using 4.10 mannequin with standard walk/run animation blend
using default crowd params (not detour)
RVO disabled
EPIC settings
With 100 spawned agents on screen (all visible) my stat units are:
FRAME: 25 ms
GAME: 20 ms
DRAW:0.14 ms
GPU: 25 ms
FPS: 40
I am wondering if this can be optimised further.
Ideally I would like to have around 500 agent on screen at around 40-50 FPS so I have sufficient FPS quota available for everything else.
Can anyone offer any thoughts/tips on whether further optimisation is possible?
Is 40FPS for 100 characters (with my system specs) average, above average or below average performance?
This puppet model is modestly high-poly and has a fair amount of bones.
If you setup LOD settings for your crowd characters, reducing poly count and bones per actor, you’ll gain significant boost in performance.
You can also force on lower LODs single frame animations to play when they are very far from camera
You can also disable LOD distance auto-control and build some algorithms that takes into account not only the LOD distance to camera, but also checks to see how many other characters it has around; the more people around that character the less detail from the body is visible and you can force a low LOD level for him while he’s in middle of the crowd, even when close to the camera, and increase LOD for the adjacent actors closer to the camera whose body-parts are constantly visible.
(this however involves complex programming and could require a fair level of experience)
I don’t think polycount is the main offender.
Toggling visibility of the crowd during PIE has little effect on performance
so I am suspecting the bones and/or animations are the culprits.
It would be great if num bones to process use could be changed dynamically
for distant characters where animation quality won’t be noticed.
However, I don’t yet know if that is possible. I suspect not, and I would have to
switch to a duplicate secondary character with less bones.