Current Configuration

(List below serves as quick links to each section)

Chassis

Superchip

Boot Options

Networking

Drives EDSFF

Power Supply Units | PSU

ECS Warranty

The request has been sent.
Thank you.

Price as Configured:
$0

Error Submitting Cart. Please get in touch with us for support.

Please select all the mandatory items in the configuration list before submitting the cart.

One or more items in your cart has no price. This will be submitted as a "Request for a Quote".

HPE ProLiant Compute DL384 Gen12 Configurable

 

Multi-Quantity Discounts Available for ALL Line Items

** LARGE DISCOUNTS available for complete systems. Pricing displayed is for single item purchases **

Use the configuration tool to create your solution and submit it us for a quote OR send us a message with your complete build requirements to receive best pricing

READ ME

  • Some options may not be displayed until the compatible parent option is chosen ie. Chassis – Drives, Processor – RAM etc.
  • “Quote” items can be added to your cart and sent to us via the cart page
  • Click the blue bar to close/open a section from view after choosing your options

Chassis

Sold only as a completely factory-assembled system. Superchip count, drive count, PCIe and OCP layout are set at the factory and are not field-upgradeable.

Superchip

NVIDIA GH200

Boot Options (Optional)

Up to 2 M.2 NVMe SSDs on the M.2 pass-through module (not in hardware RAID). M.2 is PCIe Gen3, so HPE recommends M.2 for boot drives, not local cache. Boot can also be set in iLO to EDSFF NVMe or network PXE.

Networking (Optional)

PCIe slot count is set by the factory layout (1-2 slots single GH200, 2-4 slots dual GH200). BlueField: one B3220 P66386-H21 north-south and up to four B3140H P66387-H21 east-west per server; other NIC/fabric adapters up to four per server. 200/400Gb cards have ambient-temperature limits per layout (QuickSpecs). DAC/transceivers sold separately.

Drives EDSFF (Optional)

Up to 8 EDSFF E3.S NVMe Gen5 drives with dual GH200 NVL2 (2 per superchip standard; 4 per superchip with the 4xEDSFF / 8xEDSFF factory layouts). Hardware RAID and hot-plug drives are not supported. No drives ship standard.

Power Supply Units | PSU

Single GH200: minimum 2 power supplies (2+0, 2+1 or 2+2). Dual GH200 NVL2: 4 power supplies required (3+1). One C13 power cord per power supply.

ECS Warranty

Helpful Tip: Once desired configuration is selected click "Add to Cart".
From the cart page you can submit a submit a quote request for best pricing

What’s New
  • 2U air-cooled HPE ProLiant Compute DL384 Gen12 – first HPE ProLiant rack server with the NVIDIA GH200 Grace Hopper™ Superchip
  • Single GH200 or dual superchips with NVIDIA GH200 NVL2 (NVLink between the two GH200)
  • Per superchip: NVIDIA Grace CPU with 72 Arm Neoverse V2 cores + NVIDIA Hopper GPU
  • Up to 480 GB LPDDR5X + 144 GB HBM3e per superchip – up to 1248 GB coherent memory with dual superchips
  • NVLink-C2C CPU-to-GPU bandwidth: 900 GB/s (1800 GB/s with GH200 NVL2)
  • Up to 8 EDSFF E3.S NVMe Gen5 drives + up to 2 M.2 NVMe boot drives
  • Up to 4 PCIe Gen5 x16 FHHL slots, or 2 FHHL + 2 OCP 3.0 (factory configuration)
  • NVIDIA InfiniBand, Ethernet and BlueField DPU adapters
  • Up to 4 HPE 1800W–2200W Flex Slot Titanium hot-plug power supplies
  • 6 hot-plug redundant fans
  • HPE iLO 6 with HPE Silicon Root of Trust

HPE DL384 Gen12 Base Configuration

The HPE ProLiant Compute DL384 Gen12 is a 2U NVIDIA GH200 platform optimized for AI inferencing of large language models that need large memory capacity, and for non-AI workloads such as large-scale simulation, EDA and weather forecasting. Every build starts with the CTO server plus one superchip kit (single GH200 or dual GH200 NVL2); the superchip kits include the enablement items for any supported configuration. The system is sold only as a completely factory-assembled server.

