I finally caved and had a play with uGrids using BoobPhysics' calculation guide:
http://www.neogaf.com/forum/showpost.php?p=32733748&postcount=3024
Comparison is zoomed 300% (full-screen, non-zoomed version here:
http://www.abload.de/img/tesvugridcomparisonq87mb.gif), default uGrids setting is 5:
Results:
- I finally eliminated most of the shitty distance trees in my test area using uGrids=7.
- Performance decreased by 5-10 frames per second in multiple test locations using uGrids=7, and only 4-6 in the displayed area.
- uGrids=9 eliminated the remaining LOD3, zero-detail trees.
- Performance decreased by only ~9 frames per second in my test area compared to uGrids=5 due to there being only a few extra visible trees compared to uGrids=7, but in areas overlooking huge forests my frame rate dropped by about 20. This performance decrease may have been lower had I disabled other changes I've made to increase the draw distance of trees, effect and particle LOD, etcetera.
- Regardless, uGrids=9 crashed after about 30 minutes of running around and entering/exiting interiors. uGrids=7 did not crash in the hour I tried it for.
- Furthermore, I encountered several performance nosedives in some of the more demanding locations as my hard drive thrashed to load data. Again, this was not experienced with uGrids=7.
- The uGrids=9 crashing may be mitigated by Large Address Aware, though in my basic observations of system RAM usage during the test Skyrim never came close to hitting 2 or 3GB.
- uGrids=9 performance nosedives would likely be mitigated by using a SSD.
Given the above I will keep using uGrids=7 for the time being to up fidelity without compromising performance or stability.
Once again, this is my system for reference:
3-Way GTX 580 SLI, 1.5GB VRAM
i7-2600K @ 4.4GHz
8GB RAM
Western Digital Black WD1002FAEX-00Y9A0 C:\ HDD