Creative Compute
0

HPE ProLiant Compute DL384 Gen12

Configurable systemConfigure to orderManufacturer documented

Built around NVIDIA GH200 superchips, where the processor and accelerator share coherent memory rather than copying data between them. That matters for models too large to fit comfortably in GPU memory alone.

Ask Carl about this

No payment is taken online. Every request is an enquiry answered by an engineer.

Representative system image

Where does this sit?

  1. 01Local
  2. 02Workstation
  3. 03Departmental
  4. 04Enterprise
  5. 05Cluster
  6. 06Rack-scale

Enterprise. Four to eight enterprise accelerators in a rack chassis, sized for production load rather than experiments. Typically bought by an enterprise ai or platform team. Datacentre or a well-provisioned rack.

Fit

Is this the right thing for you?

Who it's for

  • Teams running models that exceed conventional GPU memory
  • Research groups working on memory-bound generative AI
  • Organisations evaluating the Grace Hopper architecture

What it runs

  • Generative AI training and inference
  • Memory-bound model serving
  • Graph and HPC workloads

When it is the wrong answer

  • General-purpose enterprise virtualisation
  • Anyone whose software has not been tested on Arm
Where it goes
Datacentre rack
Complexity
specialist
Cooling
Awaiting verification
Specification

What the numbers mean

Every figure below is explained in plain English. Switch to the technical view for the bare specification.

System

Architecture

Up to 2x NVIDIA GH200 NVL2 Grace Hopper Superchips

MeansA combined CPU and GPU package sharing one coherent memory space.

MattersNo copying data back and forth between CPU and GPU memory.

Form factorUnverified

Rack server — exact rack units awaiting verification

Coherent memory capacityUnverified

Awaiting verification

Storage baysUnverified

Awaiting verification

NetworkingUnverified

NVLink plus PCIe and NIC options — exact configuration awaiting verification

CoolingUnverified

Awaiting verification

Verification

Manufacturer documented — HPE ProLiant Compute DL384 Gen12 product page and QuickSpecs a50009209enw. Reviewed 2026-08-31.

Manufacturer source
Before you commit

Practical considerations

The things that catch people out after the hardware has already been ordered.

  • The CPU here is Arm-based, not x86. This is the single most important thing to check before ordering — verify every piece of software you depend on.
  • Coherent memory is the reason to choose this. If your models fit comfortably in ordinary GPU memory, a conventional server is the cheaper answer.
Completeness

Nothing arrives working on its own

What else will I need?

  • Confirmation that your software stack runs on Arm
  • Rack space and power
  • Storage and networking sized to the workload
AI networkingAI storageInstallationAsk Carl what I’m missing
Goes with

What normally sits alongside it

Relationships documented by the manufacturer, or by us during a deployment.