Enhancing PyTorch Compatibility
for Nvidia-Optimized Chips
Google has begun collaborating with its artificial intelligence (AI) development rival, Meta, on AI chip software development. This move is seen as an effort to reduce its reliance on development environments optimized for Nvidia chips.
On December 17 (local time), Reuters reported, citing sources, that Google has started developing technology to optimize the open-source software (SW) 'PyTorch,' developed by Meta to run AI chips, for its own AI chip, the Tensor Processing Unit (TPU).
PyTorch is open-source software that assists in the development and execution of AI models. Used by AI developers worldwide, it is virtually a standard tool, but it is currently optimized for Nvidia chips. As a result, developers wishing to use chips other than Nvidia have had to learn new tools, which has been an obstacle for Google in expanding the TPU ecosystem.
To address this, Google has launched an internal project called 'TorchTPU' to ensure that PyTorch works seamlessly on TPUs. Through this project, developers can continue using the familiar PyTorch while switching only the hardware from Nvidia chips to Google chips.
Sources said that Meta, which develops and manages PyTorch, is also closely cooperating on this project. Through this collaboration, Google can increase its cloud service market share and AI chip sales, while Meta, which has declared its commitment to next-generation AI development, can save on infrastructure costs. The two companies, which compete in the online advertising market, are now cooperating to expand the AI chip ecosystem.
A Google Cloud spokesperson stated, "We are focused on providing developers with the flexibility and scalability they need, regardless of which hardware they choose."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
