Neural program modeling! Our newest server sports eight Nvidia RTX PRO 6000 GPU cards and 1.5 TB of RAM to train machine learning models that replicate the behavior of existing programs. The resulting models can instantly predict inputs for desired execution features, predict the coverage of given inputs, and even replace the original program for safe testing. (When not busy, the server can run Doom, too – 17,208 instances in parallel) Read more in our recent TOSEM paper: https://dl.acm.org/doi/10.1145/3845620