4 ms·
This is an abstraction on top of CUDA. It takes care of writing the low-level code for operations that you would otherwise have to write yourself like matrix mu
by sjtrny 12y ago
This is an abstraction on top of CUDA. It takes care of writing the low-level code for operations that you would otherwise have to write yourself like matrix multiplication.