Brief
Google releases open‑source TorchTPU to run PyTorch on its TPUs
The new project aims to reduce Nvidia's dominance in AI hardware by simplifying migration of PyTorch apps to Google’s chips.
By Felo News Desk · Published
On Tuesday, Google announced TorchTPU, an open‑source version of the popular PyTorch toolkit that enables AI applications to run on the company’s Tensor Processing Units (TPUs) and Google Cloud, TechCrunch reported.
What happened
Google made the TorchTPU code publicly available, allowing developers to move existing PyTorch models to TPUs with a simple configuration change and no code rewrite. The announcement follows Amazon’s release of TorchNeuron, which ports PyTorch to Amazon’s Trainium chips and AWS.
What the reports add
TechCrunch notes that PyTorch powers roughly 85% of AI applications and was originally created by Meta in 2016 before being transferred to the Linux Foundation in 2022. The article explains that Google worked directly with Meta, Hugging Face, Inferact, and RadixArk to ensure TorchTPU’s compatibility with existing AI development tools. It also highlights Google’s new TPU agent suite and an updated LiteRT tool that improves support for PyTorch models on phones, laptops and embedded devices.
What was said
Julie Bort, TechCrunch’s Startups/Venture Desk editor, wrote that “Google may be one‑upping Amazon in a few ways,” citing the company’s collaborations and tooling support.
How it came about
Both Google and Amazon are seeking to weaken Nvidia’s hardware advantage in the AI market, where most AI apps default to Nvidia GPUs. By providing open‑source pathways to alternative chips, the firms aim to broaden hardware choice for developers.
Key facts
- Google released TorchTPU, an open‑source version of PyTorch for TPUs. (techcrunch.com)
- TorchTPU lets developers migrate PyTorch models to Google Cloud TPUs without rewriting code. (techcrunch.com)
- Amazon announced a similar tool, TorchNeuron, for its Trainium chips. (techcrunch.com)
- PyTorch is used by about 85% of AI applications, according to the article. (techcrunch.com)
- Google collaborated with Meta, Hugging Face, Inferact and RadixArk to ensure compatibility. (techcrunch.com)
Sources
- [1] techcrunch.com — originally reported as “Google has joined Amazon in attacking Nvidia's stronghold with AI app developers”
