In my environment, everything from R2020b to R2016a had the same error with no gpu available.
The GPU is Tesla K80 * 2 and the OS is Ubuntu 18,04.
When I checked, nvidia-smi was OK, but even python tensorflow could not call the GPU.
The reason was that my predecessor had left behind a blacklist.conf configuration file with a fixed driver version.
So, all drivers could not be loaded by modprove.
Therefore, I think that the dependencies were broken with the version upgrade.