Spectrum®-2 Ethernet Switch ASIC
6.4Tb/s Ethernet Switch Purpose-built for the Software Defined Data Center
Spectrum-2, the second generation of the field-proven Spectrum™ switch ASIC, provides industry-leading Ethernet connectivity for up to 16 ports of 400GbE, 32 ports of 200GbE, 64 ports of 100GbE, or 128 ports of 50/25/10/1GbE.
Spectrum-2 features a fully shared and monolithic packet buffer that provides a fair, deterministic and high-performance data path that is critical for storage and multi-tenant workloads. Spectrum-2 supports best in class network virtualization features and overlay scale for software defined datacenter networks. Also, Spectrum-2 dramatically reduces mean time to resolution/innocence with its What Just Happened (WJH) visibility features.
With its rich set of features and innovative capabilities - including increased flexibility and port density - Spectrum-2 is optimized for cloud, hyperscale, big data, artificial intelligence, financial, storage and more applications.
Spectrum-2 switch enables end-to-end automated RoCE (RDMA over Converged Ethernet) support with a comprehensive ESF (Ethernet Storage Fabric) solution.
Spectrum2 ASIC - Build your Cloud Network without Compromise
- Advanced Network Virtualization with support for high performance single pass VXLAN routing and IPv6/MPLS segment routing
- Programmable Pipeline with the ability to programmatically parse, process and edit packets
- Deep Packet Inspection - 512B Deep
- Cloud Scale NAT – 100K+ connections
- MPLS or IPv6 based Segment routing
- Best in class VXLAN scale with support for 10X more tunnels and tunnel end points
- 512K shared forwarding entries that can be flexibly shared across ACL, LPM routes, Host routes, MAC, ECMP and Tunnel applications
- Extensible forwarding tables with support for up to 2M LPMs with off-chip lookups
- Fully Shared Packet buffer provides fair, predictable and high-performance data path that is essential for scale out software defined storage and modern multi-tenant cloud deployments
- Robust RoCE transport to power NVMe over fabric and Machine Learning applications that leverage GPUdirect
- Consistent and low cut-through latency
- Automated QoS treatment for Elephant/Mice flows with Adaptive Flow Prioritization
- Adaptive Routing with flowlet based dynamic load balancing to maximize cross-sectional network bandwidth
- Reduced Mean Time to Recovery/Innocence. Detailed and contextual telemetry with What Just Happened (WJH). Instant answers to When, What, Who, Where and Why questions as soon as an issue happens.
- Hardware accelerated histograms track and summarize queue depths at a sub-microsecond granularity. Avoids false-alerts common to simple watermarks/thresholds
- Inband Network Telemetry
- 512K on-chip flow counters