If you need that many concurrent connections, invest in a connection pooler. If no connection pooler is used, knock off 800 connections if you can.
shared_buffers = 8000MB
Test with a lower setting like 4GB, depending on your workload, that may or may not perform better.
work_mem = 4096MB
This magnitudes too high. It is per query * sort and aggregation steps. If you expect queries with large result sets (or distinct, group by, etc), 100MB is probably a reasonable starting point.
maintenance_work_mem = 512MB
Depending on your relation size, you might want to double this.
checkpoint_segments = 512
Is that a reason you chose such a high number?
enable_nestloop = off
Unless you have a clear reason to disable this (it applies to all queries), leave this enabled.
autovacuum = off
autovacuum should be on unless you know your data churn patterns and have scheduled manual vacuums.
