Cisco Nexus Dashboard Physical Appliance Options: All Four Hardware Models Compared

Physical vs Virtual ND — When to Go Bare Metal

Nexus Dashboard runs on four deployment platforms: VMware ESX, Linux KVM, Nutanix, AWS, and physical appliances. For production data center environments where performance, dedicated resources, and hardware predictability matter, the physical appliance form factor is the recommended choice.

Cisco offers purpose-built physical nodes based on UCS rack servers. These are not generic UCS servers you configure yourself — they’re pre-configured appliances purpose-built for ND, locked to run only Cisco Nexus Dashboard software. If any other operating system is installed on one of these nodes, it can no longer be used as an ND node.

Important: Component replacement on physical ND nodes is handled via Return Material Authorization (RMA), not field service. You don’t swap DIMMs or drives yourself — you RMA the appliance.

As of Release 4.2, Cisco supports four physical hardware models:

The Four Physical Node Options at a Glance

All four models are orderable as either a single-node appliance or a 3-node cluster chassis. The cluster PIDs above include three appliances pre-configured to work together.

ND-NODE-G5L — Current Generation, Large Profile

The G5L is the flagship physical ND node based on the UCS C225 M8 server platform. It’s the highest-capacity option in the current lineup.

CPU

AMD 9655P processor — 96 cores, 2.6GHz base, 400W TDP, 384MB cache, DDR5 6000MT/s memory support.

Storage Configuration

The G5L ships with a mix of storage tiers for different purposes:

• Boot: 240GB M.2 SATA SSD (Intel) via Cisco Boot Optimized M.2 RAID controller
• Data (HDD): 2.4TB 12G SAS 10K RPM SFF HDD (4Kn format)
• Data (SSD): 960GB 2.5” Enterprise Performance 6G SATA Micron G2 SSD (3x)
• High-performance data: 3.2TB 2.5” Enterprise Performance 24G SAS Kioxia PM7 SSD (3x)
• RAID controller: 24G Tri-Mode M1 RAID Controller with 4GB FBWC, 16 drive support

Memory

64GB DDR5-6400 RDIMM, 2Rx4 configuration.

Network Cards

Management network (mgmt0/mgmt1): Uses the ND-P-ID10GC-D — a Cisco-Intel X710T2LG 2x10GbE RJ45 PCIe NIC installed in PCIe Riser 01. Copper connections, 1G or 10G.

Data network (fabric0/fabric1): Uses the ND-M-V5Q50GV2-D — a Cisco VIC 15427 Quad Port 10/25/50G CNA installed as the Modular LAN-on-Motherboard (mLOM). Requires optical SFP+ connections. Recommended cable: SFP-10G-AOC3M (3-meter AOC), with 5-meter and 7-meter options also available.

Cabling Rules

• Management: use mgmt0 and mgmt1 on the PCIe card. All ports same speed (1G or 10G).
• Data: use the VIC 15427 mLOM. Supported port pairs are Port-1/Port-2 OR Port-3/Port-4. Port-1 corresponds to fabric0, Port-2 to fabric1.
• Do NOT configure port channels or vPC on the switch side for data connections.
• For 25G/50G connections, FEC must be configured: FEC AUTO or FC-FEC on the Nexus 9000 side matches cl74 on the CIMC port.

Power

Dual redundant 1200W AC Titanium power supplies (1+1 redundancy). DC option: 1050W -48V DC available.

ND-NODE-G5S — Current Generation, Small Profile

The G5S shares the UCS C225 M8 platform with the G5L but uses a lower-core-count CPU and less memory, making it appropriate for deployments that don’t require the full resource footprint of the G5L.

CPU

AMD 9454P processor — 48 cores, 2.75GHz base, 290W TDP, 256MB cache, DDR5 4800MT/s memory support.

Storage Configuration

• Boot: 240GB M.2 SATA Micron G2 SSD via Cisco Boot Optimized M.2 RAID controller
• Data (HDD): 2.4TB 12G SAS 10K RPM SFF HDD (4Kn)
• Data (SSD): 960GB 2.5” Enterprise Performance 6G SATA Micron G2 SSD (3x)
• RAID controller: 24G Tri-Mode M1 RAID Controller with 4GB FBWC, 16 drive support

Memory

32GB DDR5-5600 RDIMM, 1Rx4 configuration — half the memory of the G5L.

Network Cards

Management network (mgmt0/mgmt1): Uses the ND-O-ID10GC-D — an Intel X710T2LOCPV3G1L 2x10GbE RJ45 OCP 3.0 NIC. This is an OCP 3.0 form factor card (not PCIe), providing 1Gb copper connection for management only.

Data network (fabric0/fabric1): Uses the ND-P-V5Q50G-D — a Cisco VIC 15425 Quad Port 10/25/50G CNA installed in the PCIe Riser 01 slot. Requires optical SFP+ connections with the same recommended cables as the G5L.

