Google Releases TorchTPU to Loosen Nvidia's Grip on AI Development

Google unveiled TorchTPU on Tuesday, an open-source version of the PyTorch toolkit that lets AI applications run on the company's Tensor Processing Units, or TPUs. PyTorch, originally built by Meta in 2016 and handed to the Linux Foundation in 2022, is used by some accounts in 85% of AI apps. Because Nvidia's GPUs were already standard for AI research when PyTorch was created, many of its features were tailored to that hardware.

TorchTPU allows existing PyTorch apps to move to TPUs and Google Cloud through a simple configuration change, with no rewriting required. Google said it worked directly with Meta, Hugging Face, Inferact, RadixArk and others to ensure compatibility with popular AI development tools. It also launched a TPU agent suite to help developers migrate apps and fix errors, and updated its LiteRT tool to work better with PyTorch apps.

The move follows Amazon's Monday release of TorchNeuron, which ports AI apps to its Trainium chip and AWS without rewriting. Amazon also said it is submitting code to the main PyTorch project that decouples features from Nvidia's chips, making the toolkit more hardware-agnostic by default.

Neither development guarantees developers will abandon Nvidia's hardware en masse, but both companies are plainly working to erode its advantage.

Source: TechCrunch