In February 2025, data center architect David was struggling to select optical modules for his 32-port 400G spine switch. The datasheets listed six variants — SR8, SR4, DR4, FR4, 2xFR4, and LR4. Each had different reach, fiber requirements, connectors, and pricing. After hours of comparing supplier datasheets, he still couldn’t make a confident decision. He needed a clear, side-by-side guide covering all options.
The 400G OSFP transceiver family provides a complete portfolio engineered for every reach and application in modern data center networks — from 50-meter NVIDIA InfiniBand links to 10km DCI metro connections.
Whether you are planning a greenfield 400G deployment, upgrading from 100G QSFP28, or optimizing an existing fabric, understanding the differences between these six variants is essential. This guide walks you through each type with technical specifications, ideal applications, breakout options, and a practical selection framework.
Looking for the right 400G OSFP modules for your network? Browse our 400G OSFP catalog for SR8, SR4, DR4, FR4, 2xFR4, and LR4 variants with platform-tested compatibility.
The table below summarizes the major 400G OSFP module variants currently used in modern data center and AI networking environments.
| Module Type | Reach | Fiber | Connector | Power | Modulation | Application |
| OSFP SR8 | 100m | OM4 MMF | MTP/MPO-16 | ~10W | 8×50G PAM4 | Intra-rack Ethernet |
| OSFP SR4 | 100m | OM4 MMF | MPO-12 UPC | ~8.5W | 4x100G PAM4 | InfiniBand NDR |
| OSFP DR4 | 500m | OS2 SMF | MPO-12 APC | ~10W | 4x100G PAM4 | Spine-leaf, campus |
| OSFP FR4 | 2km | OS2 SMF | Duplex LC | ~12W | 4×100G PAM4 LAN-WDM | DCI short-haul |
| OSFP 2xFR4 | 2km | OS2 SMF | Dual duplex LC | ~14W | 2×200G FR4 | Cost-optimized DCI |
| OSFP LR4 | 10km | OS2 SMF | Duplex LC | ~12W | 4×100G PAM4 LAN-WDM | Metro aggregation |
Several clear patterns emerge from this comparison.
The correct module depends on the specific combination of reach, protocol, switch platform, and existing fiber infrastructure.
The OSFP SR8 is the standard short-reach multimode module for 400G Ethernet networks. This module has eight 53.125 Gbps PAM4 electrical lanes that align with eight 850nm VCSEL optical lanes, appealing to 400 Gbps over multimode fiber.
OSFP SR8 uses:
Each lane operates at 53.125 GBaud PAM4, delivering an effective data rate of 100Gbps per lane pair.
On the optical side, the module uses:
This parallel optics architecture avoids wavelength multiplexing complexity and keeps latency and module cost relatively low.
Regarding the possible link spans, OSFP SR8 supports up to 70 meters over OM3 and 100 meters over OM4 multimode fiber. Some manufacturers extended OM4 range to 150 meters through the use of improved link budgets, but 100 m is the IEEE 802.3bs standard.
For intra-rack switch-switch links and very short leaf-spine interconnects within the same row, SR8 is unmatched. For AI and HPC clusters that are using Ethernet (as opposed to InfiniBand), SR8 is preferred to links between a ToR switch and a spine switch down the same hot aisle.
OSFP SR8 commonly supports breakout into 4×100G SR2 or 8×50G connections, depending on switch ASIC support. Some platforms also support 2×200G SR4 breakout configurations.
OSFP SR8 modules are available in two thermal formats:
Finned-top modules are typically preferred in high-power switch environments where airflow and thermal density are more demanding.
OSFP SR4 is primarily associated with NVIDIA InfiniBand NDR deployments. While physically similar to SR8, SR4 uses a different lane architecture optimized for NVIDIA’s 4×100G PAM4 InfiniBand design.
