Meta launches AI software tools to make switching between Nvidia and AMD chips easier


Oct 3 (Reuters) – Facebook’s parent company Meta Platforms Inc (META.O) said on Monday it has launched a new set of free software tools for artificial intelligence apps that could make it easier for developers to switch between the various underlying chips.

Meta’s new open-source AI platform is based on an open-source machine learning framework called PyTorch, and can help code run up to 12x faster on Nvidia Corp’s flagship A100 chip (NVDA.O) or up to four times faster on Advanced Micro Devices Inc’s (AMD.O) MI250 chip, he said.

But just as important as the speed boost is the flexibility the software can provide, Meta said in a blog post.

Join now for FREE unlimited access to

Software has become a key battleground for chipmakers looking to create an ecosystem of developers to use their chips. Nvidia’s CUDA platform has been the most popular for AI work so far.

However, once developers adapt their code to Nvidia chips, it’s difficult to run it on graphics processing units, or GPUs, from Nvidia competitors like AMD. Meta said the software is designed to easily switch between chips without being locked.

“Unified GPU back-end support gives deep learning developers more hardware vendor choices with minimal migration costs,” Meta said in its blog post.

Nvidia and AMD did not immediately respond to requests for comment.

Meta’s software is designed for AI work called inference, which is when machine learning algorithms that have been previously trained on huge amounts of data are called upon to make quick judgements, like decide whether a photo is of a cat or a dog.

“This is a cross-platform software effort. And it speaks to the importance of software, especially for deploying neural networks in machine learning for inference,” said David Kanter, founder of MLCommons. , an independent group that measures the speed of AI. .

Kanter added that this new Meta AI platform would be “good for customer choice.”

Join now for FREE unlimited access to

Reporting by Jane Lanhee Lee and Stephen Nellis; edited by Jonathan Oatis

Our standards: The Thomson Reuters Trust Principles.


Comments are closed.