Trying to use CUDA, I have observed some delays in cuInit performing.
By the deep exploration, I have found that the reason is in very slow opening
of nvidia device files.
For example, the operator
int fd = open( “/dev/nvidia0”, O_RDWR);
is performing in 1,5 seconds for Device 0: “Tesla M2050”:
$ ./openTest open time = 1.5190 secs.
The same thing is occurred for Device 0: “Tesla C1060”,
but the time is slightly less:
$ ./openTest open time = 0.8532 secs.
My question is what it is and how to fix it ?
I am using Linux CentOS release 5.4 (Final) with