NVIDIA RTX 3060 Ti : Half-precision floating-point support

Hello all,

I use Ampere-generation NVIDIA GPU and get errors while using halfn elements. Also clGetDeviceInfo with CL_DEVICE_PREFERRED_VECTOR_WIDTH_HALF returns 0.

But since the Pascal generation, NVIDIA supports fp16 (half-precision floating-point operations), why am I getting this result?

Thanks,
Harish