Hardware capability problem

I have a CUDA program build on my computer(windows 10,1060,visual studio 2015,pgi 18.4),and it can run successfully,but when i copy the exe file and run on another computer(windows 10,TITAN V),it failed,even i set -Mcuda=cc70 on compiler.
Is there any way to fix?

Hi HuangXiaoHui24283,

Can you please describe the failure?

Did you dynamically link the application? If so, did you also copy over the PGI runtime DLLs with the application?

Does the other computer have the Microsoft OpenTools redistributable packages installed?

Please see the following for considerations when deploying applications on Windows: https://www.pgroup.com/resources/docs/19.1/x86/pgi-user-guide/index.htm#deploy-app-win


Hi Mat,
I do copy over the runtime DLLs with the application,the application can run the CPU part,and then throw out an error like this:
0:allocate:copyin symbol memcpy failed:13(invalid device symbol)
by the way the application can run on a computer with windows 10,1050.So i suppose maybe it’s because different architecture,but even i set -Mcuda=ccall it’s still the same error.
Is it possible the other computer didn’t install the Microsoft Open Tools makes so?
Thank you for your help.

Thanks. This is a runtime error in your program and has nothing to do with MS Open Tools nor missing PGI DLLs. If the program was missing the DLLs, it wouldn’t run.

As for the actual error, unfortunately, there’s not enough information to give you good advice.

It’s possible that it’s a system issue, in which case I’d want to know:

  • If you’re running out of device memory
    What the output from “pgaccelinfo” utility is
    What CUDA driver you have installed
    What CUDA version you compiled with
    If you’re able to run a simple CUDA program

While less likely, it could also be a problem with your program. Understood that it works on another system, but sometimes bug are masked on one system can show-up on another.