HPE ProLiant Compute DL380a Gen12

Configurable systemConfigure to orderManufacturer documented

A flexible enterprise GPU server. It takes up to ten double-width accelerators of up to 600W each, which makes it one of the more practical routes into serious AI inference without committing to an eight-way HGX platform.

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?

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

Enterprises putting an AI service into production
Teams wanting GPU choice rather than a fixed baseboard
Organisations mixing inference and conventional workloads

What it runs

Production AI inference
Mixed training and inference
Virtualised GPU workloads

When it is the wrong answer

Office deployment
Very large distributed training runs — that is an HGX conversation

Where it goes: Datacentre or a rack with 4U free and adequate cooling
Complexity: high
Cooling: Air cooled

Specification

What the numbers mean

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

System

Form factor

4U, two-socket, rack-optimised

Processors

2x Intel Xeon 6, up to 144 cores, up to 250W TDP

GPU capacity

Up to 10 double-wide GPUs, up to 600W each

MattersUnusually flexible — you are not locked to one accelerator generation.

GPU interconnect

NVIDIA 4-way NVLink Bridge across four neighbouring GPUs

MeansFour cards can pool memory and behave more like one larger accelerator.

WhenOnly when a model will not fit in a single card's memory.

Power architecture

Three power domains; two domains with six PSUs dedicated to the GPUs

MattersGPU power is separated from system power, which is why the density is achievable.

MemoryUnverified

DDR5 — maximum capacity awaiting verification

Storage baysUnverified

SFF NVMe — exact count awaiting verification

Networking

OCP 3.0 and PCIe Gen5 slots

CoolingUnverified

Air-cooled; liquid-cooling option awaiting verification

Verification

Manufacturer documented — HPE ProLiant Compute DL380a Gen12 product page and datasheet PSN1014846947. Reviewed 2026-08-31.

Manufacturer source

Before you commit

Practical considerations

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

Ten GPU slots does not mean ten of any GPU. The power domains and thermal envelope constrain which combinations are supported — we will confirm against your chosen accelerator.
Four-way NVLink bridging is available across neighbouring GPUs, which matters for models that will not fit on one card.

Completeness