DPU

A data processing unit (DPU) is a programmable processor designed to offload data-centric tasks from a host CPU. DPUs are typically implemented on accelerator cards that sit in servers alongside conventional CPUs, handling the data plane while the host runs applications.

DPU architectures usually include general-purpose cores (often ARM or RISC-V), hardware accelerators for packet processing, encryption, and compression, DMA engines, and high-bandwidth memory. They may run Linux or a real-time OS and expose software development kits for developing data-path functions.

Common workloads are line-rate packet processing, TLS offload, firewall and intrusion prevention, storage offload, and virtualization-related tasks. DPUs can also provide telemetry and policy enforcement, reducing CPU load and improving latency for cloud and enterprise networks.

Deployment often positions DPUs as smartNICs or infrastructure processing units. They are used by cloud providers, telecom operators, and data centers to accelerate networking, security, and storage services. Key considerations include software maturity, ecosystem support, power and form factor, and integration with virtualization technologies such as SR-IOV and VT-d.

Prominent examples include NVIDIA’s BlueField DPU family and Marvell OCTEON-based DPUs, with ongoing developments from other vendors as data-path offloading becomes more widespread.