To truly understand perf, ideally one should compare many types and sizes of models. I suspect some model types perform substantially better on the newer ANE /

To truly understand perf, ideally one should compare many types and sizes of models. I suspect some model types perform substantially better on the newer ANE / OS compared with others.