We run Postgres on local NVMe

That should be quite obvious for people who respect the hardware they work on

Having in mind rough numbers of all the latency involved on each "hop" from CPU is crucial

[dead]