MSN3420-CB2FC - 48x25GbE + 12x100GbE ToR Switch | NVIDIA
部品番号: MSN3420-CB2FC ✓ 生産中| 数量 | 単価 | 小計 |
|---|---|---|
| 1 | $17762 | $17,762.00 |
| 5 | $17200 | $86,000.00 |
| 10 | $16800 | $168,000.00 |
| 25 | $16500 | $412,500.00 |
| 50 | $16200 | $810,000.00 |
MSN3420-CB2FC 概要
An Open Ethernet top-of-rack switch is a network switching platform that interconnects servers within a data center rack and uplinks them to the leaf or spine layer of a Layer-2/Layer-3 fabric. Within the data center network hierarchy, the ToR switch sits at the bottom of the leaf-spine topology, above server NICs and below aggregation switches, and is a critical element of any cloud, storage, or hyperconverged infrastructure.
Key features include the second-generation Spectrum-2 ASIC with 6.4 Tb/s of bidirectional switching capacity, a full rack of 25GbE server connectivity, 100GbE uplink readiness for leaf-spine or super-spine deployment, and x86 CPU-based management. The -CB2FC configuration specifies back-to-front (C2P/P2C per configuration) airflow and dual AC power supplies for redundant operation, with a rail kit included for standard 19-inch rack mounting.
Architecturally, the Spectrum-2 ASIC delivers consistent low latency and supports advanced software-defined networking features across Layer-2 and Layer-3 forwarding. Running NVIDIA Cumulus Linux, the platform exposes industry-standard Linux networking tools and automation interfaces, making it a natural fit for web-scale IT operations that manage fabrics through configuration-as-code pipelines rather than vendor-proprietary CLIs.
Typical applications include leaf-spine data center fabrics migrating server access from 10GbE to 25GbE, cloud-scale Layer-2/Layer-3 networks, hyperconverged storage clusters, and data analytics back-end fabrics, where high port density and consistent low latency are decisive.
When deploying, verify airflow direction (front-to-back versus back-to-front) matches your hot/cold aisle layout, and confirm 100GbE uplink breakout requirements (4x25G breakout cables) before ordering optics.
This page synthesizes distributor pricing, drop-in alternatives, and practical deployment notes not found in the manufacturer datasheet.
ドロップイン代替品: MSN3420-CB2FC — 同一パッケージでピン互換。PCBの作り直しが必要な異なるパッケージの部品は除外されています。
このシリーズのバリエーション
同一シリーズでドロップイン互換のモデル: MSN3420-CB2FC （同一形状・フットプリント） — 相違点: Airflow, Power Supplies, Product Type, Form Factor, Switching Capacity.
クイック比較ツール — 並列比較する代替部品を選択:
MSN3420-CB2RC
✅ ドロップイン✓ 在庫あり
$17725 / 個
データシートを見る →MSN3410-CB2FC
✅ ドロップイン ⚠️ 仕様未検証📋 参考代替品（カタログ未掲載）
SSE-SN3420-CB2RC
✅ ドロップイン📋 参考代替品（カタログ未掲載）
MSN3700-CS2FC
✅ ドロップイン ⚠️ 仕様未検証✓ 在庫あり
$1 / 個
データシートを見る →MSN3420-CS2FC
✅ ドロップイン ⚠️ 仕様未検証📋 参考代替品（カタログ未掲載）
MSN3420-CB2FC 仕様（メーカー公開）
| Product Type | Managed Open Ethernet switch system (1U rack-mount) |
| ASIC | NVIDIA Spectrum-2 (2nd generation) |
| SFP28 Ports (10/25GbE) | 48 |
| QSFP28 Ports (40/100GbE) | 12 |
| Total Ports | 60 |
| Switching Capacity | 6.4 Tb/s bidirectional |
| Forwarding Layer | Layer 2 / Layer 3 |
| Network OS | NVIDIA Cumulus Linux |
| Management | x86 CPU based, managed |
| Form Factor | 1U, rack-mountable, standard (short) depth |
| Airflow | Back-to-front (as listed by Insight/PartsSource for -CB2FC) |
| Power Supplies | 2 x AC (redundant) |
| Rail Kit | Included |
| Max Port Speed | 100 Gb/s per QSFP28 port |
| Target Applications | ToR leaf-spine, super-spine, cloud, storage, analytics |
MSN3420-CB2FC インターフェースとコネクタ
MSN3420-CB2FC について、メーカーが公開しているインターフェース一覧はありません。コネクタとヘッダーの詳細については、メーカーのドキュメントを参照してください。
詳細は メーカー資料 をご参照ください（拡張ヘッダーのピン配置とコネクタのピン番号の詳細）。
主な用途
MSN3420-CB2FC の用途: 6 件s: Top-of-Rack 25GbE Server Access (Leaf Tier), Leaf-Spine Data Center Fabric, Cloud-Scale Layer-2/Layer-3 Networks, Hyperconverged Storage (HCI) Fabrics, Big Data and Analytics Back-End Networks, Enterprise ToR Migration from 10G with 100G Readiness.
