QSFP-100G-SR4 vs QSFP-100G-LR4 for Nexus 9000

Use QSFP-100G-SR4-S for short, in-row 100G links and LR4 for long-haul. SR4 runs over multimode fiber (MMF) with an MPO-12 connector and reaches 100m on OM4; QSFP-100G-LR4-S runs over single-mode fiber (SMF) with a duplex LC connector and reaches 10km. The deciding factor is your fiber plant and distance: if leaf and spine sit in the same row or adjacent racks on MMF, SR4 is cheaper and supports 4x25G breakout; if you are crossing a data hall, between buildings, or already have SMF, choose LR4. Both are QSFP28 modules supported on Nexus 9300 and 9500 100G ports.

The short answer: fiber type and distance decide it

For a Nexus 9000 leaf/spine fabric, the choice between QSFP-100G-SR4-S and QSFP-100G-LR4-S comes down to two physical facts you cannot change with config: what fiber you have in the ground, and how far the two switches sit apart.

SR4 is the short-reach module. It transmits four parallel 25G lanes over multimode fiber using an MPO-12 connector at 850nm, and it reaches 100m on OM4 (70m on OM3). That covers virtually any intra-row or adjacent-rack leaf-to-spine run inside a single data hall. It is the lower-cost option and, because its four lanes are physically separate fibers, it is the one that supports 4x25G breakout.

LR4 is the long-reach module. It multiplexes four 25G lanes onto four CWDM wavelengths around 1310nm and sends them down a single single-mode fiber pair using a standard duplex LC connector, reaching 10km. That is what you want for spine uplinks crossing a large data center, inter-building links on campus, or any site already cabled with SMF. LR4 cannot break out, because all four lanes ride one fiber pair.

If your runs are short and you have MMF, SR4 saves money. If they are long or you have SMF, LR4 is the only one that will link. Browse the full lineup in Optics & Transceivers.

SR4 vs LR4 at a glance

Both modules are QSFP28 form factor, both deliver a single 100GbE port, and both support Digital Optical Monitoring (DOM) so you can read live optical power on the switch. Everything else differs:

Why only SR4 can break out to 4x25G

This is the most practically useful difference for a leaf/spine build, and it is purely physical.

SR4 carries its four 25G lanes on four separate pairs of fibers inside the MPO-12 cable. Because those lanes are physically distinct, you can split one 100G QSFP28 port into four independent 25G links using an MPO-to-4x-LC breakout harness, terminating each lane into a separate SFP-25G-SR-S module in a top-of-rack switch or 25G-capable server NIC. One spine port becomes four leaf-facing 25G ports.

LR4 multiplexes all four lanes onto different wavelengths sharing a single fiber pair. There is no way to separate them in the field, so LR4 is always a single 100G link, end to end.

If your design calls for 100G spine ports fanning out to 25G access, SR4 plus breakout is the standard, cost-effective pattern. On NX-OS you configure the port group for breakout, for example `interface breakout module 1 port 1 map 25g-4x`, then reload the affected ports. Confirm the specific platform's breakout-capable ports in its hardware guide before you cable, because not every QSFP28 port on every Nexus 9000 model supports breakout.

Other 100G options worth knowing

SR4 and LR4 are the two you will reach for most, but a Nexus 9000 fabric often mixes in a few neighbors depending on distance and budget. Reach for these when SR4 is too short and LR4 is more than you need, or when a copper run is short enough to skip optics entirely:

• QSFP-100G-CWDM4-S — duplex LC over single-mode to 2km. The sweet spot when you have SMF and need more than 100m but nowhere near 10km, at a lower cost than LR4.
• QSFP-100G-AOC (Active Optical Cable) — a fixed-length optic-plus-fiber assembly (1m to 30m) with no separate connectors. Clean for structured cabling between rows.
• QSFP-100G-CU DAC (Direct Attach Copper) — passive twinax for very short in-rack links, typically 1m to 3m or 5m. The cheapest way to connect a leaf to a spine or server one or two racks over; no optics needed.

A common, sensible mix is DAC inside the rack, SR4 across the row, and LR4 for the longest spine uplinks. For switch selection on either end of these links, see our Nexus 9300 vs 9500 comparison and browse current Cisco Switches.

Nexus 9000 and NX-OS support notes

All four modules above are QSFP28 optics supported across the Nexus 9300 and 9500 100G platforms, but support is always tied to the specific switch model and NX-OS release, so verify the Cisco Optics-to-Device Compatibility Matrix for your exact PID and software train before ordering.

A few operational pointers:

• Verify the link, not just the LED. Run `show interface transceiver details` to read per-lane optical Tx/Rx power, temperature, and current. This is how you catch a dirty MPO connector or a marginal long run before it flaps.
• Mind OM3 vs OM4 with SR4. SR4 only reaches 100m on OM4; on older OM3 plant you are capped at 70m. Measure your longest run against the right number.
• Breakout needs a reload. Changing a port to breakout mode on NX-OS resets the port group, so plan it into a maintenance window.
• Match the connector to the patch panel. SR4 needs MPO-12 cassettes and trunks; LR4 and CWDM4 use ordinary duplex LC. Ordering the wrong patch leads is the most common day-one delay.

Genuine Cisco-coded optics also avoid the `%TRANSCEIVER ... not supported` log noise and ensure DOM and TAC support work as expected.

Quick decision guide

Use this to land on a part number fast:

• In-rack, 1–5m, cheapest: QSFP-100G-CU DAC.
• Across the row, MMF, up to 100m: QSFP-100G-SR4-S.
• Need to fan 100G out to 4x25G: QSFP-100G-SR4-S with an MPO-to-LC breakout harness into SFP-25G-SR-S.
• SMF, 100m to 2km: QSFP-100G-CWDM4-S.
• SMF, up to 10km / inter-building: QSFP-100G-LR4-S.

When you are ready to spec quantities and patching for both ends of the fabric, we can build the optics-and-switch bill of materials for you. Send the link distances and fiber type and request a quote — as an authorized Cisco partner we will confirm compatibility against the device matrix before anything ships.