Here are some of my tests. I used a material with nothing plugged into its bump slot. It was only using displacement.
Rendered for 5 minutes into 4096x2304 resolution.
Used Corona Renderer 4 Hotfix 1 and an internal daily build (not yet released to the public) - 2019-10-04 grid-autobump
1) 3px with autobump on - V5 2.5D displacement vs V4 classic displacement
RAM: 1.8 vs 6.1
Parsing: 1.8 vs 14.7
https://corona-renderer.com/comparer/c7s2om2) 3px classic displacement with autobump on - V5 vs V4
RAM: 5.1 vs 6.1
Parsing: 12.3 vs 14.7
https://corona-renderer.com/comparer/Hpo6XL3) 1px classic displacement with autobump off (pure geometry) - V5 vs V4
RAM: 41.7 vs 49.5
Parsing: 122.4 vs 138.6
https://corona-renderer.com/comparer/MwcRUQ4) 1px without autobump (pure geometry) - V5 new displacement vs V4 old displacement
RAM:
3.5 vs 49.5 - 14x differenceParsing:
17.3 vs 138.6 - 8x differencehttps://corona-renderer.com/comparer/VnB46Y5) V5 new displacement only - 1px without autobump (pure geometry) vs 3px with autobump
RAM: 3.5 vs 1.8
Parsing: 17.3 vs 1.8
https://corona-renderer.com/comparer/NrVvT26) V5 new displacement 1 px without autobump (pure geometry) vs V4 old displacement 3px with autobump
RAM: 3.5 vs 6.1
Parsing: 17.3 vs 14.7
https://corona-renderer.com/comparer/FrWJGm7) Geometry normals - 1 px without autobump - new 2.5D displacement vs old displacement
https://corona-renderer.com/comparer/BqzLGUNote: to quickly see which example is rendered in old version and which in new version, just check out the render stamp. The new version is the one with antialiasing in render stamp font. :)
These tests were done just to get some numbers.
In the next ones I will try to get numbers + show the benefit of using the new displacement when it comes to image quality.