CUDA program crashing video card

Accelerated Computing CUDA CUDA Programming and Performance
1

I am very new to CUDA. I am trying to write a program that draws a mandelbrot fractal. When I set the resolution of the image to 2250x2250 pixels, my monitor goes black and when it turns back on I get a little message telling me that the video card has crashed. When I set the resolution to 2000x2000 pixels this does not happen.

int main(void)

{

        //Set various parameters of the image

	float lowx = -2, lowy = -2, highx = 2, highy = 2; //bounds on image location in xy-plane

	unsigned int xres = 2000, yres = 2000;            //resolution of image

//This line comes from the Julia set example in Ch. 4 of CUDA by example by Sanders and Kandrot

	dim3 grid(yres,xres);

	

        //Allocate memory

	char *bmp = (char *)calloc(xres*yres,sizeof(char));

	char *bmpDev;

	cudaMalloc((void**)&bmpDev, xres*yres*sizeof(char));

//Call drawing function

	evalPoint<<<grid,1>>>(bmpDev,highx,highy,lowx,lowy);

//Save image

	cudaMemcpy(bmp,bmpDev,xres*yres*sizeof(char),cudaMemcpyDeviceToHost);

	printBMPbw(bmp, xres, yres);

//Deallocate memory

	free(bmp);

	cudaFree(bmpDev);

}

The above is the function that defines the various parameters of the image and calls the drawing function (evalPoint). Is the grid too big, and if so how do I get around this issue?

Also, when I call nvcc in the cmd window, I also have to type -ccbin “Path to cl.exe” with it. Is there any way around this issue?

2

Your kernel probably takes too long to execute, so that the watchdog timer triggers. This is meant as a last line of defense against your computer becoming unusable because the screen in not updated while a CUDA kernel runs.

Divide the work between multiple kernel invocations to avoid this.