Yeah, I know how that works. And no, none of that. In my case it was renting a GPU for training a model. No put traffic involved in either of the cases, the data was in a s3 prior to that. It was simply tweaking hyperparameters in the code, nothing more. On both occasions it took the exact same time, exact same to train, exact same resources. There were apparently no changes to the pricing but still. I'm referring to aws's quantum bills as a whole. Never had such issues with gcp... Yet...
In that case, that sounds like Amazon is charging you differently based on how other people are using their resources. In the times of surge pricing, I wouldn't be surprised if they are doing that to make it so they make the same amount of profit, regardless of how much you are directly using, but depending on whether they are getting enough money from all tenants of a particular shared host. So if other people are paying less, they jack up your pricing to equalize things out so at the end of the day, their systems are still hitting their profit target.