Contrary to the eight 50G lanes on the SR8, the OSFP SR4 deploys four 100 Gbps PAM4 lanes. Each lane operates at 53.125 GBaud using PAM4 signaling. This architecture aligns with NVIDIA ConnectX-7 adapters, BlueField DPUs, and Quantum-2 InfiniBand switches.
SR4 uses an MPO-12 UPC connector instead of MPO-16. This difference is important during structured cabling planning because SR4 and SR8 are not directly interchangeable at the cabling layer.
OSFP SR4 is commonly used in:
For NVIDIA-specific deployment guidance, see our OSFP for NVIDIA InfiniBand NDR Guide.
NVIDIA commonly references SR4 modules using part numbers such as:
When deploying third-party optics, platform validation is strongly recommended. For deeper InfiniBand-specific guidance, see our OSFP for NVIDIA InfiniBand NDR guide.
Need OSFP SR4 modules tested for NVIDIA ConnectX-7? Contact our optical engineers for platform compatibility verification.
The OSFP DR4 is the most versatile 400G OSFP module for modern data center architectures. It transmits 4x100G PAM4 over single-mode fiber, reaching 500 meters with no optical amplification.
DR4 transmits:
The module uses an MPO-12 APC connector with:
for a total of 8 active fibers.
Like most 100G-per-lane PAM4 optics, DR4 requires KP4 forward error correction (FEC) to maintain signal integrity and BER performance.
OSFP DR4 is there to be in a spine-leaf architecture where racks are situated too far apart for multimode reach along with catering to last-row-to-last-row connections and point-to-point connections to campus buildings within up to 500 meters.
One of the standout feature advancements is breakout. Using an MPO-12 to 4× duplex LC breakout harness, one 400G DR4 port can connect to four 100G DR links. This significantly improves port efficiency during 100G-to-400G migration.
OSFP FR4 extends 400G connectivity to 2 kilometers while using only a duplex LC fiber pair. This makes it highly attractive for environments with limited fiber availability.
FR4 multiplexes four 100G PAM4 wavelengths onto a single duplex fiber pair using LAN-WDM technology. It puts four wavelengths, 1271 nm, 1291 nm, 1311 nm, and 1331 nm, on one fiber pair. An offer of 100 Gbps PAM4 per wavelength results in an aggregated 400G.
The most outstanding advantage of FR4 over DR4 is that it uses duplex LC connectors. While DR4 requires MPO-12 (8 fibers active), FR4 requires just 2 fibers. For data centers with existing duplex LC infrastructure, FR4 often reduces deployment complexity and cabling cost.
FR4 is commonly used for:
Physical optical breakout is generally not supported for FR4 because all wavelengths are multiplexed onto a single fiber pair. If 4×100G breakout is required, DR4 is typically the preferred architecture.
2×FR4 differs from the other modules in this guide because it is primarily used in 800G environments rather than standard standalone 400G switching.
A 2×FR4 module contains:
The module typically exposes:
2×FR4 is commonly used for:
Because it consolidates multiple links into a single module, it can improve switch port utilization in high-density environments.
2×FR4 modules are among the highest-power OSFP optics currently deployed, typically operating between 14W and 18W. Thermal planning is especially important in fully populated AI switch systems.
OSFP LR4 extends 400G transmission reach up to 10 kilometers over single-mode fiber. It is designed for longer-distance enterprise and metro connectivity where FR4 reach is insufficient.
LR4 uses the same general wavelength approach as FR4 but with higher optical launch power, improved receiver sensitivity and larger optical link budget. This enables stable transmission across longer fiber spans.
OSFP LR4 rules the metro Ethernet aggregation, 5G transport network (backhaul, midhaul), and binds up regional data center interconnects. It’s useful for inter-building WAN connections in campuses, corporate setups, and selected government facilities.
At 10km distances, optical budget planning becomes increasingly important. Standard single-mode fiber attenuation is approximately 0.25 dB/km at 1310 nm. After accounting for: connector loss, splice loss and patch panel loss, most LR4 deployments still maintain sufficient operating margin under normal fiber conditions.
