MSN3420-CB2RC - 48x25GbE 12x100GbE Spectrum-2 Switch | NVIDIA
部品番号: MSN3420-CB2RC ✓ 生産中| 数量 | 単価 | 小計 |
|---|---|---|
| 1 | $17725 | $17,725.00 |
| 10 | $17725 | $177,250.00 |
| 100 | $17725 | $1,772,500.00 |
| 500 | $17725 | $8,862,500.00 |
| 1,000 | $17725 | $17,725,000.00 |
MSN3420-CB2RC 概要
A data center Ethernet switch is a Layer 2/Layer 3 managed networking device that interconnects servers, storage, and uplinks within top-of-rack (ToR), leaf, and spine topologies. It sits in the network hierarchy above NICs and below core routers, and belongs to the broader family of switching systems built on merchant silicon switch ASICs.
Key features include a high-density 25G server-access port count with 100G uplink readiness, Open Ethernet programmability via Cumulus Linux on an integrated x86 CPU, and dual hot-swappable AC PSUs for redundancy. The Spectrum-2 ASIC supports port speeds spanning 1GbE to 400GbE across the SN3000 family, giving migration flexibility from 10G to 25G access.
Technically, the Spectrum-2 generation uses a monolithic on-chip packet buffer supporting high-bandwidth packet reads and writes, enabling consistent, predictable cut-through performance. The architecture simplifies buffer management and traffic scheduling while supporting intelligent congestion management, including Explicit Congestion Notification (ECN), which matters for RoCE and cloud-scale workloads.
Typical applications include top-of-rack server access in web-scale IT, leaf switches in leaf-spine cloud networks, hyperconverged storage fabrics, and data analytics clusters - environments where 25G server connectivity and low, consistent latency are decisive.
Design consideration: verify airflow direction before ordering - CB2RC uses C2P (front-to-back exhaust per NVIDIA documentation); pairing it with the wrong rack cooling direction causes thermal throttling. Operating range is 0 to 40 degrees C.
This page synthesizes distributor pricing, sibling-SKU alternatives, and deployment design notes not consolidated in the NVIDIA datasheet.
ドロップイン代替品: MSN3420-CB2RC — 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-CB2RC （同一形状・フットプリント） — 相違点: Airflow, Form Factor, Max Port Speed, Network OS, Power Supplies.
クイック比較ツール — 並列比較する代替部品を選択:
MSN3420-CB2FC
✅ ドロップイン✓ 在庫あり
$16200 / 個
データシートを見る →MSN3420-CS2RC
✅ ドロップイン ⚠️ 仕様未検証📋 参考代替品（カタログ未掲載）
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-CB2RC 最大定格＆電気的特性
| Product Type | Managed Layer 2/Layer 3 Open Ethernet switch |
| Switch ASIC | NVIDIA Spectrum-2 (SN3000 series) |
| 25GbE SFP28 Ports | 48 |
| 100GbE QSFP28 Ports | 12 |
| Total Ports | 60 |
| Max Port Speed | 100000 Mbps |
| Network OS | NVIDIA Cumulus Linux (pre-installed) |
| Management CPU | x86 CPU |
| Form Factor | 1U rack-mountable (rail kit included) |
| Power Supplies | 2x AC PSU (redundant) |
| Airflow | C2P (connector-to-panel), front-to-back |
| Chassis Depth | Standard depth |
| 動作温度 | 0 to 40 C |
| Non-Operational Temperature | -40 to 70 C |
| Operating Humidity | 10% (operational, per distributor listing) |
| NVIDIA Factory Part Reference | 920-9N213 |
| NVIDIA SKU Code | 920-9N213-00R7-0C0 |
| Port Speed Flexibility | 1GbE to 400GbE across SN3000 family |
| Congestion Management | ECN / intelligent congestion management (Spectrum-2) |
MSN3420-CB2RC standard ピン配置ガイド
MSN3420-CB2RC (standard パッケージ) のピン配置。 ピン番号、機能、接続図はメーカーのデータシートで定義されています。正確なフットプリントとはんだ付けのガイドラインについては、データシートを参照してください。
詳細なピン配置データはありません: MSN3420-CB2RC.
