HPE ProLiant Compute DL384 Gen12
NVIDIA GH200 NVL2 Platform — Grace Hopper AI Inferencing
Our Price: Request a Quote
Our Price: Request a Quote
Our Price: Request a Quote
Click here to jump to more pricing!
Please Note: All Prices are Inclusive of GST. The DL384 Gen12 is a factory-configured system — additional PCIe, OCP, drive, and networking configuration options are available; contact us for a tailored quote.
Overview:
Do you need to run inference on large language models that demand enormous memory capacity and bandwidth?
The HPE ProLiant Compute DL384 Gen12 is the first HPE ProLiant server enabled with the NVIDIA GH200 NVL2 platform, supporting up to two NVIDIA GH200 Grace Hopper Superchips. Each superchip tightly couples an NVIDIA Grace CPU with an integrated NVIDIA Hopper GPU over NVIDIA's coherent NVLink-C2C fabric, and the dual-superchip NVL2 configuration links both superchips together for roughly double the memory and performance of a single-superchip system.
HPE positions the DL384 Gen12 as delivering the best performance per GPU in its ProLiant lineup, suited to mixed or memory-intensive workloads spanning both AI and traditional HPC. It carries the same HPE iLO management and Silicon Root of Trust security foundation used across the broader Gen12 portfolio, so it slots into an existing HPE fleet without new management tooling. HPE cites up to 2x higher inference performance versus the previous-generation H100, and the platform is NVIDIA OVX™ certified for AI workloads — well suited to large language model inferencing, fine-tuning, retrieval-augmented generation (RAG), and large-scale simulation, EDA, and weather-forecasting workloads.
What’s New
- The first HPE ProLiant rack-mount server with the NVIDIA GH200 Grace Hopper™ Superchip.
- Support for dual superchips via NVIDIA GH200 NVL2, with NVLink between the two GH200s for roughly double the memory and performance of a single-superchip system.
- Support for the latest NVIDIA InfiniBand, Ethernet, and BlueField adapters to keep an AI fabric running at full speed.
- NVIDIA OVX™ certified for AI workloads.
- High-performance LLM inference built to maximise data centre utilisation, with up to 2x the inference performance of H100.
- HPE Silicon Root of Trust, built on HPE's zero-trust architecture, to protect infrastructure, workloads, and data.
Features:
Large Coherent Memory for Large Models
The NVIDIA GH200 Grace Hopper architecture unifies CPU and GPU memory over a high-bandwidth coherent fabric, giving inference workloads far more addressable memory than a conventional discrete-GPU server.
A dual-superchip NVL2 configuration provides up to roughly 1.2 TB of combined coherent memory (LPDDR5X CPU memory plus HBM3e GPU memory) with up to 5 TB/s of aggregate bandwidth — enough headroom for large language models, LLM fine-tuning, retrieval-augmented generation, and simulation workloads that would exhaust the memory of a standard GPU server.
Next-Level Security, Chip to Cloud
HPE's Silicon Root of Trust establishes a zero-trust security framework at the silicon level to help ensure firmware integrity, continuously checking for compromised firmware and preventing an affected server from booting.
This security approach is applied consistently across the HPE ProLiant Compute Gen12 portfolio, so the DL384 Gen12 carries the same chip-to-cloud protection as the rest of your Gen12 fleet.
Enterprise Manageability and Lifecycle
Manage the DL384 Gen12 with the same HPE tooling as the rest of your fleet, including HPE iLO 6 and HPE Compute Ops Management for visibility, automation, and simplified lifecycle management.
Available as a standalone purchase or as a service through HPE GreenLake, the DL384 Gen12 integrates into an existing HPE environment without requiring new management tooling.
High-Speed Storage and I/O
Up to 8 EDSFF NVMe Gen5 (E3.S) drives deliver the sustained throughput needed to feed large-model inference pipelines, with 2 dedicated M.2 boot devices (not part of a hardware RAID array) keeping the OS separate from workload storage.
Up to 4 PCIe Gen5 x16 full-height, half-length slots and up to 2 OCP 3.0 slots provide flexible, high-bandwidth networking and I/O expansion — including support for the latest NVIDIA InfiniBand, Ethernet, and BlueField DPU adapters — for demanding AI fabric requirements.
