Cisco Programs on Tuesday launched networking chips for AI supercomputers that can compete with choices from Broadcom and Marvell Applied sciences.
Its silicon sequence chips are being examined by 5 of the six main cloud suppliers, Cisco mentioned, with out naming the businesses. Main cloud gamers embrace Amazon Internet ProvidersMicrosoft Azure and Google Cloud, collectively dominate the cloud computing market, in keeping with BofA International Analysis.
The rising reputation of AI functions reminiscent of ChatGPT, that are powered by a community of specialised chips known as graphics processing models (GPUs), has made the velocity at which these particular person chips talk extraordinarily vital.
Cisco is a serious provider of networking gear, together with Ethernet switches, which join units reminiscent of computer systems, laptops, routers, servers, and printers to native space networks.
It mentioned the most recent era of its Ethernet switches, known as the G200 and G202, have twice the efficiency of the earlier era, and may join as much as 32,000 GPUs concurrently.
“The G200 and G202 are going to be probably the most highly effective networking chips available on the market enabling probably the most power-efficient networks for AI/ML workloads.” Cisco mentioned Rakesh Chopra, a accomplice and former principal engineer.
Cisco mentioned the chips may also help carry out AI and machine studying duties with 40% fewer switches and fewer lag whereas being extra energy environment friendly.
In April, Broadcom introduced the Jericho3-AI chip that may join 32,000 GPU chips collectively.
