x86 AI Compute Extensions (Ace) Specification [pdf] (x86ecosystem.org)

The recently released x86 AI Compute Extensions (ACE) Specification marks a significant step forward in enhancing AI performance within the x86 architecture, developed collaboratively by Intel and AMD. This document outlines new instructions and features aimed at optimizing AI workloads, such as matrix multiplication and outer product operations, with particular focus on enhanced data formats and conversion functions. The ACE introduces specific assembler mnemonic names and tile management instructions that allow for more efficient data processing and memory handling in AI applications. The significance of the ACE Specification lies in its potential to standardize AI-related compute capabilities, thereby facilitating developers in leveraging advanced AI/ML functionalities on x86 platforms. By providing tools like AVX-VNNI-INT8 and new tile instructions, the specification enables more effective execution of neural network operations, particularly in handling smaller, mixed-precision data types like FP8 and INT8. This initiative could accelerate innovation within the AI/ML community, enabling improved performance and efficiency in developing AI solutions across various applications.