Juniper QFX5250 Veri Merkezi Switch

Discover the HPE Juniper Networking QFX5250 Switch Series, a liquid-cooled, ultra-high-performance switch delivering 102.4 Tbps throughput, AI-optimized Ethernet, and scalable 1.6TbE connectivity for next-generation AI and HPC data centers.

AI and HPC data centers continue to drive requirements for more bandwidth, along with more power efficiency within the data center fabric. The QFX5250 line of 1.6TbE switches, delivering 102.4 Tbps of throughput, is a next-generation fixed-configuration platform built on Broadcom’s Tomahawk 6 silicon and is designed to support NVIDIA® Vera Rubin and AMD MI450 platforms. The QFX5250 is optimized for scalability, density, and sustainability, delivering 64x 1.6TbE ports, which can also be used to deliver 128x 800GbE ports, and increased efficiency with direct-to-chip liquid cooling.

In addition to high-throughput hardware, the QFX5250 runs on the HPE Networking Junos Operating System and delivers AI-optimized Ethernet to drive performance for AI data centers. It implements congestion avoidance technologies through dynamic load balancing (DLB), global load balancing (GLB), static load balancing (SLB), and weighted packet spraying (WPS). When congestion can’t be avoided, congestion management features control congestion to prevent packet drops, creating a lossless environment. These features include DCQCN (PFC+ECN), interface and queue level telemetry, PFC watchdog, and per-Q Alpha Threshold value support.

The QFX5250 sets a new benchmark for AI fabric switching—combining hardware-level performance with the operational intelligence modern data centers demand. It is engineered for the full operational lifecycle of largescale AI/ML data center environments—from day-one fabric deployment through continuous, at-scale operation. While the hardware delivers the raw switching performance AI workloads demand, it is the intelligent software stack underneath that transforms the QFX5250 into a future-ready, feature-rich network platform built for the evolving demands of modern AI infrastructure. HPE Networking Data Center Director orchestrates complex fabric provisioning and day-two operations into a streamlined, automated workflow—eliminating operational bottlenecks and freeing teams to focus on outcomes rather than infrastructure management. Layered on top, HPE Marvis continuously analyzes traffic behavior across the fabric, proactively surfacing root causes of network degradation and triggering corrective actions before they impact GPU utilization or slow down training runs—dramatically reducing MTTR at machine speed. Together, they turn operational complexity into a competitive advantage for teams running the most demanding AI and HPC workloads at scale.

• ROCEv2 for AI/ML workloads
• DCQCN-PFC, ECN for congestion management
• Support for PFC watchdog for storm avoidance
• DLB for better load balancing
• Configurable hash-bucket size to suit different flow scale
• Selective-DLB—enable/disable DLB based on opcode/byte match
• Leaf/spine/super spine in EVPN-VXLAN fabric
• 139K MAC scale
• 2,000,000 routes (RIB), 880,000 LPM routes (FIB) and 880,000 host routes (FIB)
• 2,000,000 routes (RIB), 875,000 LPM routes (FIB) and 875,000 host routes (FIB)

Nerelerde kullanılır?

Kurumsal ağlarda VLAN ile kamera, misafir ve ofis trafiğini ayırmak, uplink ve omurga bağlantılarını yönetmek için kullanılır.

Tüm özellikler

• ROCEv2 for AI/ML workloads
• DCQCN-PFC, ECN for congestion management
• Support for PFC watchdog for storm avoidance
• DLB for better load balancing
• Configurable hash-bucket size to suit different flow scale
• Selective-DLB—enable/disable DLB based on opcode/byte match
• Leaf/spine/super spine in EVPN-VXLAN fabric
• 139K MAC scale
• 2,000,000 routes (RIB), 880,000 LPM routes (FIB) and 880,000 host routes (FIB)
• 2,000,000 routes (RIB), 875,000 LPM routes (FIB) and 875,000 host routes (FIB)