HPE ProLiant Compute DL384 Gen12 with NVIDIA GH200

The HPE ProLiant Compute DL384 Gen12 represents HPE's next-generation GPU-dense rack server platform, purpose-built to harness the transformative compute density of NVIDIA GH200 Grace Hopper Superchips. Each GH200 Superchip integrates a 72-core NVIDIA Grace Arm-based CPU with a Hopper-architecture H100 GPU through NVLink-C2C interconnect delivering 900 GB/s of chip-to-chip bandwidth, effectively collapsing the traditional CPU-to-GPU memory bottleneck. The DL384 Gen12 accommodates multiple GH200 modules within a 4U chassis footprint, enabling organizations to deploy extreme AI compute density per rack unit while leveraging HPE's enterprise-grade reliability, serviceability, and systems management ecosystem.

Targeted at enterprises, cloud service providers, national laboratories, and AI-native organizations, the DL384 Gen12 excels at workloads demanding massive memory bandwidth and tightly coupled CPU-GPU compute. Each GH200 Superchip exposes up to 96 GB of HBM3 GPU memory alongside up to 480 GB of LPDDR5X CPU memory, all accessible to the GPU across NVLink-C2C at unprecedented bandwidth. This unified memory architecture eliminates traditional PCIe bottlenecks and enables model parallelism across trillion-parameter large language models (LLMs) without memory-capacity constraints that plague conventional discrete GPU designs. Multi-node NVLink configurations via NVLink Switch further scale collective operations across DGX SuperPOD-style deployments.

From a platform management perspective, the DL384 Gen12 integrates HPE's iLO 7 Baseboard Management Controller with NVIDIA's GPU Telemetry Infrastructure, providing unified health monitoring, firmware lifecycle management, and AI-optimized thermal controls. HPE's high-density direct liquid cooling (DLC) options address the substantial thermal design power of the GH200 modules, supporting data center deployments transitioning from air to hybrid or fully liquid-cooled infrastructure. Full integration with HPE GreenLake for Compute Ops Management enables cloud-consistent operations, predictive analytics, and automated remediation across on-premises AI clusters.

Ideal for

• Training and fine-tuning large language models and multimodal foundation models with trillion-scale parameter counts requiring terabytes of aggregate GPU-accessible memory
• High-throughput generative AI inference serving for enterprise LLM deployments where low-latency token generation at scale demands maximum memory bandwidth per accelerator
• Molecular dynamics simulation, quantum chemistry, and computational fluid dynamics workloads in life sciences and engineering that benefit from the Grace CPU's memory capacity paired with Hopper GPU compute
• Seismic imaging and reservoir simulation in energy sector applications requiring sustained FP64 double-precision throughput at scale
• AI-accelerated genomics pipelines combining rapid CPU-side bioinformatics preprocessing with GPU-accelerated deep learning variant calling and protein structure prediction
• National laboratory and academic HPC clusters running mixed precision scientific computing that demands both high MPI performance from the Grace CPU mesh interconnect and Tensor Core acceleration for AI-enhanced simulations

Technical specifications

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