Amazon open-sources TorchNeuron in bid to chip away at Nvidia's software edge
A single setting lets PyTorch AI code run on Amazon's Trainium chips, the company says. The free software is released Tuesday and takes aim at Nvidia's CUDA platform.
Graphic: TERRA WORLD NEWS (own graphic based on the sources listed)
Amazon is trying to make it easier for developers to move from Nvidia's graphics processors to its own AI chips. Its new TorchNeuron software lets existing code for the widely used PyTorch AI framework run on Trainium2 and Trainium3 chips with a single configuration change and no rewriting, Amazon said. The software is being released Tuesday under the open-source Apache 2.0 license as the PyTorch Conference opens in San Jose, California.
Reuters described the move as an attempt to undercut the software advantage behind Nvidia's dominance: the chipmaker spent more than a decade building its CUDA platform, which helps PyTorch code run well on its chips. Amazon says 85% of AI researchers and developers write their code in PyTorch. Google is also courting developers for its own AI chips with better PyTorch tools.
More than 30 early customers in robotics, world models and voice AI are already using the software, Colin Brace, vice president of software at Amazon's Annapurna Labs chip unit, told Reuters. "This has been about meeting customers where they're at," he said. Named users include voice AI startup Cartesia and the Lieber Institute for Brain Development, while Hugging Face has added native Trainium support to its tools.