Meta AI has developed and is open-sourcing AITemplate (AIT), a unified inference system with separate acceleration back-ends for both AMD and NVIDIA GPU hardware.
AITemplate delivers close to hardware-native Tensor Core (NVIDIA GPU) and Matrix Core (AMD GPU) performance on various widely used AI models such as convolutional neural networks, transformers, and diffusers. With AIT, it is now possible to run performant inference on hardware from both GPU providers.
Meta said its open-source AI platform is based on an open-source machine learning framework called PyTorch. According to Meta, AITemplate can help code run up to 12 times faster on Nvidia’s A100 chip or up to 4 times faster on Advanced Micro Devices (AMD) MI250 chip.
The software has become a battleground for chipmakers looking to build up an ecosystem of developers that utilize their chips. Nvidia’s CUDA platform continues to be the most popular so far for AI work.
Once developers tailor their code for Nvidia chips, running it on graphics processing units or GPUs from Nvidia competitors like AMD is not easy. AITemplate is designed to quickly swap between chips without being locked in, Meta said in a blog post.
Meta said that the GPU back-end support gives deep learning developers more hardware vendor choices with minimal migration costs. The company added that proprietary software toolkits such as TensorRT provide ways of customization, but they are often insufficient to satisfy software developers’ needs. AITemplate can help with that.