CUDA on different graphic cards

Hello. I created library that do some calculations using CUDA. I implemented it on computer that have GTX 1050Ti and installed CUDA 10.2. Everything works fine. But when I tested this library on computer that have GTX 650Ti or GTX 770, I don’t receive expected result. These computers also have installed same CUDA version and updated drivers. I also checked functions that I use, in order to know if computing capability corresponding to GPU. Seems like it’s not problem of capability. Could you please suggest me where the problem can be and what should I check? Thank you.

I don’t receive expected result

This doesn’t really tell us anything about the nature of the issue, in particular since we also don’t know anything about the code you are running. The “unexpected results” could be anything from minor numerical discrepancies to NaNs all over the place, or occasional abnormal termination of the application.

Have you run this application under control of cuda-memcheck on the failing system(s)? If so, did it report any issues? You would want to address all issues it reports.