AWS EC2 instances tend to scale key metrics in line with one another 1 vCPU + 1 Ram Unit, 2 vCPU + 2 Ram Unit, 4 vCPU + 4 Ram Unit, etc. So it is impossible to source an instance with a High Ram / vCPU ratio. I found this annoying as I wanted lots of memory to hold a large data object but wasn't so worried about vCPU performance. Also when I wanted to upgrade on EC2 to an instance with a more powerful GPU (more cuda cores, more onboard RAM) again I encountered a similar problem. I had to use an instance with 8 GPUs - an instance type that no longer has capacity on us-east-1. This larger instance type was almost too expensive for my requirements. It was also pretty slow, probably because it was a virtualized machine (?? I am out of my depth here but it would seem this could be a cause of weak performance).
I have discovered that unless you are boiling-the-ocean either mapping genomes, folding protein structures, or training CV models, you probably can get by with older technology provided you can run on an actual piece of hardware and not a virtualized instance (again could be out of my depth here).
I have successfully launched two very stable, powerful enough for my purposes, GPU-enabled instances. I thought that if other people had a similar need for low-priced but powerful computers for a short period of time, there might be some community / club that could be built around this need.