system crash when running cuda programm


I’m junior on cuda programming. These days, when I run my cuda code, the system frequently crash and I need to reboot. For example, if I reference the memory on host (cpu) in the kernel, it certainly crash (it is a junior stupid mistake). And I think there are many other issues whish call the system crash. Can anyone state about what opeations could cause the crash? So I can avoid the situation in the future. Thanks.

my gpu is gtx 280
the code is running on windows 2003 server
the computer is a Dell Precisionâ„¢ 470 workstation which has
cpu - dual 2.80 GHz Intel Xeon
memory - 4030MB
I use the machine by vnc

anyone has some idea?

Define _DEBUG and use CUDA_SAFE_CALL everywhere.

Thanks a lot. But for the kernels designed by ourselves, how can CUDA_SAFE_CALL be used. It is used for the cuda’s own functions.

don’t use CUDA_SAFE_CALL anywhere, check errors yourself. cutil should not be used by anyone except SDK authors for any reason. it’s not robust enough and often doesn’t do the thing that you may want.

if you find some way to consistently crash the machine, let us know what driver version you’re using as well as posting full source of a repro case.

Hi, Thanks. Maybe there’s something incompatible between cuda and windows 2003 server. Now I move to winxp and the crash problem has gone.