I have been profiling a ML workload and have been coming across kernels named nvjet_hsh_128x128_64x6_1x1_h_bz_splitK_NTN
and nvjet_hsh_192x192_64x3_1x2_h_bz_coopB_NTN
and so on. This is a new kind of kernel names I am seeing. May I know the purpose of nvjet
kernels?
I believe those are Jetson specific kernels.
I think you will get a better answer in the Accelerated Computing > GPU-Accelerated Libraries forum.
1 Like