On June 20, 2026, Tom's Hardware reported that Intel and AMD have jointly released the full specification for the ACE...

On June 20, 2026, Tom's Hardware reported that Intel and AMD have jointly released the full specification for the ACE (Advanced Compute Extensions) CPU instruction set, designed to accelerate AI workloads on x86 processors. The new extensions leverage existing AVX10 registers but add dedicated silicon for matrix multiplication, enabling up to 16x more operations per instruction compared to AVX10 (Tom's Hardware, also via Reddit). This improvement significantly boosts power efficiency and density for matrix math, the core operation in neural networks. ACE supports a wide range of data types, including INT8, FP8, BF16, and Open Compute Project's MX block-scaled formats. The extensions are implementation-agnostic, allowing machine learning frameworks like PyTorch and TensorFlow to use a single code path across all compatible x86 CPUs. For AMD, this strengthens the value proposition of its EPYC and Ryzen processors for AI inference tasks, especially in latency-sensitive or GPU-less environments. While the announcement is a joint effort with Intel, it positions AMD to better compete in the growing AI inference market on CPU platforms. The low confidence reflects reliance on a single news report, but the technical details and credible source make it a noteworthy positive signal for AMD's AI infrastructure roadmap.