Top-of-Rack 25GbE Server Access (Leaf Tier)
The MSN3420-CB2FC fits the ToR leaf role in data centers upgrading server access from 10GbE to 25GbE. Its 48 SFP28 ports run both 10GbE and 25GbE, so legacy 10G NICs and new 25G servers coexist on one switch, enabling rack-by-rack migration without forklift replacement. The Spectrum-2 ASIC's 6.4 Tb/s bidirectional capacity comfortably line-rates all 48 x 25G ports (1.2 Tb/s) with substantial headroom for the 12 x 100G uplinks (1.2 Tb/s more). Deployed with Cumulus Linux, the leaf node participates in standard BGP underlay fabrics; each server port can be trunked or bonded (MLAG/bond LACP) for dual-ToR redundancy. The back-to-front airflow of the -CB2FC variant must match the rack's front-to-back cooling pattern.
おすすめ
Leaf-Spine Data Center Fabric
In a leaf-spine topology, the MSN3420-CB2FC can serve as either leaf or small-scale spine. NVIDIA's SN3000 datasheet explicitly designates the platform for leaf, spine, and super-spine applications. As a spine, its 12 QSFP28 ports (40/100GbE, breakable into 4x25G) provide 12 to 48 fabric-facing links; as a leaf, its 48 x 25G ports attach servers while the QSFP28 ports uplink. The Spectrum-2 ASIC delivers consistent low latency across Layer-2 and Layer-3 forwarding, which matters for east-west traffic in virtualized and containerized clusters. Cumulus Linux supports BGP unnumbered interfaces, a widely used web-scale pattern that eliminates per-interface IP management. Redundant AC supplies and 1U depth keep the fabric footprint and power budget predictable at rack scale.
おすすめ
Cloud-Scale Layer-2/Layer-3 Networks
The SN3000 series is described by NVIDIA as ideal for building cloud-scale Layer-2 and Layer-3 networks, and the MSN3420-CB2FC embodies this: EVPN/VXLAN overlays in Cumulus Linux extend L2 segments across an L3 routed underlay, enabling multi-tenant isolation at web scale. The Spectrum-2 ASIC supports advanced software-defined networking features, so cloud operators can automate fabric provisioning through Cumulus AutoMation and standard Linux tooling instead of proprietary CLIs. The dual-speed port design lets a single cloud pod SKU serve 10G, 25G, 40G, and 100G tenants simultaneously, simplifying spares pools. Consistent low latency across the 60 ports keeps inter-tenant and storage traffic predictable under full bi-sectional load, which is a stated design goal of the platform.
おすすめ
Hyperconverged Storage (HCI) Fabrics
Hyperconverged storage clusters are bandwidth-hungry and latency-sensitive: each node replicates and re-builds data across the network constantly. The MSN3420-CB2FC addresses this with up to 48 x 25GbE storage-node ports in 1U, and 100GbE uplinks (or 4x25G breakouts) to scale the fabric. According to NVIDIA's positioning, the SN3000 series targets hyperconverged storage among its primary workloads. The Spectrum-2 ASIC's consistent low latency and non-blocking 6.4 Tb/s capacity keep replication and erasure-coding traffic from starving VM I/O. Using Cumulus Linux, storage VLANs or VXLAN segments can be isolated per cluster with MLAG dual-ToR attachment for node-level fault tolerance - essential when a single switch failure would otherwise strand an entire HCI ring.
おすすめ
Big Data and Analytics Back-End Networks
Distributed analytics frameworks (Spark, Hadoop, data lakehouse shuffle stages) generate heavy east-west bursts where a single slow link throttles the whole job. The MSN3420-CB2FC provides full rack connectivity at 25GbE per node with 1.2 Tb/s of uplink bandwidth into the spine layer, backed by the Spectrum-2 ASIC's non-blocking 6.4 Tb/s fabric. NVIDIA lists data analytics as a core SN3000 use case, noting high performance with consistent low latency. The twelve QSFP28 ports can be broken out into 4x25G each to connect up to 96 nodes at 25G when shuffle-heavy clusters need more endpoints than raw uplink speed. Jumbo-frame-friendly L2/L3 forwarding and Cumulus Linux QoS controls let operators prioritize shuffle or replication traffic classes deterministically.
