[ju-st]

The newest Xeons have a higher performance when you use AVX2 instead of AVX512 because of the low clock when using avx512 instructions (source in German: https://www.heise.de/ct/ausgabe/2017-20-Von-Suenden-Fehlern-... )

[slededit]

The first gen is usually slower and a prototype so software can get written. AVX2's scatter gather instructions were similarly slower initially.