AI BriefingIntel and AMD unveil ACE for x86
·2026.05.04 01:52
Key point
Intel and AMD released the ACE white paper and a standard matrix acceleration direction for x86.
1 / 2
Details
Intel and AMD, through the x86 Ecosystem Advisory Group, unveiled the AI Compute Extensions (ACE) white paper, presenting a standard matrix acceleration extension for x86 CPUs.
- ACE is an outer-product-based structure that combines with AVX10, processing 1024 multiplications instead of the baseline 64 for INT8, claiming a 16x increase in compute density.
- It uses 8 2D tile registers and 32-bit accumulation, supporting INT8, BF16, OCP FP8, OCP MXFP8, and OCP MXINT8.
- On the software side, compiler, debugger, and profiler integration is underway, with support for NumPy, SciPy, PyTorch, and TensorFlow also announced.
- No ACE-supporting CPUs have been disclosed yet, but the direction aims to expand AI workloads across the same x86 stack from laptops to data centers.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
AI Compute Extensions (ACE) SpecificationBreaking the Limits of Ternary LLM Weight Compression: Achieving 1.485 bits/weight with BITCOS
AI Bubble Whiff Effect: Cascading Bottlenecks from GPUs to Data CentersGoogle Collaborating with AMD on Next-Gen TPU Design — Potential Hybrid AI ASIC with On-Package CPU Cores for Reinforcement Learning