Like a kid that just got a new toy, Intel has cast aside its line of Nervana neural networking processors (NNP) in favor of the shiny new ones from recently acquired Habana Labs.

"After acquiring Habana Labs in December and with input from our customers, we are making strategic updates to the data center [artificial intelligence] acceleration roadmap," an Intel spokesperson wrote in an email to SDxCentral. "We will leverage our combined AI talent and technology to build leadership AI products."

Intel now plans to integrate the current and next generation of Habana's Goya and Gaudi AI processors with its other AI hardware and software.

"The Habana product line offers the strong, strategic advantage of a unified, highly-programmable architecture for both inference and training," the Intel spokesperson explained. "By moving to a single hardware architecture and software stack for data center AI acceleration, our engineering teams can join forces and focus on delivering more innovation faster to our customers."

Moor Insights & Strategy Senior Analyst Karl Freund broke the news Friday in an article published in Forbes, revealing Intel will end the development of its NNP AI chip designs and focus its efforts on developing Habana Lab's existing inference and AI machine learning chips.

Intel acquired the Israeli AI startup seven weeks ago for $2 billion in an effort to bolster its position in the highly competitive data center processor space. Nervana was also the result of a previous Intel acquisition in 2016.

At the time of the purchase, Habana Labs's portfolio included two AI chips. The first, its Goya AI Inference Processor, launched in late 2018 and has been used to power Facebook's Glow machine learning compiler. Habana's second chip, the Gaudi Training Processor, which was designed for machine learning training, hit the market less than six months prior to the acquisition.

The decision to drop Nervana comes barely a year after Intel launched the platform alongside partner Facebook and just two months after the chipmaker began shipping the first Nervana chips to customers. Intel and its competitors are seeking to develop a new class of chips designed for AI inference workloads, which is when a machine acts on a new data sample to infer an answer to a query.

Intel “can’t be happy with its Nervana efforts to date. Intel must get this right; I don’t believe it will get a third chance, but it is still early enough in the game to switch horses,” Freund wrote after the Habana Labs acquisition in December.

However, Freund now suggests that the decision to can Nervana in favor of Habana's technology was likely motivated in part by a failure to outperform competitor Nvidia's AI chips, as well as economic considerations associated with Habana's integrated 100-Gb/s Ethernet fabric.

Intel's Nervana platform, by comparison, required a proprietary interconnect for scaling.

"Since a Mellanox NIC with [remote direct memory access over converged Ethernet] can cost well over $1,000 per card, Intel will now have a chip that can scale at low cost to thousands of nodes to handle the emerging very large rural network models used in applications such as natural language processing," he wrote, adding that Nvidia's acquisition of Mellanox for $6.9 billion may also have influenced the decision.

It was a decision that Freund ultimately called a "smart and bold" move on the part of the chipmaker.