Cisco and NVIDIA have announced plans for a new pluggable optical module, called Ultra-Dense Small Form Factor Pluggable, or USFP, aimed at the large data centers the companies call AI factories.
The module is designed to carry electrical signals at 400 gigabits per second per lane, twice today's 200-gigabit lanes, for a total capacity of up to 12.8 terabits per second. The companies said it would let operators grow their Ethernet AI networks while keeping the plug-in transceiver model data centers have used for several generations.
Compared with optics built on 200-gigabit lanes, USFP is designed to deliver four times the bandwidth density and about 1.3 times the power efficiency. The companies said that would let operators link more servers without stacking on extra switch layers, while spending less energy per bit.
The first USFP transceivers are planned with 16 full-duplex lanes in a package about the size of an existing OSFP module, for use in both switches and SuperNICs, with room to grow to 32 lanes or more later. With twice the lanes and twice the speed per lane, the design is meant to connect more 1.6-terabit endpoints without extra network tiers.
"AI builders need an interconnect architecture that can scale performance without locking them into a more complex operating model," said Bill Gartner, Cisco's senior vice president and general manager of optical systems and optics.