GeForce 1660 Ti Mobile GPU Poor Performance with Mixed Precision

Hi everyone.
I’m using a 1660Ti mobile gpu with tensorflow for machine learning and I am trying to speed up my training with mixed precision. 1660 Ti’s compute power is listed as 7.5 since it is Turing based and it is said that it will gain significant performance using mixed precision. But instead when I use mixed precision training slows down by 2-3 times. I suspect the absence of tensor cores in this card might be the reason. But still even there is the absence of tensor cores it shouldn’t be slower than 10*0 series. Since this card is listed with the same compute capacity of the RTX 2000 series could there be a bug of some sort. Is there any solutions to solve this problem. Thanks.

1 Like