> The ACE v1 extensions introduced an outer product architecture with native support for modern AI datatypes. ACE and AVX10 provide complementary matrix and vector compute
capabilities. With ACE v2, this capability is further extended with a focus on Reduced-precision AI: New support for 4-bit formats and expanded 8-bit format
capabilities; Half-precision floating point: FP16; Single and double precision floating point: FP32 and FP64. The AVX10.3 extensions complement this focus, and provide further enhancements to general
purpose vectorization.
> The ACE v1 extensions introduced an outer product architecture with native support for modern AI datatypes. ACE and AVX10 provide complementary matrix and vector compute capabilities. With ACE v2, this capability is further extended with a focus on Reduced-precision AI: New support for 4-bit formats and expanded 8-bit format capabilities; Half-precision floating point: FP16; Single and double precision floating point: FP32 and FP64. The AVX10.3 extensions complement this focus, and provide further enhancements to general purpose vectorization.