cuDNN error

Accelerated Computing GPU-Accelerated Libraries
#1

Hi,

I am new to cuda and decided to dive right in with cuDNN as i require it for my master project. I am simply trying to run the following cuDNN program for testing:

#include <iostream>
#include "cudnn.h"
using namespace std;
int main(int argc, char const *argv[])
{
	cout << "hello cuda" << endl;
	cudnnStatus_t status;
	cudnnHandle_t handle;
	status = cudnnCreate(&handle);
	cout << "status " << status << endl; 
	cudnnDestroy(handle);
	return 0;
}

And i compile the program using the following command:

nvcc -I ./include -I /usr/local/cuda-6.5/include -L ./lib ./main.cpp -L /usr/local/cuda-6.5/lib64/ -lcudnn -lcuda -lcudart -o prog

The lib folder contains the cuDNN lib’s and the include folder contains cudnn.h.

When running the compiled code i get:

hello cuda
status 6

I am operating on a remote university machine with a working cuda 6.5 toolkit, however i do not know which NVIDIA card is in the machine.

Might anyone know why i get the 6 status signal failure?

#2

Have you looked inside cudnn.h for a list/description of the error codes?

#3

Ai, i assumed the order of the error descriptions were the same as in the library manual. But now that you explicitly told me i see it returns CUDNN_STATUS_ARCH_MISMATCH. Turn’s out compute capabilities 3.0 are required which this graphics card does not posses. Thank you for pointing this out to me.