Selecting the correct 400G OSFP module becomes much easier when approached step by step.
Reach is the primary selection criterion:
Existing fiber plant constrains your options:
A short reach of 50 meters with NVIDIA InfiniBand NDR deployments, SR4 OSFP is required. Ethernet deployments use SR8 for the same range.
For a long reach, the SR4/SR8 distinction is less relevant as beyond DR4. DR4, FR4, and LR4 are widely used for both Ethernet and InfiniBand environments, depending on platform support.
The module consumes as much as 14W (2xFR4) in power, and as little as 8.5W (SR4). Thus, a fully loaded switch of 32 ports would need about 270W to 450W in power consumption by all of its optics. Verify the power and cooling budgets of your switch before going for a high-power variant like 2xFR4.
A pragmatic approach to selection:
David, from our opening scenario, ended up specifying a mix: OSFP DR4 for 16 cross-row links (500m), OSFP SR8 for 12 intra-row leaf connections (under 100m), and OSFP FR4 for 4 DCI uplinks (2km). The blended approach reduced his total optics cost by 28% versus a uniform DR4 deployment while matching each port to its actual requirement.
Want a platform-specific recommendation for your 400G deployment? Request a quote and our engineers will review your switch model, link distances, and protocol requirements.
Power planning is critical for dense 400G deployments. Here is the complete power comparison:
| Module Type | Typical Power | Max Power | Thermal Variant |
| OSFP SR4 | 8.5W | 10W | Flat-top or Finned-top |
| OSFP SR8 | 10W | 12W | Flat-top or Finned-top |
| OSFP DR4 | 10W | 12W | Flat-top or Finned-top |
| OSFP FR4 | 12W | 14W | Flat-top or Finned-top |
| OSFP 2xFR4 | 14W | 18W | Finned-top recommended |
| OSFP LR4 | 12W | 14W | Flat-top or Finned-top |
For a 32-port switch fully populated with high-power 2xFR4 modules, total optics power approaches 450W. Add the switch base power (200-400W) and a single 1U switch can draw 850W just for the data plane. AI infrastructure deployments with multiple 400G switches per rack need careful power and cooling planning.
For broader context on form factor decisions, see our QSFP-DD vs OSFP comparison guide covering the trade-offs between competing 400G architectures.
The 400G OSFP transceiver types form a complete portfolio designed for every modern networking application. Whether you are building an AI training cluster with NVIDIA InfiniBand, expanding a hyperscale spine-leaf fabric, or extending metro DCI capacity, there is a specific 400G OSFP variant engineered for the job.
Key takeaways include:
The best module is not necessarily the longest reach or highest specification—it is the module that best matches your actual network architecture, cabling infrastructure, and future scalability requirements.
Ready to source 400G OSFP transceivers? Contact AscentOptics for SR8, SR4, DR4, FR4, 2xFR4, and LR4 modules with platform validation, MSA compliance, and factory-direct pricing for deployments at any scale.
OSFP SR8 uses eight 50G PAM4 optical lanes and is primarily deployed in 400G Ethernet environments. OSFP SR4 uses four 100G PAM4 lanes and is commonly used in NVIDIA InfiniBand NDR deployments.
OSFP LR4 supports transmission distances up to 10 kilometers over single-mode fiber and is commonly used for metro aggregation and long-distance enterprise connectivity.
Yes. A 400G DR4 port can typically break out into four independent 100G DR connections using an MPO-12 to 4× duplex LC breakout harness.
OSFP generally provides higher thermal headroom and improved power handling compared with QSFP-DD, making it popular in AI and HPC deployments, especially in NVIDIA-based environments.
Q5：Which 400G OSFP module uses the fewest fibers?
FR4 and LR4 use duplex LC connectors and require only two fibers, making them more fiber-efficient than parallel-optics modules such as SR8 and DR4.
NVIDIA InfiniBand documentation
