In October 2024, AMD and Intel came together to form the x86 Ecosystem Advisory Group (EAG), with a shared vision of propelling the evolution of the x86 computing architecture. Shortly after its establishment, the EAG unveiled four pivotal features: FRED, AVX10, ChkTag, and ACE. On April 29, 2026, a significant milestone was achieved as AMD and Intel jointly published the ACE white paper, formally introducing the 'x86 Standard Matrix Acceleration Architecture' instruction set to the developer community at large.
The primary goal of ACE is to substantially boost the matrix multiplication capabilities of x86 chips. It accomplishes this by incorporating a matrix acceleration mechanism rooted in outer product operations, which delivers a computational density 16 times greater than that of equivalent AVX10 multiply-accumulate operations, all while utilizing the same input vectors. ACE inherently supports the prevalent precision standards in today's AI landscape, encompassing INT8, OCP FP8, BF16, and more.
As an extension of the AVX10 instruction set, the software ecosystem is already adapting to ACE, with leading machine learning frameworks embarking on integration initiatives. AMD and Intel underscore that ACE is crafted with a philosophy of minimal disruption and broad applicability, empowering developers to sidestep the need for code rewrites across diverse hardware platforms.