Intel and AMD jointly released a whitepaper on April 29, 2024, introducing a new standard called AI Compute Extensions. This initiative aims to establish a unified matrix acceleration solution for the x86 architecture. The two companies collaborated to define the technical specifications that will guide future hardware development.
The ACE specification integrates seamlessly with the existing AVX10 instruction set. It supports INT8, OCP FP8, and BF16 data formats, which are common in artificial intelligence workloads. The supported data formats enable efficient matrix operations through seamless integration with AVX10.
Joint specification targets unified matrix acceleration for x86 architecture
Performance claims in the whitepaper indicate a significant improvement in computational density. The document states that ACE provides 16 times higher compute density compared to equivalent AVX10 multiply-accumulate operations for the same number of input vectors. This metric highlights the potential efficiency gains for matrix multiplication tasks.
AI Compute Extensions (ACE) Specifications
| Feature | Detail |
|---|---|
| Compute Density | 16x improvement over AVX10 |
| Data Formats | INT8, OCP FP8, BF16 |
| Integration | AVX10 |
The whitepaper release marks the beginning of the standardization process rather than the immediate availability of products. The specific timeline for hardware implementation or adoption of ACE remains unconfirmed. The whitepaper does not specify a timeline for hardware implementation or adoption of ACE.
Discussion
0 comments
Log in to join the thread with a thoughtful take, question, or correction.