AMD and Intel are collaborating on a new instruction set extension called AI Compute Extensions, or ACE. The x86 Ecosystem Advisory Group released a whitepaper detailing the joint effort to standardize matrix acceleration for the x86 ecosystem. This initiative aims to provide developers with a unified approach to handling AI workloads across different hardware platforms.
ACE functions as an extension of the existing AVX10 instruction set. It supports several key data formats including INT8, OCP FP8, OCP MXFP8, OCP MXINT8, and BF16. These formats allow for efficient processing of the numerical data types commonly used in modern artificial intelligence applications.
Joint whitepaper details ACE instruction set extension
The extension delivers up to 16 times higher operation density compared to standard AVX10 multiply-accumulate operations when using the same input vector count. According to the whitepaper, this architecture significantly improves matrix multiplication performance, scalability, and energy efficiency. The increased density enables more complex calculations within the same computational footprint.
AMD and Intel continue to cooperate on the ACE and AVX10 roadmap to capture new opportunities in the AI and general workload sectors. The x86 Ecosystem Advisory Group stated that the broad market and high performance of x86 offer developers an exceptional choice. Integrating the ACE instruction set will further strengthen the x86 ecosystem's position in the industry.
Discussion
0 comments
Log in to join the thread with a thoughtful take, question, or correction.