Organization of threads

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi. I try to make a program which will run in GPU GeForce 9500 GT.
My problem is on organizing of treads in blocks. If i knew some values of my device and cuda parameters , i could completed the program. The parameter which i know are:

Maximum number of threads per block: 512 (depend on CUDA)
Dimensions of grid: 65535 x 65535 x 1 (depend on CUDA)
Dimensions of blocks: 512 x 512 x 64 (depend on CUDA)
Number of multiprocessors: 4 (depend on device)

But i can not find the below values for my GPU:

  1. Maximum number of blocks which can be executed simultaneously on one multiprocessor.
  2. Maximum number of threads which can be executed simultaneously on one multiprocessor.

Where can i found these values for my GPU? I looked at my GPU specifications in site of CUDA but cant found something.

Also, i want to explain me somebody, what mean exactly “cuda cores”. Each multiprocessor has 8 cores? Each core execute a block each time?

Thanks

2

The CUDA Programming Guide lists these values in Appendix F.

CUDA cores work together to execute the instructions for an entire warp at once. In your GPU, there are 8 CUDA cores per multiprocessor, so the multiprocessor can complete one warp instruction in 4 clock cycles. Warps are executed in any order from the blocks assigned to that multiprocessor.