3 ms·

So is this enough to add AMD support to llama.cpp? I would like to know if HIP is faster than OpenCl.