Comment by catlifeonmars

I guess it depends on your application, but 100ms is pretty slow. Plus there is the model footprint (memory + cpu/gpu).