Intel has announced the cessation of development for its Nervana AI processors, shifting its focus to advancing processors based on technology from Habana Labs, a company it acquired in December for $2 billion. The decision marks Intel’s strategic move to bolster its position in the competitive AI processor market, where it faces rivals such as Nvidia, AMD, and numerous startups.
While Intel plans to fulfill existing customer commitments for the Intel NNP-I inference accelerator, it will discontinue the development of the Intel NNP-T, a processor designed for AI training. This decision reflects Intel’s commitment to making rapid strides in the AI processor domain, as it acknowledges being a few years behind its competitors.
Intel AI hardware
The NNP-I, code-named “Spring Hill,” was introduced in January 2019 at the Consumer Electronics Show. It is a 10 nm chip specifically designed to accelerate inference for organizations dealing with AI workloads. Facebook was an early adopter of this chip, collaborating with Intel on its development.
On the other hand, the NNP-T, tailored for AI training, had not yet been released. Intel initially acquired Nervana Systems in 2016 with the aim of expanding its hardware and software AI portfolio.
The shift in Intel’s AI hardware roadmap raises questions about the future of customers running workloads on technology without a clear roadmap. However, industry experts, including Peter Rutten, research director at IDC, believe that Intel will prioritize supporting existing customers and gradually transition them to the Habana processors when the time for renewal arrives.
Habana processors
Habana Labs, now at the forefront of Intel’s AI hardware strategy, has developed two powerful processors: Goya, a fast and energy-efficient inference processor, and Gaudi, a processor designed for AI training. These processors are equipped with Tensor cores, providing a performance edge over the Nervana chips.
Intel aims to swiftly bring these advanced processors to the market, leveraging the acquisition of Habana Labs to accelerate its competitive position. The company states that it will enhance current and future generations of Goya and Gaudi with Intel AI hardware and software innovation. The unified, highly programmable architecture of the Habana product line is positioned as a strategic advantage for both inference and training, enabling Intel’s engineering teams to collaborate and deliver innovations more rapidly to customers in the data center AI acceleration space.
