NVIDIA BlueField-3 P3175 DPU+GPU Combo 200GbE for HPE ProLiant DL380 Gen11

The NVIDIA BlueField-3 P3175 DPU+GPU Combo (HPE part number P60636-B21) is a uniquely integrated accelerator card purpose-built for HPE ProLiant DL380 Gen11 servers, combining a full NVIDIA BlueField-3 Data Processing Unit with GPU compute capabilities on a single PCIe form-factor adapter. The BlueField-3 silicon is built on TSMC 5nm process technology and integrates 16 Arm Cortex-A78AE cores alongside dedicated hardware engines for network offload, security, and storage acceleration, while the 200GbE connectivity fabric delivers up to 200 Gb/s of line-rate network throughput with RDMA over Converged Ethernet (RoCE v2) support. This converged architecture eliminates the latency penalty of crossing PCIe boundaries between discrete DPU and GPU components, making it particularly effective in environments where network-intensive AI inference and data-plane acceleration must coexist on the same platform.

The card supports NVIDIA's DOCA (Data-center Infrastructure-on-a-Chip Architecture) software framework, enabling operators to program the DPU as an isolated, trusted compute domain independent of the host CPU and operating system. Hardware-accelerated functions include IPsec and TLS cryptographic offload, Open vSwitch (OVS) acceleration, NVMe-oF storage target offload, and network function virtualization (NFV) packet processing. The BlueField-3 P3175 variant is specifically validated and factory-integrated for HPE ProLiant Gen11 platform compatibility, carrying HPE's system qualification and firmware alignment to ensure stable operation within iLO 6-managed environments and HPE's Smart Array and OneView ecosystem.

Targeted at enterprise data centers, cloud service providers, and AI infrastructure deployments across UAE, GCC, EMEA, and APAC regions, the P60636-B21 addresses the growing need for infrastructure disaggregation, zero-trust security enforcement at the network edge, and high-throughput AI data pipelines. Its ability to run an independent Arm-based software stack on the DPU while simultaneously handling GPU-accelerated compute workloads positions it as a foundational building block for software-defined data center architectures where CPU offload, network security, and AI acceleration are required concurrently.

Ideal for

• AI inference pipeline acceleration where high-bandwidth 200GbE network connectivity and GPU compute must operate on a shared, low-latency fabric within a single ProLiant DL380 Gen11 node
• Zero-trust security enforcement leveraging hardware-accelerated IPsec and TLS offload on the BlueField-3 DPU, isolating cryptographic operations from the host CPU and hypervisor
• NVMe-over-Fabrics (NVMe-oF) storage disaggregation, using the DPU's storage offload engines to present remote NVMe targets to the host at near-local latency without consuming host CPU cycles
• Network function virtualization and telco cloud deployments requiring line-rate packet processing, Open vSwitch acceleration, and deterministic forwarding performance at 200 Gb/s
• Secure multi-tenant cloud infrastructure where the BlueField-3's isolated Arm subsystem runs tenant-independent control-plane services, preventing host-side compromise from affecting network policy enforcement
• High-performance computing (HPC) and data analytics workloads that demand concurrent GPU compute and ultra-low-latency RDMA networking within HPE's validated ProLiant Gen11 server ecosystem

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.