firearrows wrote:
I have this issue too. Turning off shadows and water quality kinda fixed it but I don't know why the drop in FPS.
Which GPU do you have?
I spent most of the today investigating this and trying to pinpoint the things that affect FPS on my desktop with ATI and my laptop with NVIDIA (Core i5, Geforce GT540M). For an unknown reason, the change from using texture wrap mode with building and wall texturing to texture clamp mode and arranging vertices in code so that they end up making similar wrapping visually, boosted my performance on the desktop from that ~50 to ~85 FPS on that one measuring point. Other optimizations like the new water affected just a few FPS for the better in total when everything was enabled.
On the laptop with NVIDIA, the change from texture wrap to clamp didn't affect a thing, 68 FPS on 720p (I don't think I should be able to run 1080p on that laptop). My best guess is that ATI, in general or my GPU, has poor support for this texture wrap mode, which has been hindering the performance big time in my case.
As that was the case on NVIDIA, it means that the additions we made are eating out the FPS then. The thicker grass is one of the bigger ones of them, that's why it's optional at this point so you can (almost) revert to 0.76 in terms of grass to reduce plenty of polygons for now. On my lappy, I'm getting a 10% increase in FPS if I remove the grass.
I studied the character model improvements and effects at another measuring point, having 30 immobile soldiers posing on the screen, and the difference was minimal between 0.76 and 0.77 character rendering methods. The shading effect seems to be practically free on both my systems, the integrated outline which was made to reduce batch count didn't actually affect at all in performance which is a shame.
The optimizations will continue for sure, this episode helps me to understand that my desktop GPU is in fact rather powerful, with 0.76 it wasn't, and paying extra attention to optimization is even more important now.
We've already identified plenty of batches to remove for the next version and I'm hoping to get good results from the collision detection optimizations I'm working with. Getting to the next release shouldn't take a month this time, it never should of course.