Factory configuration setting (choose one, set at the factory – not field-upgradeable):

  • Single GH200: 2 PCIe + 2 EDSFF P72451-B21 (recommended) · 1 PCIe + 1 OCP + 2 EDSFF P72454-B21 · 1 PCIe + 4 EDSFF P72457-B21
  • Dual GH200 NVL2: 4 PCIe + 4 EDSFF P72460-B21 (recommended) · 2 PCIe + 2 OCP + 4 EDSFF P72463-B21 · 2 PCIe + 8 EDSFF P72466-B21
NVIDIA GH200 Grace Hopper™ Superchip

CPU, GPU and memory are integrated on each superchip – there are no separately orderable processors or memory. The GH200 NVL2 option links two GH200 with NVLink for twice the memory and performance. Superchip table: QuickSpecs Standard Features.

GH200 (single superchip)
  • 72 Arm Neoverse V2 cores · 3.1 GHz base / 3.0 GHz all-core SIMD
  • L2 1 MB per core · L3 114 MB
  • 480 GB LPDDR5X · 512 GB/s
  • 144 GB HBM3e · up to 4.9 TB/s
  • NVLink-C2C CPU-to-GPU 900 GB/s
  • FP64 34 TFLOPS · FP8 Tensor Core 3958 TFLOPS (with sparsity)
GH200 NVL2 (dual superchip)
  • 144 Arm Neoverse V2 cores
  • L3 228 MB
  • 960 GB LPDDR5X · 1024 GB/s
  • 288 GB HBM3e · up to 9.8 TB/s
  • NVLink-C2C CPU-to-GPU 1800 GB/s
  • FP64 68 TFLOPS · FP8 Tensor Core 7916 TFLOPS (with sparsity)
Cooling & Fans

The DL384 Gen12 is air cooled and populated with six dual-rotor, hot-plug redundant fans. Configurations that populate all available PCIe slots require a PCIe fan kit in slot 1, except dual superchip systems with 4x PCIe slots. Source: QuickSpecs Standard Features, User Guide Fan mode behavior / Network adapter guidelines.

Boot Device – M.2 NVMe

Up to two M.2 NVMe SSDs install on the dual-slot M.2 pass-through module (2280 and 22110). M.2 is PCIe Gen3 and not in hardware RAID, so HPE recommends M.2 for boot drives rather than local cache. Boot can also be set in iLO to the EDSFF NVMe drives or network PXE. Source: QuickSpecs Internal Storage Devices.

480GB NVMe Gen4 M.2 2280 PE9010 SSD 480GB M.2 NVMe
P80318-B21
960GB NVMe Gen4 M.2 2280 PE9010 SSD 960GB M.2 NVMe
P80321-B21
960GB NVMe Gen4 M.2 2280 Self-encrypting PE9010 SSD 960GB M.2 NVMe SED
P80327-B21
1.92TB NVMe Gen4 M.2 2280 PE9010 SSD 1.92TB M.2 NVMe
P80324-B21
EDSFF NVMe Storage

Up to 8 EDSFF E3.S NVMe Gen5 drives with dual GH200 NVL2. Each superchip supports 2 EDSFF drives standard, or 4 per superchip with the 4xEDSFF / 8xEDSFF factory settings (PCIe reduced to one slot per superchip). No drives ship standard; drive blanks are populated at the factory. Source: QuickSpecs Overview / Internal Storage Devices.

  • No hardware RAID: hardware RAID controllers are not supported this generation for either the M.2 or EDSFF NVMe drives, and hot-pluggable drives are not supported.
  • Single superchip: all drives are configured to the boot superchip.
  • Dual superchip – Unbalanced P72469-B21: drives fill the first GH200 (up to 4) before the second.
  • Dual superchip – Balanced P72472-B21: an even number of drives split equally across both GH200.
  • Drives may be upgraded in the field; the factory maximum drive / PCIe / OCP layout is not field-upgradeable.
Networking & OCP 3.0

Rear OCP 3.0 slots 21/22 are available with the 1 PCIe / 1 OCP (single GH200) or 2 PCIe / 2 OCP (dual GH200) factory settings. HPE supports one north-south BlueField B3220 DPU (P66386-H21) and up to four east-west B3140H (P66387-H21) per server; other NIC and fabric adapters are supported up to four per server. 200/400Gb cards have supported ambient-temperature limits per configuration – see QuickSpecs. DAC or fiber transceivers and cables are purchased separately; for RJ-45 select SFP-to-RJ-45 adaptor 453154-B21.

