[cramr5]

UPDATE:
I have been using AWS for a while now and pretty happy with it.
It is quite complex to set up yourself but once you get the hang of it it's quite straight forward and user friendly. On the good side, there is plenty of detailed and good documentation on how AWS works in Amazon and youtube that is very helpful.

Even using a single instance you can go up to 192c + 768GB RAM on new EPYC dual CPU systems [hpc7a] and it's pretty quick for a reasonable 7.7USD/h.
That can also be way cheaper if you are willing to use "spot instances". The downside is that there is around 5% chances that they will need your resources and then they will kill your job with 2 minutes notice. Could be fine during working hours if you are able to monitor the system.

I have also tried their Graviton ARM system and it's the best in terms of cost/performance since it's as quick as AMD systems but with reduced cost. However, there are some ARM limitations that can make it less convenient for some people.

I have also learnt to use the AWS cluster as "compute" and keep the client locally to perform the rendering and most pre-post which works great. The only downsize is that our network speed is not that great and trasnfering big files can be a bit slow.
The best workflow I found was to do the setup + surface mesh (that is single core) locally, upload the case with the surface mesh to AWS and there do the parallel volume mesh and run.

So far pretty happy with it. The speed increase is so far linear and I have not yet used all the potential of AWS by using ParallelCluster service and so on (but I don't think we need that for our needs right now).

Hope this update helps someone.