your123 -

AMD tries to differentiate in AI market with acquisition

The acquisition of the open source startup helps the chip vendor compete in the optimization market and offer flexibility compared to competitors such as Nvidia.

AMD will acquire startup and its AI software optimization technology.

The chip vendor revealed on Oct. 10 that it has signed a definitive agreement to continue the development of what AMD called Nod's "open AI software capabilities."

Based in Santa Clara, Calif., California-based was founded in 2013 as a developer of machine learning systems and open source AI technologies.

Nod is mainly known for its SHARK software, which reduces manual optimization and the time to deploy AI models.

According to Nod, SHARK enables AI models to run faster than PyTorch, a popular framework for building neural networks, and Torchscript, an open source technology that optimizes neural network inference speed.

AI optimization

By acquiring Nod, AMD is signaling it wants to be player in the AI optimization arena, said Jack Gold, an analyst at J. Gold Associates.

Because Nod has a decade of experience in optimizing AI workloads, the acquisition creates a significant opportunity for AMD to run AI workloads faster and compete with AI chipmaker and software giant Nvidia, Gold said.

"It's trying to make sure they have an optimized environment from the software perspective that will run on their chips," he said.

An opportunity for flexibility

The agreement to acquire Nod comes about two weeks after AMD revealed a partnership with large language model provider Lamini to run its high-performance large language models on AMD's systems.

The two acquisitions reflect the strategic path that AMD is now trying to take by expanding into a full-fledged AI hardware and software provider, said Daniel Newman, an analyst with Futurum Group.

"It's very much like a plug in to a bigger ecosystem play," Newman said.

As an open source developer, Nod will help AMD stand out in a market dominated by Nvidia, with its CUDA parallel computing platform and programming model.

While many developers use CUDA, they are sometimes locked into CUDA libraries and are unable to use other libraries outside of that.

Meanwhile, Nvidia has a set of other closed architectures and frameworks and has models that are customized to run on only Nvidia, Newman noted.

AMD woos AI business

That creates an opportunity for AMD and veteran chipmaker Intel to position themselves as open and offer users more flexibility in moving between hardware and using more software languages to develop their AI capabilities.

"That's where AMD is leaning in and saying, 'Hey, work with us. We're friendly; we're open,'" Newman said. He added that this flexibility could help AMD make significant inroads in AI hardware.

It's very much like a plug in to a bigger ecosystem play.
Daniel NewmanAnalyst, Futurum Group

Going with AMD could also be a more inexpensive option for enterprises that sometimes spend millions of dollars training AI models, Gold said.

Moreover, the highest-end chips, like the Nvidia H100, may not be necessary for some workloads.

"It's not just about having the fastest silicon. It's about having a [product] that is optimized for particular workload around cost of ownership," Gold said.

While AMD can go after some parts of the AI hardware market with its powerful GPU configurations, the vendor still need some help at the developer level -- an area in which Nvidia has been successful, Newman continued.

"Something like Nod gives them the opportunity to support AI customers with open source software that enables them to successfully deploy their AI projects and do it even better on AMD hardware," he said.

AMD did not disclose financial terms of the acquisition.

Esther Ajao is a TechTarget Editorial news writer covering artificial intelligence software and systems.

Dig Deeper on AI technologies

Business Analytics
Data Management