Based on Broadcom’s scalable 400G Ethernet controller architecture, the N2200G 2x200G OCP 3.0 adapter is designed to build highly scalable, feature-rich networking solutions in servers for AI/ML, cloud, high-performance computing and storage applications.
The N2200G 2x200G OCP 3.0 adapter builds upon the success of the widely deployed Broadcom architecture by combining a high-bandwidth Ethernet controller with a unique set of highly optimized hardware acceleration engines to enhance network performance and improve server efficiency.
The adapter supports fourth-generation, standards-based RDMA over Converged Ethernet (RoCE) with hardware-based congestion control. Broadcom’s RoCE congestion control delivers the lowest latency in real-life scenarios and dramatically reduces RoCE complexity deployment at scale.
With its market-leading hardware acceleration technologies, the N2200G 2x200G OCP 3.0 adapter addresses these performance and service demands of mega-scale data center networks with high throughput and advanced flow processing. Features such as TruFlow™ increase virtual machine density by freeing up CPU cycles. The adapter supports technology-leading security enabling the industry's most secure server platform with secure boot and attestation anchored in Broadcom's silicon Root-of-Trust (RoT).
Features
- Standards-compliant PCIe 5.0 QSFP112 OCP 3.0 adapter with line-rate, full-duplex throughput
- Single-port 400/200/100/50/25G
- Dual-port 200/100/50/25G
- High-speed SerDes capable of 100/50G PAM4 and 25G NRZ
- Large-scale RoCE deployment with Switch Aware Rate Adjustment (SARA) and Drop Congestion Notification (DCN)
- Optimized performance with StrataXGSⓇ and StrataDNX™ switches
- Inline encryption engine enables CPU offload and improves application performance
- Peer memory direct for low-latency inter-GPU communication
- On-chip tunneling protocol processing for Geneve, VXLAN, and NVGRE
- Industry's most secure OCP 3.0 adapter solution provides unparalleled platform security via silicon Root-of-Trust
- TruFlow™ engine line-rate offload for virtual switch processing
- TruManage™ enhances server manageability and security for data center deployments
- Multi-host allows multiple CPUs to connect to a single Ethernet adapter
- NIC partitioning supporting 16 physical functions (PFs)
- Seamless integration with Broadcom's PCIe Gen 5 PCIe switches for GPU-based AI/ML servers
Applications
- Frontend and GPU server
- Artificial Intelligence (AI) and Machine Learning (ML)
- High performance computing (HPC)
- Cloud and enterprise data center servers
- Network Function Virtualization
- NVMe storage disaggregation
- Mobile Edge Computing
- Storage servers