NVIDIA Spectrum-4 SN5400 Ethernet Switch

NVIDIA Spectrum-4 SN5400 Ethernet Switch

Brand: NVIDIA | Category: Network Switches

SKU: NVID-MSN5400VS2FO | Part #: MSN5400-VS2FO | MPN: MSN5400-VS2FO

Contact for Pricing — Request a Quote

Request a Quote Contact Us

About the NVIDIA Spectrum-4 SN5400 Ethernet Switch

The NVIDIA Spectrum-4 SN5400 is a 64-port 400GbE Ethernet switch based on the Spectrum-4 ASIC, offering 51.2Tb/s of aggregate non-blocking switching bandwidth in a 1U form factor. It is purpose-built for leaf switching roles in AI data center fabrics, providing direct 400GbE connectivity to GPU servers or DGX systems while upstreaming to SN5600 spine switches. The combination of high port density and a compact 1U chassis makes the SN5400 ideal for racks requiring maximum GPU server concentration with high-bandwidth ToR switching.

The SN5400 shares the same Spectrum-4 silicon and feature set as the SN5600, including full RoCEv2 lossless RDMA support with hardware-enforced PFC and ECN, EVPN/VXLAN for multi-tenant segmentation, and wire-rate telemetry with nanosecond granularity. Breakout capabilities allow each 400GbE port to be split into multiple 100GbE or 200GbE connections, enabling flexible server and storage connectivity in mixed-speed environments. NVIDIA Cumulus Linux running on the SN5400 provides a standard Linux-based NOS with full programmatic control via APIs, NetQ telemetry, and integration with popular automation frameworks such as Ansible, Terraform, and Salt.

The SN5400 is widely selected by enterprise AI infrastructure teams deploying NVIDIA-recommended reference architectures such as DGX BasePOD and DGX SuperPOD, where it serves as the compute fabric leaf switch. Its ability to deliver consistent, low-latency RoCEv2 performance while supporting the full enterprise networking feature set makes it a versatile choice for organizations modernizing both their AI training and general data center networking infrastructure.

Ideal for

  • Leaf switch in AI data center leaf-spine architecture for DGX H100/H200 pod connectivity
  • Top-of-rack switch providing 400GbE server-side ports for GPU-accelerated servers
  • RoCEv2 lossless RDMA fabric leaf node for distributed AI training over Ethernet
  • Aggregation switch in enterprise data centers migrating to 400GbE server connectivity
  • Storage switch for high-performance NVMe-oF and parallel file system traffic
  • Multi-tenant cloud edge fabric with EVPN/VXLAN support

Technical specifications

ManufacturerNVIDIA
Part NumberMSN5400-VS2FO
Switch ASICNVIDIA Spectrum-4
Ethernet Speed400GbE per port (native)
Total Ports64 x QSFP-DD 400GbE ports
Breakout Support64 x 400GbE, 128 x 200GbE, 256 x 100GbE
Aggregate Bandwidth51.2 Tb/s non-blocking
Forwarding Rate38.4 Bpps
Latency<900 ns port-to-port (cut-through mode)
Buffer ArchitectureShared and dedicated buffers with Spectrum-4 advanced buffer management
RDMA SupportRoCEv2 with PFC, ECN, hardware-based congestion control
Layer 2 FeaturesVLAN, MSTP, LACP/LAG, MLAG, VxLAN, EVPN
Layer 3 FeaturesBGP, OSPF, IS-IS, ECMP, BFD, PIM, IGMP
Management Interface1 x 10/100/1000 RJ-45 OOB; USB console; 1 x 10GbE management SFP+
Operating SystemNVIDIA Cumulus Linux / NVIDIA Onyx NOS
TelemetryStreaming telemetry (gRPC/gNMI), INT, nanosecond timestamps, NVIDIA NetQ
Form Factor1U rack-mount
Dimensions44 x 440 x 560 mm (H x W x D)
Power SupplyDual redundant hot-swap PSUs, 1200W typical
CoolingHot-swap redundant fans, port-side exhaust or intake
Operating Temperature0°C to 45°C
StandardsIEEE 802.3bs (400GbE), 802.1Q, 802.3ad, 802.1X

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.

Technical Specifications

BrandNVIDIA
CategoryNetwork Switches
SKUNVID-MSN5400VS2FO
Part NumberMSN5400-VS2FO
ConditionNew
Switch ASICNVIDIA Spectrum-4
Ethernet Speed400GbE per port (native)
Total Ports64 x QSFP-DD 400GbE ports
Breakout Support64 x 400GbE, 128 x 200GbE, 256 x 100GbE
Aggregate Bandwidth51.2 Tb/s non-blocking
Forwarding Rate38.4 Bpps
Latency<900 ns port-to-port (cut-through mode)
Buffer ArchitectureShared and dedicated buffers with Spectrum-4 advanced buffer management
RDMA SupportRoCEv2 with PFC, ECN, hardware-based congestion control
Layer 2 FeaturesVLAN, MSTP, LACP/LAG, MLAG, VxLAN, EVPN
Layer 3 FeaturesBGP, OSPF, IS-IS, ECMP, BFD, PIM, IGMP
Management Interface1 x 10/100/1000 RJ-45 OOB; USB console; 1 x 10GbE management SFP+
Operating SystemNVIDIA Cumulus Linux / NVIDIA Onyx NOS
TelemetryStreaming telemetry (gRPC/gNMI), INT, nanosecond timestamps, NVIDIA NetQ
Form Factor1U rack-mount
Dimensions44 x 440 x 560 mm (H x W x D)
Power SupplyDual redundant hot-swap PSUs, 1200W typical
CoolingHot-swap redundant fans, port-side exhaust or intake
Operating Temperature0°C to 45°C
StandardsIEEE 802.3bs (400GbE), 802.1Q, 802.3ad, 802.1X

Frequently Asked Questions about NVIDIA Spectrum-4 SN5400 Ethernet Switch

What does the NVIDIA Spectrum-4 SN5400 Ethernet Switch do?

The NVIDIA Spectrum-4 SN5400 Ethernet Switch is enterprise networking hardware for data-center, campus, and branch deployments. Common roles include core/distribution switching, server uplinks, top-of-rack fabric, and edge connectivity in mixed Cisco/Juniper/Arista environments.

What are the headline specs of the NVIDIA Spectrum-4 SN5400 Ethernet Switch?

Key specifications for the NVIDIA Spectrum-4 SN5400 Ethernet Switch: new condition; manufacturer NVIDIA; part number MSN5400-VS2FO; switch asic NVIDIA Spectrum-4; ethernet speed 400GbE per port (native); total ports 64 x QSFP-DD 400GbE ports; breakout support 64 x 400GbE, 128 x 200GbE, 256 x 100GbE. Manufacturer part number MSN5400-VS2FO. For the full datasheet with electrical, environmental, and compliance details, contact our pre-sales engineering team.

Is the NVIDIA Spectrum-4 SN5400 Ethernet Switch compatible with my infrastructure?

The NVIDIA Spectrum-4 SN5400 Ethernet Switch interoperates with standard switching protocols (LACP, MLAG/VPC, 802.1Q, OSPF/BGP). For specific cross-vendor scenarios — Cisco-to-Arista MLAG peering, fabric integration with Cumulus/SONiC, or migration from legacy Catalyst — our network engineers map a compatibility plan as part of the RFQ.