The specifications for 'ACE,' an x86 instruction set for high-speed execution of AI, have been announced.

The x86 Ecosystem Advisory Group , a joint venture between Intel and AMD to promote the x86 architecture, released the specifications for AI Compute Extensions (ACE), an x86 instruction set for AI processing, on June 15, 2026.
AI Compute Extensions (ACE) Specification – x86 Ecosystem Advisory Group
https://x86ecosystem.org/resource/ai-compute-extensions-ace-specification/
ACE: A Shared Path to Faster Matrix Math on x86 – x86 Ecosystem Advisory Group
https://x86ecosystem.org/blog/ace-a-shared-path-to-faster-matrix-math-on-x86-519/
Both Intel and AMD CPUs use the 'x86' architecture. The x86 Ecosystem Advisory Group is an advisory body comprised of Intel and AMD, as well as companies such as Broadcom, Dell, Google, Hewlett-Packard Enterprise, HP, Lenovo, Meta, Microsoft, Oracle, and Red Hat, and works to advance the architecture.
Intel and AMD establish 'x86 Ecosystem Advisory Group,' with major technology companies such as Google, Microsoft, and Meta also participating - GIGAZINE

The newly announced ACE is an extension instruction set designed to speed up AI execution. AI model inference requires the execution of calculations called 'matrix operations,' and ACE is an extension instruction set to speed up these matrix operations. The x86 Ecosystem Advisory Group explains that 'while traditional methods performed matrix operations in one dimension, ACE performs them in two dimensions.'
ACE is also characterized by its native support for computational precisions such as 'INT8,' 'BF16,' 'MXFP8,' and 'MXINT8.' Development of compilers, debuggers, and profilers that support ACE is already underway, and integration with machine learning frameworks such as PyTorch and TensorFlow is also planned.
The specifications for ACE are available at the following link.
AI Compute Extensions (ACE) Specification - ACE_v1_Specification_public_1_15.pdf
(PDF file) https://x86ecosystem.org/wp-content/uploads/2026/06/ACE_v1_Specification_public_1_15.pdf

Related Posts:






