matrix_mul with max_size

Accelerated Computing CUDA CUDA Programming and Performance
1

hi.

now i am experimenting max size about matrix_multiple example.

my experimental result, problem size larger than 3000X3000, program didn’t work.

but i think i am using global memory just 108Mb.

(
3000X3000 X4 X3 =108000000byte.
(matrix_size) (floating size) (3element, MATRIX M, MATRIX N, MATRIX P)

108000000byte=108Mb
)

i think the program must work much large size more than 3000X3000(almost 6000X6000)

what is the reason???
thank you very much~!!!

2

Actually, what is your aim. you are using Matrix multiplication program in the SDK. right. If this is giving wrong result in 3000x3000 size, then how it gets correct result in a bigger size than 3000x3000.
First you have to check how much memory is in your card. Next thing is what is actual CPU memory size allocated for this program. (this can be checked from task manager). If allocation is 108MB (around) , then you have to check the logic in your code.