TRX vs. LPO vs. CPO: Comparing Transceiver Technologies for 400G/800G/1.6T Optical Modules

OptechTW

Introduction

As data center bandwidth demand grows rapidly—driven by AI workloads, RoCE fabrics, and ultra-low latency switching—the choice of optical transceiver architecture becomes crucial. Today, three architectures dominate the landscape for high-speed modules:

  • TRX (Traditional Transceivers)

  • LPO (Linear Pluggable Optics)

  • CPO (Co-Packaged Optics)

Each of these has unique performance characteristics, benefits, and limitations across 400G, 800G, and 1.6T generations.

 What is TRX (Traditional Optical Transceiver)?

TRX modules integrate DSP (Digital Signal Processing), driver, and TIA components inside the module itself.

Pros:

  • Fully digital with FEC (Forward Error Correction)

  • Host-agnostic, simplifying design

  • Mature and widely deployed (e.g., QSFP28, QSFP-DD, OSFP)

Cons:

  • High power consumption (especially in 800G+)

  • Larger form factor (limiting density at 1.6T)

  • Thermal management becomes a challenge

What is LPO (Linear Pluggable Optics)?

LPO removes the DSP from the transceiver, pushing analog signaling directly to/from the host ASIC.

Pros:

  • Lower power consumption (~30–50% less than TRX)

  • Reduced latency

  • Smaller form factor for high-density switches

Cons:

  • Requires analog-aware host ASICs (limited to certain switch vendors like Broadcom/NVIDIA)

  • Challenging signal integrity and link margin

  • Susceptible to system noise and skew

Best Use Cases:

  • AI/ML clusters with NVIDIA/Broadcom RoCE

  • Hyperscale data centers pushing toward lower cost/power

What is CPO (Co-Packaged Optics)?

CPO integrates the optical engines and switch ASIC into the same package, enabling ultra-short electrical traces.

Pros:

  • Best power efficiency at >1.6T speeds

  • Highest density possible

  • Eliminates signal integrity issues from host-to-module

Cons:

  • Very high integration complexity

  • Difficult to service/replace failed optics

  • Still early in adoption (limited ecosystem)

 TRX vs. LPO vs. CPO – Feature Comparison Table

Feature TRX LPO CPO
DSP Location In-module Host ASIC Co-packaged
Power Efficiency Moderate High Highest
Latency Moderate Very Low Very Low
Host Compatibility Universal Limited Proprietary
Hot Swappable Yes Yes No
Ecosystem Maturity High Medium Low
Max Speed (Today) 800G–1.6T 800G–1.6T 1.6T–3.2T

Technology Use Across 400G, 800G, 1.6T Modules

400G:

  • TRX (e.g., QSFP-DD DR4, FR4) remains dominant

  • LPO is being trialed in hyperscalers

  • CPO is not common at this level

800G:

  • TRX is widely adopted (e.g., 800G OSFP 2xFR4)

  • LPO gaining traction in AI/ML clusters (especially NVIDIA CX7 & Broadcom Tomahawk 5 platforms)

  • CPO early prototypes being tested

1.6T:

  • LPO required to meet power limits in air-cooled environments

  • CPO emerging as the go-to architecture for switch-ASIC vendors

Conclusion

Choosing between TRX, LPO, and CPO depends on your deployment scenario:

  • TRX offers maturity and compatibility—ideal for general-purpose data centers.

  • LPO delivers ultra-low latency and power—best suited for AI/ML networks like NVIDIA DGX clusters.

  • CPO is the future of ultra-high-speed networking, ideal for >1.6T deployments with extreme scale.

Back to blog

Contact form