CUDA block sizes

Question

I have the computing ability of NVIDIA GTX 570 2.0 running cuda-4.0.

The deviceQuery executable in the CUDA SDK gives me information about my CUDA device and its various properties. Two lines of output:

Maximum number of threads per block: 1024

Maximum dimensions of each block dimension: 1024 x 1024 x 64

Why is the 3rd block dimension limited to only 64 threads, only if the size of X and Y can vary up to 1024 threads?

Answer 1

EDIT2: , , ; . , , 64 . , , , .

, : -, , ; -, 32, - 2 32; -, , , , ( , X Y, Z).

CUDA, , 1D-, , . , X 1024 .

, , CUDA 512 X ( , X , ) 32 Y. 1024 X, , , X , Y, Y 1024. , , 32 Y.

3D-, X, Y >= Z 1024, , X = Y = Z = 10; , Z 10,

, , (1024, 32, 10). - (1024, 1024, 64)? , , - X >= Y >= Z.

: , : .

Answer 2

, threadIdx.x, threadIdx.y threadIdx.z 32- , , . , warp id? , , , , ?

, , , .