In situ video encoding of floating-point volume data using special-purpose hardware for a posteriori rendering and analysis

Scientific simulations typically store only a small fraction of computed timesteps due to storage and I/O bandwidth limitations. Previous work has demonstrated the compressibility of floating-point volume data, but such compression often comes with a tradeoff between computational complexity and the achievable compression ratio. This work demonstrates the use of special-purpose video encoding hardware on the GPU which is present but (to the best of our knowledge) completely unused in current GPU-equipped super computers such as Titan. We show that lossy encoding allows the output of far more data at sufficient quality for a posteriori rendering and analysis. We also show that the encoding can be computed in parallel to general-purpose computation due to the special-purpose hardware. Finally, we demonstrate such encoded volumes are inexpensive to decode in memory during analysis, making it unnecessary to ever store the decompressed volumes on disk.