Number of Active Blocks

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello.

If each block allocates 16kb of shared memory. Only one block will be alocated for each multiprocessor at time?

thanks.

2

Hello.

If each block allocates 16kb of shared memory. Only one block will be alocated for each multiprocessor at time?

thanks.

3

cc 1.x - Only one active block at a time, yes
cc 2.x - Possibly three due to 48 KB smem.

4

cc 1.x - Only one active block at a time, yes
cc 2.x - Possibly three due to 48 KB smem.

5

thanks Jimmy.

6

thanks Jimmy.

7

Actually, with compute capability 1.x it won’t execute at all. 16368 bytes is the largest amount of shared memory your kernel may use there if it has no arguments (less if it has).

8

Actually, with compute capability 1.x it won’t execute at all. 16368 bytes is the largest amount of shared memory your kernel may use there if it has no arguments (less if it has).

9

That’s true.

10

That’s true.

11

Exactly. I was aware of that.

12

Exactly. I was aware of that.