In GTX20xx - int and float shaders began to work simultaneously, so these cards are faster x1.7 (instead x1.2 of classic generation up) than:
GTX10xx - float shader is always 1/2 or 1/4 lower than Tesla, so faster only +x0.5 instead +x1.0…
NVIDIA compiler use automaticly always all shaders… it’s perfect!
… how it look like with tensor shaders in TeslaV100+Volta… always auto use (and is simultaneously?) or is it different in this case?