詳細は データシート をご参照ください。
主な用途
MSN3420-CB2RC の用途: 6 件s: 25GbE Top-of-Rack Server Access, Leaf Switch in Leaf-Spine Fabrics, Hyperconverged Storage Fabrics, Web-Scale and Cloud Data Centers, Data Analytics Clusters, Enterprise Data Center Refresh (10G to 25G Migration).
25GbE Top-of-Rack Server Access
The MSN3420-CB2RC fits 25GbE top-of-rack access because it provides 48 SFP28 ports at 25GbE - enough for a full 48-server rack - while its 12 QSFP28 ports deliver 100GbE uplinks, giving roughly 2.4:1 oversubscription toward the spine with 1.2 Tbps of uplink capacity. The Spectrum-2 ASIC's monolithic packet buffer and cut-through forwarding keep server-to-server latency consistent even at line rate, which matters for east-west traffic in virtualized clusters. Deployed at the rack top with Cumulus Linux on the x86 CPU, the switch integrates with standard automation ( Ansible-style configuration management) for zero-touch provisioning. Performance consideration: verify C2P airflow matches rack cooling direction, and operate within the 0-40 C envelope to sustain full port bandwidth without thermal throttling.
おすすめ
Leaf Switch in Leaf-Spine Fabrics
As a leaf switch, the MSN3420-CB2RC aggregates server racks at 25GbE and fans up to spine switches through 12x 100GbE QSFP28 ports. NVIDIA explicitly targets the SN3000 series at cloud-scale Layer 2/Layer 3 leaf/spine/super-spine topologies. The Spectrum-2's intelligent congestion management with Explicit Congestion Notification (ECN) is a key fit for RoCE storage and AI training traffic, where congestion signaling prevents pager-buffer overruns on the storage or compute side. With Cumulus Linux, standard BGP-based fabric designs deploy without vendor lock-in, and the x86 CPU supports containerized network agents at the leaf. The trade-off versus higher-radix platforms is uplink count: 12x 100G limits spine fan-out, so large fabrics pair it with dedicated spine SKUs from the same SN3000 family for software consistency.
おすすめ
Hyperconverged Storage Fabrics
Hyperconverged infrastructure (HCI) clusters push heavy east-west replication and I/O traffic that saturates 10GbE. The MSN3420-CB2RC addresses this with 48 ports of 25GbE to HCI nodes, tripling per-node bandwidth while reusing existing 25G SFP28 cabling, and the Spectrum-2's ECN congestion management protects latency-sensitive storage replication from bursty backup traffic. NVIDIA markets the SN3000 series specifically for hyperconverged storage alongside cloud and analytics workloads. The monolithic on-chip packet buffer enables fair resource sharing across flows, preventing a single replication stream from starving VM live-migration traffic. Dual redundant AC PSUs meet the availability targets HCI platforms assume from the access layer. Cumulus Linux enables telemetry and monitoring integration so storage teams can correlate network congestion with application latency.
おすすめ
Web-Scale and Cloud Data Centers
Web-scale operators standardize on Open Ethernet to avoid vendor lock-in, and the MSN3420-CB2RC delivers exactly that model: Spectrum-2 merchant silicon plus Cumulus Linux on an open x86 management CPU. NVIDIA's SN3000 materials cite web-scale IT and cloud as primary use cases, with port speeds from 1GbE to 400GbE across the family allowing a single operating platform from access to super-spine. The 48x25G + 12x100G configuration is the canonical cloud access ratio, giving full 48-server rack connectivity with economical 100G uplinks. Consistent low latency from the cut-through architecture supports tenant-facing SLAs. Operational consideration: automation-first teams should validate their Cumulus Linux version and NOS upgrade pipeline against the factory image before at-scale rollout, since image lifecycle management dominates web-scale switch operations cost.
おすすめ
Data Analytics Clusters
Distributed analytics frameworks (Spark-class compute) shuffle large intermediate datasets between nodes, making the network the dominant job-completion factor. The MSN3420-CB2RC provides each analytics node a dedicated 25GbE port - 2.5x the typical 10GbE baseline - reducing shuffle phase duration proportionally for network-bound jobs. The Spectrum-2 ASIC's predictable cut-through latency and fair packet-buffer sharing prevent straggler nodes caused by queuing variance, which directly improves tail-latency-sensitive query performance. NVIDIA's SN3000 datasheet lists data analytics among the series' target workloads. With 12x 100GbE uplinks, aggregation toward compute or object-storage tiers scales without a blocking bottleneck. Teams should size spine uplinks so that concurrent shuffles from all 48 nodes keep aggregate oversubscription below roughly 3:1 to preserve job determinism.
