Most of what claims to be interactive visualization of big datasets relies on one of two strategies: pre-canning and sampling. However, both of these techniques have well-known limitations. Enter Map-D, a distributed end-to-end data analytics and visualization platform that can run on any number of GPUs, allowing millisecond query latencies over multi-terabyte datasets. In addition to supporting ultra-fast relational table and array querying, Map-D uses the native graphics pipeline of the GPU to render 2D and 3D visualizations of the results. By streaming these visualizations to a user''s browser via interactive 30fps H264 video, it can appear as if billions of data points are in the DOM, even on low-powered mobile clients.