Data processing unit that offloads networking, storage and security from host CPUs using NVIDIA DOCA.
NVIDIA BlueField data processing units combine Arm compute with high-speed networking to offload networking, storage and security from host CPUs. They run software-defined infrastructure services through NVIDIA DOCA.
BlueField-3 delivers 400 Gb/s and is widely deployed today; BlueField-4 doubles that to 800 Gb/s for gigascale AI factories, and a BlueField-4 storage processor powers AI-native storage.
Software-defined networking and distributed routing without host CPU overhead.
Real-time threat detection, isolation and runtime protection in silicon.
RDMA, NVMe-oF and GPUDirect Storage for AI data and inference context.
Securely share infrastructure across tenants while improving GPU and CPU utilisation.
Elastic provisioning across cloud environments.
| Peak throughput | 800 Gb/s |
|---|---|
| Purpose | Infrastructure processor for gigascale AI factories |
| Peak throughput | 400 Gb/s |
|---|---|
| Purpose | Software-defined networking, storage and cybersecurity |
| CPU | NVIDIA Vera |
|---|---|
| Networking | ConnectX-9 |
| Role | Storage-side processor and data-path engine |
| Framework | NVIDIA DOCA |
|---|
End-to-end 800 Gb/s InfiniBand platform for the largest AI training clusters.
Ethernet platform purpose-built for AI, from single-cluster fabrics to multi-site AI factories with Spectrum-XGS.
SuperNIC delivering up to 800 Gb/s for InfiniBand or Ethernet GPU-to-GPU networking.
Tell us your workload and timeline for the NVIDIA BlueField DPU. We reply within one business day with a configuration, pricing and lead time.