おすすめ
Enterprise Data Center Refresh (10G to 25G Migration)
NVIDIA positions the SN3420 class specifically for data centers evolving from 10G to 25G server access while keeping 100GbE uplink readiness. The MSN3420-CB2RC's SFP28 ports accept SFP28 DAC/AOC/optics and typically run 10GbE modes, so existing 10G servers stay connected while new 25G-capable servers are added rack by rack - no forklift replacement of the access layer. The 12 QSFP28 uplinks then carry aggregated traffic at 100GbE to the core when the upgrade completes. Cumulus Linux preserves the existing automation toolchain across the transition, and dual AC PSUs maintain current availability SLAs during the phased rollout. Planning note: audit airflow direction per rack row before ordering mixed CB2RC/CB2FC units so the migrated rows cool correctly 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-CB2RC — 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-CB2FC | MSN3420-CS2RC | MSN3420-CS2FC |
|---|---|---|---|---|
| Package | 1U rack-mountable, standard depth | 1U rack-mountable, standard depth - same | 1U rack-mountable, deep chassis | 1U rack-mountable, deep chassis |
| Brand | NVIDIA (Mellanox) | NVIDIA (Mellanox) | NVIDIA (Mellanox) | NVIDIA (Mellanox) |
| 25GbE SFP28 Ports | 48 | 48 | 48 | 48 |
| 100GbE QSFP28 Ports | 12 | 12 | 12 | 12 |
| Switch ASIC | NVIDIA Spectrum-2 | NVIDIA Spectrum-2 | NVIDIA Spectrum-2 | NVIDIA Spectrum-2 |
| Airflow Direction | C2P (connector-to-panel) | Opposite airflow code (FC) | C2P (connector-to-panel) | Opposite airflow code (FC) |
| Network OS | Cumulus Linux (pre-installed) | Cumulus Linux (pre-installed) | Cumulus Linux (pre-installed) | Cumulus Linux (pre-installed) |
主な差別化要因
- C2P airflow for front-to-back cooled racks （vs MSN3420-CB2FC)
- Cumulus Linux pre-installed on x86 CPU （vs Supermicro SSE-SN3420 (EOL))
- Standard-depth 1U chassis （vs MSN3420-CS2RC)
設計ノート
Airflow direction is the number-one ordering error on the SN3000 family. NVIDIA's official ordering information page states explicitly: 'Please pay attention to the airflow direction when ordering your system.' The CB2RC suffix indicates C2P (connector-to-panel) airflow. Mixing airflow directions in one rack row recirculates hot exhaust and leads to thermal throttling or premature PSU fan failure. Audit every rack row's cooling orientation before placing the order, and document the SKU code 920-9N213-00R7-0C0 on purchase orders to prevent receiving the wrong variant.
The system ships with 2x AC PSUs for redundancy. Connect each PSU to separate power feeds (A/B feeds from independent PDUs or UPS branches) to make the redundancy meaningful; dual PSUs on a single PDU negate fault tolerance. Verify rack power budgeting: with 48x 25G and 12x 100G optics populated, transceiver power is a substantial share of total draw, so confirm exact PSU wattage and per-port budgets in the NVIDIA Spectrum-2 SN3000 Hardware User Manual before final rack power planning.
The CB2RC is a standard-depth chassis with a rail kit included; the deep-chassis CS2 siblings require deeper racks and will not fit standard-depth cabinets. Confirm rack depth and rear-service clearance before choosing between standard and deep variants. Also confirm front/rear rack airflow management (blanking panels, cable managers) supports the C2P exhaust pattern, and leave the specified service clearance per the NVIDIA SN3000 Hardware User Manual for PSU and fan-field replacement.
コンプライアンス情報
Compliance declarations not present in the provided web data; consult the NVIDIA SN3000 Hardware User Manual or request NVIDIA's product compliance documentation.
