I’m struggling to get a very simple C++ Cuda application to run inside a docker container.

• The code:

#include <stdio.h>

__global__ void helloCUDA()
{
    printf("Hello CUDA from a CUDA Kernel!n");
}

int main()
{
    helloCUDA<<<1, 1>>>();
    cudaDeviceSynchronize();
    return 0;
}

• The base image:

nvidia/cuda:12.2.2-devel-ubuntu22.04

• I run the container as follows:

docker run –gpus all -d –name container -it –rm mycudaimage

• Inside the container, I can see my GPU when I run nvidia-smi:

+—————————————–

| NVIDIA-SMI 550.99

Driver Version: 552.74

CUDA Version: 12.4

|—————————————–+

| GPU Name Persistence-M |

| Fan Temp Perf Pwr:Usage/Cap |

| |

|=========================================+

| 0 Quadro P3200 On |

| N/A 60C P5 9W / 88W |

| |

+—————————————–+

• I compile my code using nvcc:

nvcc hello_world.cu

And run the ./a.out, but nothing happens. I have already tried to print statements outside the Cuda kernel, and it works fine.

Any suggestions?

Thank you,
Rafael.