Comment by byzantinegene

i think the main consensus here is that the actual performance is not indicative of the benchmark performance (which supposedly outperforms the previous iterations)