Juniper Networks QFX5120-48YM Enterprise Leaf Switch

The Juniper Networks QFX5120-48YM is a high-density, multi-rate leaf switch purpose-built for modern AI/ML infrastructure and enterprise data center environments demanding lossless fabric performance. Featuring 48 ports of 25G (with 10G/25G multi-rate flexibility) and 8 uplink ports supporting 100G QSFP28, the platform delivers a full 2.4 Tbps switching capacity in a compact 1RU form factor. The forwarding architecture is anchored by a high-performance merchant silicon ASIC paired with a deep, shared packet buffer pool specifically sized to absorb micro-burst traffic patterns endemic to GPU-to-GPU collective communication operations such as AllReduce and AllGather workloads in distributed training clusters.

What distinguishes the QFX5120-48YM within the QFX5120 family is the mandatory hardware MACsec (IEEE 802.1AE) support across all 48 front-panel ports simultaneously, enabling line-rate AES-256 encryption without any throughput penalty. This addresses stringent zero-trust network segmentation requirements in multi-tenant GPU cluster environments where east-west traffic confidentiality is non-negotiable. The enhanced buffer design ensures Priority Flow Control (PFC) and Enhanced Transmission Selection (ETS) operate effectively to deliver RDMA over Converged Ethernet (RoCEv2) lossless transport, a prerequisite for high-performance collective communication libraries including NCCL and OpenMPI.

Running Juniper's Junos OS with native support for EVPN-VXLAN, BGP-EVPN underlay/overlay architectures, and telemetry streaming via gRPC/gNMI, the QFX5120-48YM integrates seamlessly into Juniper's Apstra intent-based automation framework for closed-loop network management. The platform supports advanced ECMP hashing with adaptive load balancing and includes Juniper's AI-Native Networking Platform telemetry hooks, enabling AIOps-driven anomaly detection across large GPU fabric deployments. Its combination of MACsec ubiquity, deep buffers, and DCB protocol support makes it a purpose-fit spine-leaf building block for HPC, AI inference farms, and enterprise cloud-native workloads.

Ideal for

• GPU cluster east-west fabric for distributed AI/ML training using RoCEv2 lossless transport with PFC and ECN congestion management
• Zero-trust data center leaf layer with line-rate MACsec AES-256 encryption on all 25G server-facing ports for multi-tenant workload isolation
• EVPN-VXLAN leaf node in spine-leaf architectures supporting large-scale overlay networks for containerized microservices and Kubernetes CNI plugins
• High-frequency trading and financial analytics environments requiring deterministic ultra-low latency switching with nanosecond-resolution timestamping
• Enterprise private cloud NVMe-oF and storage disaggregation fabrics benefiting from enhanced buffer depth and lossless block storage transport
• Telco edge compute and 5G MEC deployments requiring compact 1RU density, encrypted east-west traffic, and streaming telemetry integration with automation platforms

Technical specifications

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