Question about L1 Cache in Memory Workload Analysis Table

Please check the box enclosed in the red circle.

I am not sure why the number of bytes is zero in this case. There are at least 26214400 global memory access requests (not bypassed). Could it be a problem with how Nsight Compute calculates the number of bytes?

Nsight Compute 2022.1.1 was used.

This is a bug in the table layout. The “Global Load” and “Global Load To Shared Store (access)” rows were accidentally merged for the “Bytes” column (but without actually accumulating their values, only the zero Global Loads value is shown). We will fix this in a future version. In the meantime, you can calculate the value that should have been in that cell for “Global Load To Shared Store (access) - Bytes” as

32 * sm__sass_l1tex_t_sectors_pipe_lsu_mem_global_op_ldgsts_cache_access.sum
1 Like

Thank you Felix!