What's your GPU? The difference is that having the backend automatically determine the correct amount of layers to offload to the GPU doesn't always work, which can then results in very slow processing.
