The vast majority of users use NVidia’s own kernels versus optimize their own. And those who do write custom kernels are typically not trying to compete with NVidia’s own GMM.