Amazon is taking its biggest shot at Nvidia yet with new TorchNeuron dev tool
Amazon wants to break Nvidia’s hold on PyTorch, the free toolkit used to build most AI apps, so that more developers run their apps on alternative chips like its own.
TL;DR
- Amazon Web Services released TorchNeuron, a tool to move AI apps from PyTorch to its Trainium chip.
- TorchNeuron requires only a single line of configuration change, without code rewrites, for PyTorch apps.
- Approximately 85% of AI apps are built using PyTorch, which was originally developed by Facebook.
- Nvidia has established a strong presence by ensuring its CUDA software seamlessly supports PyTorch apps on its GPUs.
- TorchNeuron allows PyTorch developers to easily port apps to Trainium, which Amazon claims offers better price-performance and integrates training and inference.
- AWS argues that using Trainium on its cloud infrastructure provides a more complete solution than relying on specialized cloud providers tied to Nvidia.
- Amazon is also contributing code upstream to PyTorch to make the framework more hardware-agnostic, reducing its dependency on specific chips like Nvidia's.
- This initiative aims to give developers more choice and potentially more affordable options, challenging Nvidia's "lock-in" in the AI development ecosystem.