDR5X). This architecture eliminates traditional PCIe bottlenecks between processor and accelerator, making the DL384 Gen12 particularly effective for memory-intensive generative AI and large language model inference and fine-tuning workloads.

Designed for modern AI infrastructure deployments, the DL384 Gen12 supports high-speed networking requirements essential for distributed training and multi-node inference clusters. The system is managed through HPE's iLO 7 embedded management controller, integrating with HPE's broader software ecosystem including HPE GreenLake for hybrid cloud operations and HPE InfoSight for predictive infrastructure analytics. The server's dense compute-to-footprint ratio makes it well suited for data center operators seeking to maximize accelerated compute density without disproportionate expansion of floor space or power distribution infrastructure.

As a rack-optimized platform, the HPE ProLiant DL384 Gen12 (Manufacturer Part Number P65399-B21) is positioned for enterprise organizations and research institutions building out scalable generative AI pipelines, retrieval-augmented generation systems, and high-throughput LLM serving infrastructure. HPE's ProLiant Gen12 platform introduces updated security capabilities including silicon root of trust and firmware integrity enforcement, ensuring the server meets enterprise-grade security and compliance requirements across regulated industries.

Ideal for

• Large language model inference serving for enterprise generative AI applications requiring high memory bandwidth and low-latency token generation
• LLM fine-tuning and parameter-efficient training (LoRA, QLoRA) on proprietary datasets where unified CPU-GPU memory simplifies data pipeline complexity
• Retrieval-augmented generation (RAG) pipelines combining vector database queries with real-time model inference at scale
• High-performance scientific computing and simulation workloads that benefit from the Grace CPU's high-bandwidth LPDDR5X memory subsystem
• AI model development and MLOps platforms requiring a standardized, enterprise-managed accelerated compute node within HPE GreenLake environments
• Multi-node distributed AI training clusters where NVLink and high-speed fabric interconnects are used to scale across multiple DL384 Gen12 nodes

Technical specifications

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