Ollama now includes a significantly improved model scheduling system, reducing crashes due to out of memory issues, maximizing GPU utilization and performance, especially on multi-GPU systems.