InfiniBand NDR200/Ethernet 200Gb 1-port OSFP MCX75310AAS-HEAT NDR200 200Gb 1p
P45642-H23
InfiniBand NDR/Ethernet 400Gb 1-port OSFP MCX75310AAS-NEAT NDR 400Gb 1p
P45641-H24
InfiniBand XDR/Ethernet 2x400GbE 1-port OSFP PCIe6 CX8 XDR 1p OSFP CX8
P79114-H21
InfiniBand XDR400/Ethernet 400GbE 2-port QSFP112 PCIe6 CX8 XDR400 2p CX8
P79115-H21
InfiniBand XDR Multi Host/Socket Direct Auxiliary Card XDR Aux Card
P81264-H21
Riser / PCIe Slots
  • All riser slots are x16 PCIe5 Full Height Half Length, rated up to 75 W each.
  • Primary riser: slots 1 and 2 · Secondary riser: slots 4 and 5 · slots 3 and 6 are not enabled.
  • Single superchip systems have one or two available PCIe slots; dual superchip systems have two to four.
  • Populate slots in order: Slot 4 > Slot 5 > Slot 1 > Slot 2.
  • Riser/slot layout is set by the factory configuration setting (see Base Configuration).
Power Supply Units | PSU

HPE Flex Slot Titanium hot-plug power supplies, up to 96% efficiency with a standard C-14 inlet. Choose one C13 power cord per power supply. Source: QuickSpecs Step 2.

HPE 1800W-2200W Flex Slot Titanium Hot Plug Power Supply Kit 1800W–2200W Titanium
P44712-B21

Single GH200: two power supplies are sufficient; three provide 2+1 and four provide 2+2 redundancy.

Dual GH200 NVL2: four power supplies are required (3+1 redundancy).

HPE iLO 6
Security
  • HPE Silicon Root of Trust
  • UEFI Secure Boot and Secure Start
  • TPM 2.0 embedded on the system board
Management & Access
  • Dedicated 1Gb iLO management port
  • iLO Service Port on the front for direct local access
  • iLO RESTful API (Redfish conformant)
  • RESTful Interface Tool and Scripting Tools

Monitoring & Health
  • HPE Active Health System
  • Integrated Management Log (IML)
  • Chassis intrusion detection switch connector
Platform notes
  • HPE iLO 6 ASIC; iLO Standard with Intelligent Provisioning (embedded)
  • Optional HPE iLO Advanced
  • Most commonly used iLO features are enabled on this Arm server – see the iLO 6 User Guide for exceptions
  • UEFI boot mode only (no Legacy mode) · rail kit ships standard

Base Standard Capabilities

Technical specificationsHPE ProLiant Compute DL384 Gen12
Processor typeNVIDIA GH200 Grace Hopper™ Superchip
Processor familyNVIDIA Grace CPU and Hopper GPU
Processor number1 superchip, or 2 superchips with NVIDIA GH200 NVL2
Processor core available72 Arm Neoverse V2 cores per superchip (144 cores with GH200 NVL2)
Processor speed3.1 GHz base / 3.0 GHz all-core SIMD (GH200)
Processor cache64 KB i-cache + 64 KB d-cache L1, 1 MB L2 per core, 114 MB L3 (228 MB with GH200 NVL2)
Memory typeLPDDR5X with ECC (CPU) and HBM3e (GPU), integrated on the superchip
Maximum memoryUp to 624 GB per superchip (480 GB LPDDR5X + 144 GB HBM3e); up to 1248 GB coherent memory with dual superchips
Memory slotsNone - memory is integrated on the GH200 superchip
Memory protection featuresError-correction code (ECC) on LPDDR5X
AcceleratorsOne NVIDIA Hopper GPU per superchip with up to 144 GB HBM3e (up to 4.9 TB/s); NVLink-C2C 900 GB/s CPU-to-GPU (1800 GB/s with GH200 NVL2)
System fan features6 hot plug redundant fans
Expansion slotsUp to four PCIe Gen5 x16 Full Height Half Length (FHHL) slots, or up to two PCIe FHHL and two OCP 3.0, set by factory configuration
Form factor2U rack, air cooled
Network controllerOptional - NVIDIA Ethernet (PCIe and OCP 3.0), InfiniBand NDR/XDR and BlueField DPU adapters. See QuickSpecs for network card choices.
Storage controllerHW RAID controllers are not supported this generation for either the M.2 or EDSFF NVMe drives
Drive supportedUp to 8 EDSFF E3.S NVMe Gen5 drives (dual superchip) plus up to 2 M.2 NVMe boot drives
Power supply typeUp to 4 HPE 1800W-2200W Flex Slot Titanium Hot Plug Power Supply Kits. Single GH200: 2+0, 2+1 or 2+2. Dual GH200 NVL2: 4 power supplies required (3+1).
Infrastructure managementHPE iLO Standard with Intelligent Provisioning (embedded) on HPE iLO 6; optional HPE iLO Advanced