I want to know when does the Turing GPU use Tensor core.
And, can I decide whether to use it or not?
The language features to use tensor cores from CUDA code are documented here
So,
I’m using the Quadro RTX 6000 and want to calculate Turing’s occupancy.
Is it possible to calculate occupancy( active warps / maximum number of warps per SM) through CUDA occupancy calculator in Turing GPU?
And, Where can I get details of T uring GPU like maximum number of warps per SM?
Please explain to me.
Yes, you can use the occupancy calculator with a Turing GPU in the same way as any other CUDA GPU.
The documentation lists maximum warps per SM and other data for all GPU types in Table 15 of the programming guide.