[th0114nd]

I can’t recall exactly, I want to say 16 vCPUs and 64GB. How well a project can make use of the cores depends on how wide your dependency graph is vs how tall, as the unit of parallelism is a package.

The EC2 instance was on continuously, and I would ssh+tmux in. You could probably replace this with a server plugged in under your desk, depending on how much you trade off operational costs and capital costs.

More often than not I wanted passing tests to send something for review. If I needed an artifact, I would publish a docket image on an internal registry.