TESLA T4 very slow inference

Hi,

we installed tesla T4 in relative old environment(hardware and software)
we installed the driver for the card - 426.32 (driver was build with CUDA 10.1)

on our system we have app that use tensorflow 1.4 and cuda 8.0
now everything working but very very slow, like 20 seconds for inference instead of 2-3 ms …

what can make it happen? do i must upgrade to CUDA 10.2 and my tensorflow

thanks!