HPE ProLiant DL380 Gen12 GPU Computing Server

Brand: HPE | Category: Servers

SKU: HP-P77372B21 | Part #: P77372-B21 | MPN: P77372-B21

Contact for Pricing — Request a Quote

HPE's ProLiant DL380 Gen12 GPU Computing Server (part number P77372-B21) is a 2U rack platform engineered for dual-socket deployments of either AMD EPYC 9004/9005 series processors or Intel Xeon Scalable 6th Generation CPUs. The form factor and chassis width are typically the first constraints data-centre planners evaluate, and this 2U design delivers substantial compute density without requiring a 4U footprint—a critical consideration for facilities managing rack-unit budgets across hybrid workloads. The dual-socket architecture accommodates processors with thermal design points up to 400W each, making the DL380 Gen12 capable of housing the highest-frequency and highest-core-count variants of both processor families currently available. HPE has engineered this chassis specifically to handle the thermal and power demands of GPU-accelerated inference and training workloads, which means the mechanical design, cooling topology, and electrical infrastructure all assume GPU cards will be populated and thermally active during normal operation.

The GPU compute capability is where the DL380 Gen12 departs most visibly from general-purpose servers. Up to four double-wide GPU accelerators can be installed, with native support for NVIDIA H100, H200 SXM, H100/H200 PCIe variants, and AMD Instinct MI300X cards. Each GPU riser delivers PCIe Gen5 x16 connectivity with 128 gigabytes per second of bidirectional bandwidth per slot—a specification that matters directly to infrastructure teams because it eliminates the PCIe bottleneck that plagued earlier-generation GPU servers during high-frequency model inference. The HPE ProLiant DL380 Gen12 includes dedicated thermal zones for GPU exhaust and predictive thermal control managed through iLO 6, so cooling performance scales with actual card utilization rather than assuming worst-case simultaneous load. For enterprises and AI infrastructure teams evaluating multiple GPU server platforms, this orchestrated thermal approach often becomes the tiebreaker between a platform that runs stable for years and one that drifts toward thermal throttling over time.

Memory architecture reflects the processor socket choice: systems configured with AMD EPYC 9004/9005 chips will support DDR5-5600 across 32 slots, allowing up to 6 terabytes of ECC registered memory in a single chassis. Intel Xeon Scalable 6th Generation configurations also ship with 32 DDR5 RDIMM or LRDIMM slots at identical speed and capacity. This memory headroom is deliberately over-provisioned for GPU-accelerated workloads because inference and training pipelines routinely consume multi-terabyte datasets that must reside in system RAM to avoid repeated PCIe transfers to GPU memory. Storage flexibility complements the memory tier: the DL380 Gen12 offers either up to 12 hot-plug 3.5-inch large-form-factor drives or 24 small-form-factor 2.5-inch bays, plus front-facing NVMe U.3 and U.2 drive support via a dedicated PCIe Gen5 NVMe backplane. This multi-tier storage architecture allows architects to implement all-flash NVMe tiers for model artifact caching while reserving SAS capacity for longer-term training dataset retention.

Network connectivity includes four onboard 25-gigabit Ethernet ports (Intel E810 or Broadcom BCM57504 controller) plus a dedicated 1GbE management port routed through HPE iLO 6. The iLO 6 controller itself is exceptional in this platform because it offers native GPU telemetry visibility—a feature that network engineers and data-centre operations teams have come to depend on when troubleshooting GPU utilization anomalies or validating accelerator health across deployed fleets. For deployments requiring higher aggregate throughput, the DL380 Gen12 includes an OCP 3.0 mezzanine slot capable of hosting 100-gigabit or even 400-gigabit network adapters, a specification that matters when multiple servers feed a single model-serving cluster or when training workloads require synchronized gradient updates across many nodes. Power delivery is handled by dual 1600-watt Platinum-rated HPE Flex Slot power supplies configured in N+1 redundancy with hot-plug capability; the 80 PLUS Platinum certification (≥94% efficiency at 50% load) ensures that even in partial-load scenarios typical of inference serving, the platform wastes minimal electrical overhead as heat. Storage administrators and cloud infrastructure teams appreciate that the HPE ProLiant DL380 Gen12 supports HPE Smart Array Gen11 RAID controllers with inline support for RAID 0, 1, 5, 6, 10, 50, and 60 configurations, plus optional NS204i-u NVMe RAID modules for all-flash configurations.

Security and manageability are woven throughout the platform. Silicon Root of Trust firmware validation, Secure Boot, TPM 2.0, and FIPS 140-2 compliance address regulatory and audit requirements common to financial institutions and government agencies running GPU workloads. HPE GreenLake integration allows the DL380 Gen12 to report inventory, health metrics, and utilization data directly to HPE's managed services portal, simplifying multi-site visibility for enterprises deploying these servers across GCC, Asian, and European data centres. The platform supports Red Hat Enterprise Linux, SUSE Linux Enterprise Server, Ubuntu Server, VMware vSphere 8, and Windows Server 2025, giving AI ops teams and infrastructure planners flexibility in hypervisor and operating system choice without worrying about driver compatibility or firmware qualification delays. Part number P77372-B21 began shipping in Q3 2025, placing it among the earliest production GPU servers to implement PCIe Gen5 and the latest processor generations in a unified 2U form factor.

GPU Compute & Accelerator Specifications

• Up to 4x double-wide GPU accelerators (NVIDIA H100/H200 SXM or PCIe, AMD Instinct MI300X) with PCIe Gen5 x16 per slot and 128 GB/s bidirectional bandwidth
• 32x DDR5 RDIMM/LRDIMM memory slots supporting up to 6TB capacity at DDR5-5600 MHz with full ECC protection
• Dual 1600W Platinum-rated power supplies (N+1 redundant, hot-plug) with ≥94% efficiency at 50% load
• Dedicated GPU thermal zones with predictive cooling control via iLO 6 and high-static-pressure fan arrays
• Front-facing NVMe U.3/U.2 drives via PCIe Gen5 backplane plus up to 12x 3.5-inch LFF or 24x 2.5-inch SFF SAS/SATA/NVMe bays
• Four onboard 25GbE Ethernet ports (expandable to 100GbE/400GbE via OCP 3.0 mezzanine) with dedicated 1GbE iLO 6 management port
• HPE Smart Array Gen11 RAID controller supporting RAID 0/1/5/6/10/50/60; optional NS204i-u NVMe RAID for all-flash deployments

Omnixon Global supplies the HPE ProLiant DL380 Gen12 (P77372-B21) to data-centre operators, AI infrastructure teams, and enterprises across the UAE, GCC region, and Asia-Pacific expansion markets. We maintain direct relationships with HPE and competitive allocation agreements that ensure stock availability for both immediate deployments and planned capacity builds. For infrastructure planning, cost modelling, or technical pre-sales validation of the DL380 Gen12 against your workload profile, contact our solutions team to request a formal quote or arrange a configuration discussion with our technical specialists.