At the second stop of Arm Tech Symposia 2023 in Hsinchu, Arm announced its collaboration with industry players including Intel, AMD, NVIDIA, Qualcomm, Microsoft, and Meta to promote the development of artificial intelligence technology applications through various projects, enabling more than 1500 million developers worldwide using Arm architecture to create various artificial intelligence application services through appropriate basic frameworks, technologies, and specifications.
Arm said that currently up to 70% of third-party artificial intelligence applications on mobile phones are run on Arm architecture processors. The announcement of cooperation with many industries will enable more artificial intelligence models to be executed on edge computing devices and provide developers with more technical resources.
The collaboration with NVIDIA has enabled the adaptation of the TAO toolkit to run low-code on the Ethos-U NPU, helping to create performance-optimized visual AI models. Development can also be performed in the open-source TensorFlow and PyTorch frameworks. The collaboration with Meta is also focused on bringing PyTorch to Arm-based mobile and embedded platforms. Based on the Tensor Operator Set Architecture (TOSA), in which Arm has invested heavily, developers can use ExecuTorch to deploy the neural networks required for advanced AI or machine learning workloads on various mobile and edge devices.
The new 8-bit floating-point specification, FP8, proposed in a collaborative project between Arm, Intel, and NVIDIA last year has now been joined by AMD, Google, and Meta, who have jointly created the official OCP 8-bit floating-point specification (OFP8). Arm has now added this standard-compliant OFP8 floating-point specification to its latest Cortex-A series architecture, promoting the development of artificial intelligence computing based on neural networks.
In addition, Arm recently joined the MX Alliance, a group of industry players including AMD, Intel, Meta, Microsoft, NVIDIA, and Qualcomm, to standardize narrow-bit data specifications and accelerate the scaling of artificial intelligence.
Arm emphasized that it has laid a solid foundation for the development of global artificial intelligence, and it is expected that more artificial intelligence computing developments will be built on Arm infrastructure, from sensors and smartphones to software-defined vehicles, servers and supercomputers.
