On-Prem

Networks

Synopsys unleashes Ethernet on steroids with 1.6T blueprint to turbocharge AI training

Design aims to sate the bandwidth hunger of next-gen networks


Semiconductor design outfit Synopsys is targeting datacenter customers with a 1.6 terabit (1.6T) Ethernet blueprint intended to enable chips for demanding AI networking applications.

The California-based company, best known as a provider of electronic design automation (EDA) tools, claims its latest work is the first complete 1.6T Ethernet IP solution for chip designers to build products with an eye on the burgeoning bandwidth demands of AI processing.

Although the IEEE is not expected to finalize the latest iteration of the 1.6 Tb Ethernet standard until 2026, Synopsys says its design provides chipmakers with the IP they need today to create the fast networking silicon. A baseline set of features is expected to be completed this year via the 802.3dj task force.

Synopsys has based its design on the latter, supporting 4 x 400G, 2 x 800G, and 1.6T Ethernet rates with 112 Gbps and 224 Gbps SerDes. As well as 1.6T media access control (MAC) and physical coding sublayer (PCS) components, the solution has 224G Ethernet PHY and verification IP.

With the current craze for AI driven by the latest generational models, data networks in the cloud datacenters where these are most commonly hosted have felt the strain from the massive volumes of data required for training them.

This calls for significantly faster Ethernet speeds in order to keep pace with the ever-increasing data demands, according to Synopsys senior VP of IP marketing and strategy John Koeter.

"Our complete IP solution for 1.6T Ethernet, pre-verified subsystems... allow designers to confidently integrate the necessary functionality into their SoCs with less risk," he claimed.

The move has been welcomed by industry consortium the Ethernet Alliance.

"With growing demands from large language modeling, HPC simulation, and AI training in hyperscale datacenters, network boundaries are crossing over the terabits per second threshold," chairman Peter Jones, also a Distinguished Engineer at Cisco, said in a statement.

"The availability of development tools capable of meeting these needs is critical to the success of next-generation Ethernet standards addressing this market."

Synopsys said that its 1.6T Ethernet MAC and PCS components cut the required silicon area by 50 percent while reducing latency by 40 percent via a patented Reed-Solomon Forward Error Correction architecture.

All the components of the 1.6T Ethernet blueprint are available now, and it has already been adopted by multiple customers, Synopsys claimed.

The design has been validated with multiple hardware platforms, PHYs, and Ethernet verification suites across a broad range of production processes and foundries, the company said. ®

Send us news
Post a comment

AMD says datacenter still king for profit margins amid AI buzz

CFO talks GPU development, strategy and market dynamics at Nasdaq Investor Conference

Payoff from AI projects is 'dismal', biz leaders complain

No wonder most orgs are slowing their spending

Um, what ever did happen with network automation?

Is it lingering somewhere between fusion power and self-driving cars?

Nvidia CEO brushes off Big Tech's attacks on NVLink network tech

Says UALink and Ultra Ethernet won't be threat for years as he reveals successor to Blackwell and free NIMs

AI smartphones must balance promise against hype and privacy concerns

Color us shocked: 66% of Apple users said they wouldn't switch for any reason

AMD's DC chief happy to work with Intel and others to chip away at Nvidia's AI empire

'If everybody's got their own little ecosystem, it's very inefficient'

Microsoft to spend $3.2B on expanding cloud and AI in green energy-rich Sweden

Budget to be blown on construction and 20K GPUs among other things in the next 2 years

AI chip sales predicted to jump by a third this year – then cool off

Gartner gives us a ray of hope amid ongoing hype and pressure to buy more hardware

Apple built custom servers and OS for its AI cloud

Mashup of iOS and macOS runs on homebrew silicon, with precious little for sysadmins to scry

DuckDuckGo AI Chat promises privacy for bot conversations

There's also an off switch

US senators propose guardrails for government AI purchases and operations

Bill proposes appointment of chief AI officers, privacy safeguards, and lots of testing

Italian premier taps Pope Francis to warn G7 of AI disaster if ethics ignored

Holy smokes, don't screw this up