← 목록으로
IT/기술

NVIDIA And Cisco Cram 4x OSFP’s Bandwidth Density Into An OSFP-Sized Module, Pushing AI Fabrics To 12.8 Tb/s With USFP

· 읽기 15분

🔗 원문 전체 보기 → Wccftech

NVIDIA And Cisco Cram 4x OSFP’s Bandwidth Density Into An OSFP-Sized Module, Pushing AI Fabrics To 12.8 Tb/s With USFP

요약

NVIDIA & Cisco have partnered to create USFP, an ultra-dense and compact form factor that drives up to 400 Gb/s speeds. NVIDIA's USFP Is A Compact & Pluggable Switch That Harnesses Optics To Drive Up To 400 Gb/s Per Lane & 12.8 Tb/s Capacity The need for adva…

본문

Hardware Industry NVIDIA And Cisco Cram 4x OSFP’s Bandwidth Density Into An OSFP-Sized Module, Pushing AI Fabrics To 12.8 Tb/s With USFP Hassan Mujtaba • Oct 12, 2026 at 11:00am EDT Comments Add to Preferres Sources
A close-up of an NVIDIA component with visible branding, featuring a metallic design and PCIe connection.

NVIDIA & Cisco have partnered to create USFP, an ultra-dense and compact form factor that drives up to 400 Gb/s speeds.

NVIDIA's USFP Is A Compact & Pluggable Switch That Harnesses Optics To Drive Up To 400 Gb/s Per Lane & 12.8 Tb/s Capacity

The need for advanced fabric, networking, and interconnect solutions is growing as AI factories continue to grow. The bigger the AI factory gets, the more the demand for connecting systems while managing network complexity, power, and costs increases. This not only increases the requirements for increased bandwidth per connection but also more connections per switch.

Related Story NVIDIA DLSS 5 Runs 30% Faster Now, Reduces VRAM Use By 20% & Coming To Hogwarts Legacy & Where Winds Meet In November

To address this, NVIDIA and Cisco have teamed up to design a new optical switch called the USFP. This is an ultra-compact and ultra-dense design that offers 400 Gb/s speeds.

“Building the world’s largest AI factories starts with rethinking the smallest of connections,” said Gilad Shainer, senior vice president 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.”

The USFP, or "Ultra-Dense Small Form Factor Pluggable" module, is said to be the world's first pluggable module built for 400-gig base fabrics. Moving to 400-gig CERTIs doubles the per-lane bandwidth, but it also increases power by 1.5x. It improves power efficiency by 1.3x over 200-gig.

The image is an announcement slide from NVIDIA and Cisco for the USFP—World’s First 400G per Lane Optical Module, highlighting features such as '200G / 400G SerDes per lane,' 'Up to 32 electrical lanes,' '1.3x power efficiency,' and '4x bandwidth density,' with ecosystem partners Amphenol and TE Connectivity.

This USFP module delivers four times the bandwidth density of 200-gig OSFP, making high-density 1.6 terabit fabrics practical. USFP supports up to 32 electrical lanes, which gives you 12.8 terabits per second of capacity. NVIDIA and Cisco plan to offer initial products with 16 lanes & 6.4 terabits per second capacity.

In terms of size, the USFP module is similar to existing OSFP modules but delivers 4x bandwidth uplift and front-panel bandwidth density at 1.3x the efficiency. This scale allows AI factories to increase their compute infrastructure within the same network footprint.

“AI builders need an interconnect architecture that can scale performance without locking them into a more complex operating model,” said Bill Gartner, senior vice president and general manager of optical systems and optics at Cisco. “Cisco and NVIDIA recognized that scaling pluggable optics to 400Gb/s per lane would require more than another incremental extension of today’s module designs.

Through our collaboration, USFP is designed to deliver a new level of bandwidth density while preserving the serviceability, deployment flexibility and operational simplicity of pluggable optics. By taking a full-stack approach — from silicon and systems to optics and the network operating system — we can optimize every layer to deliver the performance and scale the next generation of AI networks will require.”

To prepare for these denser fabrics, NVIDIA is working with Amphenol and TE Connectivity to help accelerate the ecosystem adoption. This collaboration will further build upon existing commitment to the NVIDIA Spectrum-X Ethernet networking solution and Cisco's Silicon One-based Ethernet networks to work together seamlessly.

Hassan Mujtaba Photo

About the author: A Software Engineer by training and a PC enthusiast by passion, Hassan Mujtaba serves as Wccftech's Senior Editor for hardware section. With years of experience in the industry, he specializes in deep-dive technical analysis of next-generation CPU and GPU architectures, motherboards, and cooling solutions. His work involves not only breaking news on upcoming technologies but also extensive hands-on reviews and benchmarking.

Follow Wccftech on Google to get more of our news coverage in your feeds.

Deal of the Day

Further Reading

NVIDIA RTX 5080 24 GB GPU Expected To Arrive In Q1 2027 As The New Flagship Blackwell GeForce GPU

A slide titled 'Announcing NVIDIA Contributes Vera Rubin MGX Innovations' features elements like a 'Cableless, Hoseless, Fanless Tray,' 'Liquid Cooled Busbar,' and 'In-tray MQD/MQDB.'

NVIDIA Hands Its Vera Rubin Rack Designs To OCP, Stripping Compute Trays Of Every Cable, Hose, And Fan

Four NVIDIA graphics cards are installed in a computer setup with visible branding and heatsinks.

NVIDIA’s 9-Year-Old Tesla V100 Still Packs Some Inference Muscle; 4 GPUs On An NVLink Board Can Outperform M5 Ultra, DGX Spark In Various AI Benchmarks By Up To 230%

A hand holding an AMD RDNA chip and an SK hynix GDDR7 chip against a background with the number 5.

AMD Rumored To Launch RDNA 5 In Second Half Of 2027; Also Quashes Rumors Of Next-Gen Using 4 GB GDDR7 Memory

Read all comments on NVIDIA And Cisco Cram 4x OSFP’s Bandwidth Density Into An OSFP-Sized Module, Pushing AI Fabrics To 12.8 Tb/s With USFP

Trending Stories

Popular Discussions

출처: Wccftech

작성자: Hassan Mujtaba

원문 게시일: 2026-10-12T15:00:04Z

이 글은 NewsAPI를 통해 자동으로 수집된 콘텐츠입니다. 원문은 위 링크를 통해 확인하세요.

Technology Cybersecurity Cloud IT Wccftech
← 목록으로