I continually have my Serverless (nearly half of them at any given time) in a "Throttled" state, which, from my front-end, means unusable. I took support's advice and moved everything to a new zone with higher availability, chose to allow only machines with CUDA 12.8, and selected only compatible GPUs. I either have to go into Runpod and decrease workers to 0, save, then increase workers to 1, or delete the worker and hope it creates a new one in Ready status. This makes the whole experience very painful as I have to administer the serverless before using the front end.