おすすめ
Enterprise ToR Migration from 10G with 100G Readiness
Many enterprise data centers cannot jump straight to 25G or 100G everywhere. The MSN3420-CB2FC is engineered exactly for this transition: its SFP28 ports natively run 10GbE today and 25GbE tomorrow without hardware change, while the twelve QSFP28 ports keep 100GbE uplinks ready for future spine upgrades. NVIDIA describes the platform as serving data centers evolving from 10G to 25G access while maintaining 100GbE uplink readiness. With dual AC supplies, short depth for shallow cabinets, an included rail kit, and back-to-front airflow, it installs into existing enterprise racks without rework. Cumulus Linux's standards-based management means existing NetDevops tooling (Ansible, ZTP) works from day one.
おすすめ
おすすめ製品の概要
エンジニアリング参考データ: MSN3420-CB2FC — 比較、設計ガイダンス、コンプライアンス情報。
選定ガイド
代替品との比較
| パラメータ | この製品 | MSN3420-CB2RC | MSN3410-CB2FC | SSE-SN3420-CB2RC | MSN3700-CS2FC |
|---|---|---|---|---|---|
| Brand | NVIDIA (Mellanox) | NVIDIA | NVIDIA | Supermicro | NVIDIA |
| Package / Form Factor | 1U rack-mount, short depth | 1U rack-mount, short depth - same | 1U rack-mount - same family | 1U rack-mount - same | 1U rack-mount - same family |
| ASIC | Spectrum-2 (2nd gen) | Spectrum-2 - same | Spectrum-2 - same | Spectrum-2 - same | Spectrum-1 (1st gen) |
| Switching Capacity | 6.4 Tb/s bidirectional | 6.4 Tb/s | 6.4 Tb/s | 6.4 Tb/s | 6.4 Tb/s |
| Network OS | Cumulus Linux | Cumulus Linux (Onyx/SONiC options per SKU) | Cumulus Linux | Open Networking OS (Cumulus/SONiC) | Cumulus Linux |
| Airflow | Back-to-front | Opposite airflow variant (R vs F suffix) | Front-to-back (per suffix) | Reverse airflow | Front-to-back (per suffix) |
| Lifecycle | Active | Active | Active | EOL / Discontinued | Active |
主な差別化要因
- Dual-speed 25G server access with 10G backward compatibility （vs MSN3700-CS2FC)
- Second-generation Spectrum-2 silicon （vs MSN3700-CS2FC)
- Actively supported versus discontinued cross-brand equivalent （vs SSE-SN3420-CB2RC)
設計ノート
Airflow direction is the single most common ordering error with the SN3000 series. NVIDIA's ordering-information documentation explicitly warns to pay attention to airflow direction when ordering systems. The -CB2FC suffix indicates the back-to-front (or front-to-back) airflow orientation; ordering the wrong direction forces hot exhaust into cold aisles, throttling the ASIC under load. Physically verify the airflow arrow molded into the chassis and match it to your rack's hot/cold aisle layout before issuing the PO - especially when mixing CB2FC and CB2RC units in the same rack row.
Plan rack power for dual redundant AC feeds: the MSN3420-CB2FC ships with two AC power supplies, and each should be connected to a separate PDU/power feed for true redundancy. A single feed failure should leave the switch fully operational. Exact power draw figures were not published in the cited sources; consult the SN3000 hardware user manual on docs.nvidia.com before finalizing circuit breaker and PDU capacity, and remember to de-rate when both supplies share one branch circuit during lab deployments.
Use the QSFP28 breakout capability early in your design. Breaking each of the twelve 100G QSFP28 ports into 4 x 25G lanes yields up to 96 x 25G endpoints from one 1U chassis, which can defer a second ToR switch purchase in dense racks. Ensure cabling plans account for breakout DAC/AOC cable bend radius in the short-depth chassis, and label both physical and logical breakout ports in your Cumulus Linux configuration to avoid port-map confusion during troubleshooting.
コンプライアンス情報
Compliance data for the complete switch system was not present in the provided web data; consult the NVIDIA SN3000 hardware user manual and product compliance documentation.