NVIDIA BlueField DPU
Vendor: NVIDIA
Category: DPU Data Processing Unit
Architecture: Mellanox ConnectX + ARM
Introduction
NVIDIA BlueField DPU (Data Processing Unit) is purpose-built for data center infrastructure acceleration. Offloads network, storage, security, telemetry and other tasks, freeing CPU and GPU to focus on computation. Supports DOCA SDK programming.
Specifications
| Model | Compute | Memory | Interface | TDP | Process |
|---|---|---|---|---|---|
| BlueField-3 | 200 Gbps Networking | 16GB DDR5 + 4GB ARM | PCIe 5.0 / 400GbE | 35W | 5nm |
| BlueField-2 | 100 Gbps Networking | 8GB DDR4 + 4GB ARM | PCIe 4.0 / 200GbE | 25W | 16nm |
Official Website
Driver Downloads
Linux
Related Documentation
OS Support
| Windows | Linux | macOS | Android |
|---|---|---|---|
| ✅ | ✅ | ❌ | ❌ |
Version History
| Version | Release Date | Description |
|---|---|---|
| DOCA 2.5 | 2024 | Full BlueField-3 feature support |
| DOCA 2.0 | 2023 | Zero-trust security framework |
Performance Benchmarks
| Model | Task | Performance Metric |
|---|---|---|
| BlueField-3 | Network Offload | 400Gbps line-rate processing |
| BlueField-3 | Storage Offload | NVMe-oF/RDMA acceleration |
| BlueField-2 | Security Offload | IPsec/TLS hardware acceleration |
Pricing Information
| Model | Reference Price | Notes |
|---|---|---|
| BlueField-3 DPU | Contact sales | Enterprise data center |
| BlueField-2 DPU | Contact sales | Widely deployed |
Quick Setup
Linux
# 1. Install DOCA SDK (NVIDIA DPU Software)
apt install -y doca-sdk
# 2. Configure BlueField
bfread -p /dev/mst/mt41686_pciconf0
# 3. Verify
mst status
DOCA SDK download from NVIDIA DOCA.
Code Examples
C (DOCA Flow - Network Acceleration)
#include <doca_flow.h>
// Create flow rules for network packet processing
doca_flow_port_cfg port_cfg = {0};
port_cfg.port_id = 0;
port_cfg.type = DOCA_FLOW_PORT_DPDK_BY_ID;
// Add flow rule: forward to VF
doca_flow_match match = {0};
doca_flow_actions actions = {0};
actions.fwd.type = DOCA_FLOW_FWD_VPORT;
actions.fwd.port_id = vport_id;
doca_flow_create_rule(port, &match, NULL, &actions, NULL);
Architecture Highlights
- Mellanox ConnectX + ARM: Combination of network controller and ARM cores, capable of running an independent OS
- DOCA Software Stack: NVIDIA's unified DPU programming framework, supporting network, storage, and security offload
- Zero-Trust Security: Hardware-level isolation, running security policies on the DPU without impacting host performance
Model Compatibility
| Model/Framework | Support Status | Notes |
|---|---|---|
| DOCA SDK | ✅ Native | NVIDIA official |
| DPDK | ✅ | Network acceleration |
| SPDK | ✅ | Storage acceleration |
| Kubernetes | ✅ | Container networking |
| AI Frameworks | N/A | DPU is not an AI accelerator |
Related Products
If you are evaluating alternatives, the following products may also fit your scenario:
- AMD Pensando — AMD (DPU Data Processing Unit)
- Intel IPU — Intel (DPU Data Processing Unit)
- NVIDIA GPU / CUDA — NVIDIA (GPU Graphics Processor)
- AMD ROCm / GPU — AMD (GPU Graphics Processor)
- Intel GPU — Intel (GPU Graphics Processor)