Jasper Yu
says to
YSITD
https://www.geekbench.com/doc/ml-0.5-inference-workloads.pdf