NVIDIA and Cisco have announced plans for a new Ultra-Dense Small Form Factor Pluggable (USFP) optical module designed to increase network capacity and power efficiency in next-generation artificial intelligence (AI) factories. The technology supports electrical signals of 400Gb/s per lane and is designed to deliver up to 12.8Tb/s of overall capacity. (12/10)

The companies are collaborating with TE Connectivity and Amphenol to develop USFP cages, connectors and modules. They also plan to establish an open industry specification through an upcoming community-led multi-source project.

USFP Supports Up to 12.8Tb/s of Network Capacity

USFP is designed to help AI factory operators connect more computing systems while managing network complexity, power consumption and costs. The new form factor aims to support the transition from 200Gb/s-per-lane to 400Gb/s-per-lane Ethernet AI fabrics without requiring a complete overhaul of existing data center infrastructure.

As AI models grow and demand for inference expands, AI factories need greater bandwidth per connection and more connections per switch. When a switch reaches its capacity, expanding the network can require additional switch layers and optical links, increasing latency, power consumption and costs.

The first planned USFP-based transceivers will feature 16 full-duplex lanes in a footprint comparable to an Octal Small Form Factor Pluggable (OSFP) module. The design supports both switches and SuperNICs, with the potential to expand to 32 lanes or more in the future.

By doubling the bandwidth per lane and increasing the number of lanes, USFP is designed to connect more 1.6Tb/s endpoints without adding network tiers. The resulting configuration can provide up to 12.8Tb/s of total capacity.

New Optical Design Increases Bandwidth Density and Power Efficiency

Compared with 200Gb/s-per-lane pluggable modules, USFP is designed to deliver four times the bandwidth and front-panel bandwidth density, along with approximately 1.3 times the power efficiency.

These capabilities are intended to help AI factory operators increase computing capacity within the same network footprint while reducing networking's share of the overall AI factory power budget.

“Building the world’s largest AI factories starts with rethinking the smallest of connections,” said Gilad Shainer, senior vice president at NVIDIA. “With USFP, NVIDIA and Cisco are building an ecosystem to reinvent pluggable optics — connecting more computing with greater efficiency to power the next generation of AI factories.”

Bill Gartner, senior vice president and general manager of optical systems and optics at Cisco, said AI builders need an interconnect architecture that can increase performance without creating a more complex operating model.

“AI builders need an interconnect architecture that can scale performance without locking them into a more complex operating model,” Gartner said.

He explained that scaling pluggable optics to 400Gb/s per lane requires more than incremental changes to existing module designs. The collaboration aims to increase bandwidth density while maintaining the serviceability, deployment flexibility and operational simplicity of pluggable optics.

Gartner also highlighted the companies' full-stack approach, which covers silicon, systems, optics and network operating systems to support the performance and scale required by next-generation AI networks.

Connector and Power Design Supports Higher-Bandwidth Networks

USFP uses a new two-dimensional array connector designed for electrical signals of 400Gb/s per lane. Its two-step insertion design engages the high-speed contacts after the module is inserted, reducing contact wear, improving signal integrity and making insertion force more predictable.

The connector is designed to support expansion to 32 full-duplex lanes or more. It also offers improved performance at 200Gb/s per lane because the connector is designed to support 400Gb/s operation.

The module supports power dissipation of up to 100 watts without requiring a separate, dedicated embedded cooling circuit within the transceiver. According to the announcement, this design reduces switch design complexity and improves reliability while enabling higher-bandwidth network fabrics.

All USFP configurations support optical designs intended to reduce latency and power consumption by simplifying signal processing. Support for different optical technologies and connection distances also gives system designers flexibility to meet the requirements of individual AI factory deployments.

These features are designed to preserve the established deployment, serviceability and manufacturability of existing pluggable transceivers while supporting higher network capacity.

NVIDIA and Cisco Develop a Multi-Source USFP Ecosystem

NVIDIA and Cisco are working with TE Connectivity and Amphenol to develop USFP cages, connectors and modules, with additional partners expected to join. The shared form factor is intended to provide suppliers with a common foundation for developing high-density pluggable infrastructure for 400Gb/s-per-lane networks.

The collaboration builds on the companies' existing work to ensure that NVIDIA Spectrum-X Ethernet networking and Cisco Silicon One-based Ethernet networks operate together seamlessly. The effort aims to simplify AI infrastructure deployment for enterprises, neoclouds and sovereign clouds.

NVIDIA and Cisco plan to create an open industry specification for USFP through an upcoming community-led multi-source project.

NVIDIA describes itself as a global leader in AI and accelerated computing, while Cisco develops networking and security technologies for organizations in the AI era.

PHOTO: NVIDIA

This article was created with AI assistance.

We make every effort to ensure the accuracy of our content, some information may be incorrect or outdated. Please let us know of any corrections at [email protected].