Meta releases a new AI platform that can switch freely between Nvidia and AMD chips

Facebook parent company Meta announced the launch of a new artificial intelligence free software platform that supports both NVIDIA and AMD chips, making it easier for developers to develop artificial intelligence programs between hardware systems based on different chips. Switch back and forth.

Meta's newly released set of artificial intelligence open source software is built on the basis of the PyTorch open source machine learning framework, which can make the code on Nvidia's flagship A100 chip run 12 times faster, and also make the code on AMD's MI250 chip run faster. 4x increase.

Meta said in a blog post that the AI ​​software platform not only speeds up code execution but also supports AI chips from different manufacturers.

At present, software development has become a key area for chip manufacturers to build developer ecosystems and use their own chips. For example, CUDA developed by NVIDIA is very popular.

However, after developers develop artificial intelligence code based on Nvidia chips through CUDA, it is difficult to run on graphics and image processing chips made by companies such as Nvidia's rival AMD. Meta said its latest artificial intelligence platform allows developers to easily switch their AI applications between hardware systems based on different chips.

"Unified GPU backend support minimizes application migration costs and helps deep learning application developers choose a wider variety of hardware," Meta said in a blog post.

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

It is reported that the software launched by Meta is specially developed for a type of artificial intelligence work called inference. During the inference process, developers make quick judgments on images by invoking machine learning algorithms that have been trained on a large amount of data before.

"This is cross-platform software work that demonstrates the importance of software, especially the deployment of neural networks for inference in machine learning," said David Kanter, founder of MLCommons, a third-party organization that measures the processing speed of AI. David Kanter) said.

The new AI platform will "give customers more choice," Kantor added.