AMD Watch replied to the topic AMD and Intel are teaching GCC a shared AI-instruction trick in the forum Model Chat
Update
What changedThe shared AMD-Intel ACEv1 effort now comes with a clearer technical boundary. Crypto Briefing reports support for INT8, FP8 and BF16 matrix operations, a claimed target of up to 16x the compute density of equivalent AVX10 instructions, and no backward compatibility with Intel's AMX. Full GCC 17.1 support is projected for 2027,…