NVIDIA and Cisco's planned USFP pluggable optics: what to know

What is USFP designed to change?

USFP, short for Ultra-Dense Small Form Factor Pluggable, is a proposed optical-module form factor for AI data-center networks. The first planned transceivers would fit an OSFP-sized footprint and support switches and SuperNICs, according to NVIDIA and Cisco.

The companies say the design is intended to fit more network capacity into a given space while preserving the serviceability of pluggable optics. Cisco senior vice president Bill Gartner said the goal is to increase bandwidth density while retaining the flexibility to deploy and service the modules. That is the design rationale, not evidence of how USFP would perform in an operating network.

The proposed connector uses a two-dimensional array of contacts and a two-step insertion process. NVIDIA and Cisco say those features are intended to reduce contact wear and improve signal integrity. The announcement does not include reliability or signal-performance results from hardware tests.

What bandwidth and power gains do the companies project?

NVIDIA and Cisco compare USFP with pluggable modules that use 200Gb/s-per-lane signaling. They say the proposed design could deliver four times the bandwidth density and approximately 1.3 times the power efficiency. Those comparisons are company projections; the announcement provides no independent measurements to verify them.

The companies also say USFP is designed to support up to 100 watts of power dissipation without a separate dedicated cooling circuit embedded inside the transceiver. That figure describes a design target. It does not establish how much power a deployed system would use or what its cooling requirements would be.

If implemented as described, more capacity within the same front-panel footprint could let operators connect more equipment. Whether USFP would reduce the need for additional network tiers or cut power use in practice remains untested.

Who is developing USFP, and when could it be available?

NVIDIA and Cisco say they are working with TE Connectivity and Amphenol on USFP cages, connectors and modules. They plan to develop an open industry specification through a community-led project intended to support multiple suppliers. The announcement does not give a timetable or further details for that work.

NVIDIA senior vice president Gilad Shainer presented the effort as a way to connect more computing systems with greater efficiency. The companies also say their collaboration builds on work to make NVIDIA Spectrum-X and Cisco Silicon One-based Ethernet networks work together. They do not say that USFP hardware is already operating in those networks.

For data-center operators, the next useful evidence would include module availability, compatibility, measured power use, reliability and performance under operating conditions. The companies have outlined a design and a plan for an industry specification, but have not announced when operators can evaluate the hardware.