Technical Specifications:
| HPE ProLiant Compute DL384 Gen12 — NVIDIA GH200 NVL2 Platform Specifications | |
|---|---|
| Form factor | 2U standard 19-inch rack design, air cooled |
| Superchip configuration | Single NVIDIA GH200 Grace Hopper Superchip, or dual superchip via NVIDIA GH200 NVL2 |
| CPU (per superchip) | NVIDIA Grace — 72 Arm Neoverse V2 cores; base/all-core SIMD frequency 3.1GHz / 3.0GHz. Dual-superchip NVL2 totals 144 cores. |
| GPU (per superchip) | Integrated NVIDIA Hopper GPU |
| CPU memory | 480GB LPDDR5X per Grace CPU (960GB total in dual-superchip NVL2); up to 512GB/s bandwidth per CPU (1024GB/s NVL2) |
| GPU memory | 144GB HBM3e per Hopper GPU (288GB total NVL2, up to 4.9TB/s per GPU, up to 9.8TB/s NVL2) |
| Combined coherent memory | Up to ~1.2TB coherent CPU+GPU memory in a dual-superchip (NVL2) configuration |
| CPU-to-GPU interconnect | NVIDIA NVLink-C2C: 900GB/s per superchip (1800GB/s combined, NVL2); NVLink between superchips: 900GB/s |
| Storage | Up to 8 hot-plug EDSFF NVMe Gen5 (E3.S) drives (max ~122.8TB using 8x15.36TB); up to 2 M.2 NVMe boot drives (not in hardware RAID) |
| Expansion slots | Up to 4 PCIe Gen5 x16 full-height half-length (FHHL) slots; up to 2 OCP 3.0 x16 slots |
| Power supplies | HPE 1800W–2200W Flex Slot Titanium Hot Plug power supplies — 2/3/4 supplies for single superchip (N+0/N+1/N+2); 4 supplies for dual superchip (N+1) |
| Dimensions | 789mm (31.08in) D × 448mm (17.64in) W × 87.5mm (3.44in) H (2U) |
| Weight (fully populated) | Single superchip: up to 26kg (58lbs); dual superchip: up to 31kg (69lbs) |
| Maximum heat dissipation | 3048 Watts (10.4k BTU/hr) — fully populated dual-superchip NVL2 configuration |
| Infrastructure management | HPE iLO 6, iLO RESTful API (Redfish), UEFI Class 3, HPE Compute Ops Management |
| Security | HPE Silicon Root of Trust |
| Operating system support | Certified for RHEL 9.4, Ubuntu 24.04, and SLES15 SP5 QU1 (Arm-based Grace platform — refer to the HPE Servers Support & Certification Matrix for the latest list) |
| Warranty | 3-year parts, 3-year labor, 3-year onsite support with next business day response |
Note: This is a factory-configured system — the fundamental superchip, PCIe, OCP, and drive configuration is set at the factory and is not field-upgradeable. Refer to the HPE ProLiant Compute DL384 Gen12 QuickSpecs for the full configuration matrix.
Hardware Specs:
HPE ProLiant Compute DL384 Gen12 server
Front View
Internal View
Rear View
Documentation:
Download the HPE ProLiant Compute DL384 Gen12 Server Quick Specs (.PDF)
*NVIDIA, Grace, Hopper, GH200, and OVX are trademarks and/or registered trademarks of NVIDIA Corporation. Arm and Neoverse are trademarks of Arm Limited. All other third-party marks are property of their respective owners.
Pricing Notes:
- All Prices are Inclusive of GST
- Pricing and product availability subject to change without notice.
- The DL384 Gen12 is sold only as a completely factory-assembled system — the base superchip/PCIe/OCP/drive configuration cannot be changed in the field. Contact us to confirm the exact configuration for your workload before ordering.
Our Price: Request a Quote
Our Price: Request a Quote
Our Price: Request a Quote
Our Price: Request a Quote
Our Price: Request a Quote
Our Price: Request a Quote
Our Price: Request a Quote
Our Price: Request a Quote
Our Price: Request a Quote