Dell PowerSwitch Z-series Data Center Switch

The Dell EMC PowerSwitch Z-series represents a next-generation data center switching platform engineered specifically for the demands of AI/ML training clusters, high-performance computing fabrics, and modern cloud-scale data center architectures. Built around a high-radix, non-blocking switching ASIC, the Z-series delivers wire-rate forwarding across a dense portfolio of 400GbE and 800GbE ports, enabling the low-latency, high-bandwidth spine and leaf interconnects that large-scale GPU and accelerator clusters require. The platform supports both Ethernet-based AI fabrics and traditional IP/MPLS data center underlay topologies, offering flexible deployment across multi-vendor environments via open standards and rich protocol support.

Architecturally, the Z-series is designed around a deep packet buffer configuration to handle the bursty, all-to-all traffic patterns characteristic of distributed AI training workloads such as AllReduce collectives across thousands of GPUs. The switch supports advanced congestion management features including Dynamic Load Balancing (DLB), Explicit Congestion Notification (ECN), and Priority Flow Control (PFC), all critical for maintaining fabric efficiency and preventing head-of-line blocking in lossless RoCEv2 RDMA environments. The platform integrates Dell's SmartFabric OS and supports SONiC-based open NOS deployments, giving operators the choice between automated lifecycle management and a fully programmable, disaggregated network operating system.

The Z-series positions itself as a cornerstone fabric switch for enterprises and hyperscalers building GPU SuperPOD architectures, large-scale inference clusters, and next-generation storage-compute fabrics. Its high port density reduces the number of switching tiers required in a Clos fabric, directly lowering total rack count, cabling complexity, and power consumption per unit of bandwidth. Integration with Dell OpenManage and Redfish-compliant APIs enables seamless incorporation into automated data center operations pipelines, while support for telemetry streaming via gRPC/gNMI provides real-time visibility into fabric utilization critical for AI workload scheduling and optimization.

Ideal for

• Spine-layer switching in GPU SuperPOD and AI training cluster fabrics requiring non-blocking 400/800GbE connectivity between thousands of accelerator nodes
• Lossless RoCEv2 RDMA fabric interconnect for distributed deep learning workloads performing large-scale AllReduce and gradient synchronization operations
• High-bandwidth east-west traffic switching in hyperscale and enterprise cloud data centers running containerized microservices and distributed storage platforms
• NVMe-oF storage fabric aggregation connecting high-density all-flash arrays to compute nodes with sub-microsecond latency and sustained multi-terabit throughput
• Leaf-spine Clos fabric construction for large-scale inference serving infrastructure requiring deterministic low-latency forwarding for real-time AI model inferencing
• Open SONiC-based programmable network underlay for telco cloud and network function virtualization deployments requiring vendor-agnostic, software-defined switching at scale

Technical specifications

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