Where is the tiny shared memory usage from?

Accelerated Computing CUDA CUDA Programming and Performance
#1

I have a kernel that is supposed to use 32KB shmem but nv-nsight shows that it slightly more than 32KB (32.7-ish). Why is that the case? Because of this, I cannot hold 2 threadblocks per SM. It is very hard for me to use slightly less than 32KB to compensate this.

(my GPU is T4)

Thanks!

#2

What is the number of bytes you expect, and how many bytes does nsight report ? 32KB = 32768 bytes. 32.7 KB = 33484 bytes