Dense, volumetric maps are essential for safe robot navigation through cluttered spaces, as well as interaction with the environment. For latency and robustness, it is best if these can be computed on-board on computationally-constrained hardware from camera or LiDAR-based sensors. Previous works leave a gap between CPU-based systems for robotic mapping, which due to computation constraints limit map resolution or scale, and GPU-based reconstruction systems which omit features that are critical to robotic path planning. We introduce a library, nvblox, that aims to fill this gap, by GPU-accelerating robotic volumetric mapping, and which is optimized for embedded GPUs. nvblox delivers a significant performance improvement over the state of the art, achieving up to a 177x speed-up in surface reconstruction, and up to a 31x improvement in distance field computation, and is available open-source.