Why do CUDA exes not work correctly on another computer?

Question

GPGPU

I am writing a program in VS2010 on a computer with GeForce 8400gs. It counts correctly on it with a large number of blocks and threads (I check with the result of the CPU).
Further, I transfer the exe c dll to other computers with GTX 650 and 8800 video cards, but there it gives an incorrect result with a large number of threads and blocks. At the same time, with one thread and one block, it counts correctly.

Answer 1

A classic manifestation of the thread race