Training Stops: CPU stays busy, GPU goes to zero

Im training an RL agent on another vision task with the simple neural net from one of the first pong papers. Everything goes smoothly for 48 epochs, each having 240 steps. 7950x CPU utilization at about 18%, RTX 3090 GPU utilization at about 8% and using about 2 gigs of memory. Temps low on both. RAM at 44 of 64 gigs. Once I reach 48 epochs CPU keeps being utilized, GPU memory stays the same, GPU utilization goes to zero and no new epochs complete. I restart the program several times, always getting to epoch 48, and then the same problem. I restart my computer and now I get to epoch 51, and then the same problem. Restart the program and epoch 51 again before the same problem. Any advice on where to begin trouble shooting this?