I was looking for access to a single high-RAM GPU to do some image and video processing using VQGAN (pytorch/CUDA) and so I was very excited to find out DataCrunch had Tesla A100 80GB available. I had only previously used the "big cloud providers" and find their overall environments and account management to be too cumbersome for my use case, but more importantly, there is little to no availability of very high RAM GPUs on their stack when I need them. So I gave DataCrunch a try and was really impressed how streamlined and focused they are for this kind of work. You basically choose a machine, the OS/Image and the storage needs, choose your SSH key and hit go. It is very easy to see where you are at with your usage credits at any moment which makes understanding your budget really easy. The hibernate feature for keeping your image stored but your machine costs down when not using them is easy to understand too. I just generally love how I was able to get up and running including installation of everything I needed and troubleshooting in about 15 minutes from start to finish because their user interface and overall value proposition is so straightforward and focused. That and the hourly costs are very reasonable and for these high RAM GPU machines they are actually cheaper than I priced them at "big cloud" providers. I really hope it stays that way because this is absolutely my goto service for high RAM GPU computing.