Scale secure AI clusters with NVIDIA Cloud Partner networking
Increase throughput, strengthen tenant isolation, and preserve host CPU capacity with F5 BIG-IP Next for Kubernetes on NVIDIA BlueField-3 DPUs—built for the NVIDIA Cloud Partner Common Networking architecture.
Accelerate and secure Kubernetes networking for AI
F5 BIG-IP Next for Kubernetes runs on NVIDIA BlueField-3 DPUs to deliver high-performance ingress and egress, integrated security, and application-aware traffic management for multi-tenant AI clusters. Offloading network and security processing to the DPU preserves host CPU resources for AI workloads.
Key benefits
Accelerate ingress and egress
Use NVIDIA DOCA and BlueField-3 DPU hardware acceleration to deliver high-bandwidth, low-latency networking for AI training and inference workloads.
Offload security from host CPUs
Move firewall ACLs and DDoS mitigation to BlueField-3 DPUs to protect sensitive AI workloads without consuming valuable host CPU capacity.
Isolate tenants at scale
Map Kubernetes namespaces to segmented network paths and apply dedicated security and traffic policies across multi-tenant AI clusters.
Automate with Kubernetes-native APIs
Use CNCF Gateway API, operators, and CRDs to automate deployment while extending Kubernetes with data-center-grade networking and security.
What you'll learn in this reference architecture
Design the NCP fabric
Understand the converged network design, NVIDIA BlueField-3 DPU placement, EVPN-VXLAN segmentation, and BGP connectivity.
Deploy accelerated services
Configure DPUs, Kubernetes node and Exit VLANs, Gateway API services, and accelerated ingress and egress.
Operate securely at scale
Apply tenant isolation, firewall policies, telemetry, logging, troubleshooting, and real-time traffic capture.
FROM ARCHITECTURE TO OPERATIONS
Build high-performance Kubernetes networking for AI
Download the reference architectureInside the reference architecture
Architecture
Design a secure, accelerated NVIDIA NCP fabric
See how F5 BIG-IP Next for Kubernetes runs on NVIDIA BlueField-3 DPUs within the NVIDIA NCP Common Networking architecture. The design connects Kubernetes application and namespace context to accelerated ingress and egress, EVPN-VXLAN segmentation, BGP routing, security policies, and observability across multi-tenant AI clusters.
Deployment and operations
Deploy and operate DPU-accelerated services
Follow the deployment model for BlueField-3 DPUs, converged networking, VLANs, and BIG-IP data plane services. The reference architecture also covers accelerated ingress and egress, Kubernetes-native configuration, firewall policies, programmability, telemetry, logging, troubleshooting, and example deployment configurations.
Converged network design
Connect compute, high-performance storage, management, and border services through a consistent NVIDIA NCP Common Networking fabric.
BlueField-3 DPU deployment
Configure DPU mode, hardware-accelerated eSwitch link aggregation, DOCA-accelerated OVS, and scalable functions for BIG-IP data plane services.
Accelerated ingress and egress
Scale Kubernetes ingress and namespace-aware egress while offloading traffic processing from host CPUs to NVIDIA BlueField-3 DPUs.
Multi-tenant security
Apply tenant-specific segmentation, firewall policies, DDoS mitigation, network address translation, and programmable traffic controls.
Observability and supportability
Use OpenTelemetry, cloud-native logging, RBAC-controlled diagnostics, support log collection, and real-time traffic captures.
Frequently asked questions
It explains how to deploy F5 BIG-IP Next for Kubernetes within the NVIDIA Cloud Partner Common Networking architecture. It covers BlueField-3 DPU integration, converged networking, accelerated Kubernetes ingress and egress, tenant isolation, security, automation, and observability.
It is designed for network architects, platform engineers, Kubernetes administrators, security teams, and infrastructure operators building multi-tenant AI and cloud-native environments with NVIDIA NCP infrastructure.
F5 BIG-IP Next for Kubernetes uses NVIDIA DOCA and BlueField-3 hardware acceleration to process ingress, egress, load balancing, and network address translation on the DPU. This reduces reliance on host CPU resources while supporting high-throughput, low-latency application traffic.
The architecture maps Kubernetes namespaces and application context to segmented network paths. Teams can apply tenant-specific firewall policies, DDoS protection, address translation, business logic, telemetry, and logging while isolating workloads within shared AI infrastructure.
The architecture includes F5 BIG-IP Next for Kubernetes, NVIDIA BlueField-3 DPUs, NVIDIA DOCA, EVPN-VXLAN, BGP, CNCF Gateway API, Kubernetes operators and CRDs, accelerated OVS networking, OpenTelemetry, and cloud-native logging.
Yes. It includes topology diagrams and configuration examples for BlueField-3 DPU mode, BFB installation, eSwitch link aggregation, DOCA-accelerated OVS, scalable functions, Kubernetes node and Exit VLANs, ingress and egress services, and BGP peering.