(CUDA) People with NVIDIA GPUs: Can you run my benchmark?

[Jamie]

I'm interested whether different gpus give different results when copying from host to device memory, if some of you could compile and run the following program it would be handy!

/*
*	copyTest.cu
*
*	Time a copy of how ever many megabytes to the device
*/

#include <stdio.h>
#define MB 32 //Number of megabytes to transfer

int main (void) {

	int size = (MB*1024*1024) / sizeof(int); //number of cells needed for MB
	int *a = (int*) malloc(size * sizeof *a);
	int *dev_a;
	
	printf("Copy Operation Test\n");

	//Fill array with garbage
	for (int i=0; i<size; i++) {
		a[i] = i+i;
	}
	
	printf("Int size: %i bytes, size of array: %i MB\n", sizeof(int), MB);

	cudaEvent_t start, stop;  
	float elapsedTime; 
	cudaEventCreate(&start); 
	cudaEventCreate(&stop); 
	//event start 
	cudaEventRecord(start,0); 

	//Allocate
	cudaMalloc((void**) &dev_a, size * sizeof(int));
	//Copy FROM, TO, NUMBER OF BYTES, DIRECTION
	cudaMemcpy(dev_a, a, size * sizeof(int), cudaMemcpyHostToDevice);

	//event stop 
	cudaEventRecord(stop, 0); 
	cudaEventSynchronize(stop); 
	cudaEventElapsedTime(&elapsedTime, start, stop); 
	 
	printf("Time: %3.3f ms\n", elapsedTime); 
	
	//Tidy up
             free(a);
	cudaFree(dev_a);
	cudaEventDestroy(start); 
	cudaEventDestroy(stop); 
	
	printf("Done!\n"); 

	while( getchar() != '\n' ); 

	return 0;
}

gpuz:

Results over 3 runs (milliseconds):
#1 14.420
#2 14.080
#3 14.713

average: 14.404333