ParentFull threaddarknoon·you can access the tensor cores from cuda, in practice you might generate the code with something like openai tritonView on HN