Hardware-accelerated histograms track and summarize queue depth with sub-microsecond granularity

In-band Network Telemetry (INT) and streaming telemetry

512K on-chip traffic counters

Fully shared packet buffers deliver fair, predictable, and high-bandwidth data paths

Low latency

Robust RoCE transport powers NVMe by leveraging the switching fabric of NVIDIA GPUs and machine learning applications

Best-in-class VXLAN scalability with 10x more tunnels and tunnel endpoints than other devices

512K shared forwarding entries, flexibly shared across ACLs, LPM routes, host routes, MAC, ECMP, and tunnel applications

Supports Layer-2, Layer-3, and RoCE

Advanced network virtualization with high-performance single-pass VXLAN routing and IPv6 segment routing

Cloud-scale NAT with 100K+ sessions

Programmable pipeline enabling programmable parsing, processing, and editing of packets

Deep packet inspection with 512B depth

Fully shared packet buffers deliver fair, predictable, and high-bandwidth data paths

Consistent and low cut-through latency

Robust RoCE transport powers NVMe over fabric and machine learning applications leveraging NVIDIA GPUDirect

Best-in-class VXLAN scalable tunnels and tunnel endpoints, 10x more than other devices

512K shared forwarding entries, flexibly shared across ACLs, LPM routes, host routes, MAC, ECMP, and tunnel applications

Supports RoCE, Layer-2, and Layer-3

Advanced network virtualization with high-performance single-pass VXLAN routing and IPv6 segment routing

Cloud-scale NAT with 100K+ sessions

Programmable packet parsing, processing, and editing

Deep packet inspection with 512B depth

支持带内网络遥测 (INT) 、流式遥测