Cubemap large memory footprint

I have a skybox composed of 6 texture of 1024x1024. The cubemap works correctly but CodeXL tells me that the cubemap texture takes 110MB. It finds no OpenGL errors.
It takes only 18.4MB on Intel GPU.

This is my code of the cubemap generation:

static const char*  names[] = { "/right.", "/left.", "/top.", "/bottom.", "/back.", "/front." };
static const GLenum types[] = {
    GL_TEXTURE_CUBE_MAP_POSITIVE_X, GL_TEXTURE_CUBE_MAP_NEGATIVE_X,
    GL_TEXTURE_CUBE_MAP_POSITIVE_Y, GL_TEXTURE_CUBE_MAP_NEGATIVE_Y,
    GL_TEXTURE_CUBE_MAP_NEGATIVE_Z, GL_TEXTURE_CUBE_MAP_POSITIVE_Z
};

glGenTextures(1, &_id);
glBindTexture(GL_TEXTURE_CUBE_MAP, _id);

glTexParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glTexParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
glTexParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
glTexParameteri(GL_TEXTURE_CUBE_MAP, GL_TEXTURE_WRAP_R, GL_CLAMP_TO_EDGE);

for (int i = 0; i < 6; ++i)
{
    SDL_Surface *image = loadImage(path + names[i] + extension);
    if (image == nullptr)
        return false;
    int     format = image->format->Amask ? GL_RGBA : GL_RGB;
    glTexImage2D(types[i], 0, format, image->w, image->h, 0, format, GL_UNSIGNED_BYTE, image->pixels);
    SDL_FreeSurface(image);
}
glBindTexture(GL_TEXTURE_CUBE_MAP, 0);
return true;

I have tested on a GTX 760m with driver 340.52 and a GTX 560ti with driver 340.65.

Stack Overflow : http://stackoverflow.com/questions/25701411/large-memory-footprint-with-opengl-cubemap