The fundamental properties of light restrict the amount of information obtainable from microscopic images. Deconvolution is a well-known software technique used to help overcome the limitations of diffraction by increasing the resolution and contrast of fluorescence microscopy. However, it has traditionally been too slow for use in parallel with image acquisition and is instead relegated to post-processing of chosen subsets. We'll showcase CUDA-accelerated Richardson Lucy-based 3D deconvolution software that can run in near real time, permitting deconvolution to be integrated with image acquisition. We'll explain the process of deconvolution and demonstrate its application with traditional and emerging microscope technologies. Our results show increased signal-to-noise ratio and resolution, further improving even newer super-resolution techniques and producing 120 nm resolution or better with much gentler imaging conditions and shorter processing than traditional super-resolution microscopy. We'll also present forthcoming developments in multi-view deconvolution for very large samples. Finally, we'll compare execution time on NVIDIA hardware ranging from embedded platforms to large-scale cluster deployments.