Key Difference from G5L — NIC Location Swap

On the G5L, the management NIC is in PCIe Riser 01 and the data VIC is the mLOM. On the G5S, this is reversed — the data VIC is in PCIe Riser 01 and the management OCP NIC is in the mLOM slot. Verify cabling diagram before racking.

Cabling Rules

• Management: use mgmt0 and mgmt1 on the mLOM card. All ports same speed (1G or 10G).
• Data: use the VIC 15425 in PCIe Riser 01. Same port pairing rules as G5L: Port-1/Port-2 or Port-3/Port-4.
• No port channels or vPC on data connections.

Power

Same dual 1200W AC Titanium / 1050W DC power supply options as G5L.

ND-NODE-L4 — Previous Generation (UCS C225 M6)

The L4 is based on the UCS C225 M6 platform, the previous generation before the M8-based G5 nodes. Still fully supported in ND 4.2 and appropriate for existing deployments or environments where M6-class hardware is already standardized.

Key Specs

• Platform: UCS C225 M6
• Storage: 10-drive SFF backplane, drive bays 1-2 support NVMe SSDs, bays 3-10 support SAS/SATA
• PCIe: Supports up to three half-height riser configurations (1H, 3H, or 2FH options)
• mLOM: Provides management connectivity (mgmt0/mgmt1)

Network Cards

Management network: mLOM card — 1G or 10G, all ports same speed.

Data network: Installed in PCIe Riser 01. Options include:

• 2x10GbE NIC (APIC-P-ID10GC) — copper
• 2x25/10GbE SFP28 NIC (APIC-P-I8D25GF) — optical
• Cisco VIC 1455 Quad Port 10/25G SFP28 — optical

Cabling Notes

• fabric0 = Port-1, fabric1 = Port-2
• When using a 4-port card on the L4, port order from left to right is: Port-4, Port-3, Port-2, Port-1 (reversed from the SE-NODE-G2)
• If using the 25G Intel NIC, disable FEC on the switch port: fec off
• No FEX, port channels, or vPC on data connections

SE-NODE-G2 — Legacy Generation (UCS C220 M5)

The SE-NODE-G2 is the oldest supported physical node, based on the UCS C220 M5 platform. Still supported in ND 4.2 but this is the previous-previous generation. Organizations running SE-NODE-G2 clusters should be planning a hardware refresh to G5-class nodes.

Key Specs

• Platform: UCS C220 M5
• Storage: 10-drive SFF backplane, drive bays 1-2 support NVMe SSDs, bays 3-10 support SAS/SATA
• Rear panel: Dual 1Gb/10Gb Ethernet LOM ports plus dedicated 1Gb management port plus quad 10Gb/25Gb ports

Network Cards

Management network: mLOM card — mgmt0 and mgmt1, 1G or 10G.

Data network: 4-port Cisco VIC 1455 card in PCIe Riser 01 slot.

• fabric0 = Port-1, fabric1 = Port-2
• Port order left-to-right on SE-NODE-G2: Port-1, Port-2, Port-3, Port-4 (standard order, opposite of the L4)

Shared Physical Deployment Requirements

Regardless of which model you’re deploying, these requirements apply to all physical ND nodes:

CIMC must be configured before deployment. Each node needs a reachable CIMC IP address set before you can proceed with ND software installation. Configure via F8 during POST:

• NIC mode: Dedicated
• NIC Redundancy: None
• DHCP or static IP

Serial over LAN (SoL) must be enabled. SoL is required for the connect host command used during initial node setup:

# SSH to CIMC, then:
scope sol
set enabled yes
set baud-rate 115200
commit
show
# Verify com0 is the SoL com port

All nodes must run the same ND software version. If hardware ships with an older version than your target release, deploy the existing version first, then upgrade. Do not skip versions.

Data interfaces must connect to individual host-facing switch ports. No FEX, no port channels, no vPC — on any physical node model.

If connecting to Catalyst switches: Add switchport voice vlan dot1p to switch interfaces where ND nodes connect. Without this, Catalyst switches tag traffic with vlan0 causing data network reachability issues.

Which Model Should You Choose?

For new deployments in 2025, the answer is straightforward:

• Go G5L if you need maximum resources for large-scale fabric management, many concurrent services, or high telemetry volume
• Go G5S if you need current-generation hardware with a smaller resource footprint and lower power draw
• Keep L4 if you have existing L4 hardware in a supported ND release — no need to replace working hardware
• Plan to replace SE-NODE-G2 clusters — they’re supported but two hardware generations behind and appropriate for refresh planning

All models support the same ND software feature set. Hardware generation affects raw performance and capacity, not which ND services are available.