One key metric when using GPUs is VRAM usage. We want to expose it next to CPU and memory usage when a Service is deployed on a GPU.