AMD AITER MI3XX/CDNA3 Kernels Patched for MI2XX/CDNA2
AMD's AITER project provides performance enhancements for MI2xx and MI3xx cards.
A user recently acquired MI210 cards and struggled to meet performance goals. AMD's AITER project offers optimized ASM kernels for CDNA3 and CDNA4 cards. Trials revealed that MI2xx cards matched MI3xx performance for INT8 W8A8. Following this patch, the user achieved up to 1.86x faster prefill and 1.72x faster decode speeds.