NVIDIA LinkX Ethernet Optical Transceivers

Discover the next generation of transceivers for GPU-accelerated computing.

NVIDIA® LinkX® Optics Ethernet transceivers are used to create high-speed, 100G–400G links supporting every configuration, reach, and speed in networks requiring detachable optical connectors. LinkX transceivers are designed to create high-speed 25G–400G optical links in Ethernet switching networks, Ethernet storage fabrics (ESF) linking hard-disk-drive (HDD) and flash memory subsystems, and NVIDIA GPU-based, end-to-end AI systems. All transceivers are extensively tested in live GPU and switching systems and used in NVIDIA-Certified Systems™.

Long Reaches and Many Configurations

Transceivers are available in one, four, and eight channels in non-return-to-zero (NRZ) and 50G-4-level pulse-amplitude-modulation (PAM4) for 850nm multi-mode and 1310nm single-mode in small form-factor pluggable (SFP), quad small form-factor pluggable (QSFP), and QSFP double density (QFSP-DD) form factors.

100GbE DR1 500m transceiver has 1x 100G-PAM4 LC 1310nm optics and includes an internal gearbox to convert to 4x 25G NRZ electrical output. It’s often used with 400G DR4 transceivers with splitter fibers to link new 400G-PAM4 systems to legacy 100G (4x 25G-NRZ)-based systems.

With 100 percent factory testing in end-to-end configurations, all LinkX transceivers offer extremely low BER compared to industry standards and competitors. There are a wide variety of transceiver options available for GPUs, CPUs and storage networks in every data center configuration.

NVIDIA LinkX Ethernet Optical Transceivers


                                                                PORT SPEEDS






                  MODULATION TYPES

10/25G NRZ,

* MPO: multi-fiber push-on

Connectorized Transceivers for Long-Reach Links

Transceivers are widely used in data centers to connect NVIDIA GPU-based AI systems to top-of-rack (TOR) switches, to leaf-spine and spine-super spine switches in link budgets requiring large optical power or long fiber reaches. The 500m PSM4/DR4, 2km CWDM4/FR4, and 10km LR4 are used to traverse complex networks over long single-mode fiber reaches. Short-reach runs up to 100m along system rows and the multi-mode SR/SR4/SR8 versions are the most popular, low-cost options.

Key Features

  • Extensively tested in live NVIDIA switching and GPU-based systems
  • Low-cost, low- latency, long reaches
  • Extensive 25/50/100/200/400G product offering for every need
  • Tested and certified in NVIDIA GPU, DGX™, EGX™, and DPU systems


  • Many parts designed and manufactured by LinkX Optics
  • 100 percent factory tested in live switching and GPU-based systems
  • Designed to beyond IEEE industry BER standards for reliable operation under harsh operating conditions

High-Speed Interconnects

The latest rapid growth in data center traffic and cloud computing has resulted in tremendous increases in machine-to-machine traffic due to rapid adoption of server virtualization, software-defined networks (SDNs), and in-cloud computing. Now add the massive bandwidth requirements for networking terabytes of data for training Artificial Intelligence (AI) machine learning (ML) GPU-based systems.

This is driving massive demands for high-bandwidth and low-latency networks. High-speed DAC and AOC cables and optical transceiver interconnects are playing a critical role in these modern data center technologies.

Ethernet Optical Transceivers

Ethernet Optical Transceivers

* Check product briefs for detailed product codes and specifications.
** An 850nm multi-mode transceiver’s typical reach is 70m with OM3 and 100m with OM4 fiber.