[ T3D Alpha Suggestion] Change maxTexSize to 2048 for 3 splits, 4096 fo 2 splits
by Joshua Halls (Xerves) · in Torque 3D Professional · 11/25/2009 (3:22 am) · 3 replies
This is in regard to splits for the dynamic lights. Was testing this a bit and found we can do a fairly decent progression here and open the options up to the client to change them.
Looks like this (using the defaults and changing to the following)
4 splits - 250 distance - 1024 texsize (2048 would be fine here, need to test on a 512MB card though)
3 splits - 400 distance - 2048 texsise
2 splits - 700 distance - 4096 texsize
For the most part each progression from 2 to 4 improves the visual quality at the cost of framerate. Even at 2 splits with a highres texture it looks good enough to me if you are trying to push a scene from 30 to 40 or 40 to 50 FPS.
You can change these yourself in the code in lightShadowMap by changing the maxTexSize. Anything to eek out a few extra frames for low-mid range cards.
Looks like this (using the defaults and changing to the following)
4 splits - 250 distance - 1024 texsize (2048 would be fine here, need to test on a 512MB card though)
3 splits - 400 distance - 2048 texsise
2 splits - 700 distance - 4096 texsize
For the most part each progression from 2 to 4 improves the visual quality at the cost of framerate. Even at 2 splits with a highres texture it looks good enough to me if you are trying to push a scene from 30 to 40 or 40 to 50 FPS.
You can change these yourself in the code in lightShadowMap by changing the maxTexSize. Anything to eek out a few extra frames for low-mid range cards.
About the author
Part of the team that works on The Repopulation, a SciFi based MMO using a heavily modified version of the Torque MMO Kit - T3D. I also take care of the T3D version of the Torque MMO Kit.
#2
For the Pacific demo we plan to do shadow performance tweaks in two ways.
First issue is lower end CPUs and too many draw calls. While yes you can go to 2 splits from 4 and reduce your draw calls.... you can also just draw less stuff into the shadows.
There is currently a $pref::ts::detailAdjust which is a scalar which can scale how quickly or slowly lod changes occur. I plan to have a seperate DTS lod scale for the pssm shadow for beta 1.1.
Second to address lower fillrate/memory on GPUs there will be a pref that controls the overal shadow texture scale (you can see where in LightShadowMap::getBestTexSize that will go). By going from 4 1024s to 4 512s you've reduced your shadow rendering fillrate to 1/4... a huge jump.
Anyway... thats how i plan to approch it in Pacific and it will probably be the default in most of the demos.
11/25/2009 (4:40 pm)
Well currently the system isn't setup for the number of splits to be adjusted as a performance tweak for end users. You could totally do it if you like... but i don't have plans to do this for the core.For the Pacific demo we plan to do shadow performance tweaks in two ways.
First issue is lower end CPUs and too many draw calls. While yes you can go to 2 splits from 4 and reduce your draw calls.... you can also just draw less stuff into the shadows.
There is currently a $pref::ts::detailAdjust which is a scalar which can scale how quickly or slowly lod changes occur. I plan to have a seperate DTS lod scale for the pssm shadow for beta 1.1.
Second to address lower fillrate/memory on GPUs there will be a pref that controls the overal shadow texture scale (you can see where in LightShadowMap::getBestTexSize that will go). By going from 4 1024s to 4 512s you've reduced your shadow rendering fillrate to 1/4... a huge jump.
Anyway... thats how i plan to approch it in Pacific and it will probably be the default in most of the demos.
#3
11/26/2009 (1:05 pm)
Thanks for that info Tom. Look forward to seeing the demo and learning all about the little tweeks. We have exposed a few controls already on our own: Sun Shadows, Pixel size occlusion, terrain distance, groundcover density/radius, and a few other things. Will probably work on some options to change split/quality as well on the client to get a nice lower end for some of the low-mid end cards can run if they choose. Wouldn't mind seeing some more work on BL either as I am pretty sure I can get better performance in TGEA now in a similiar scene. 
Torque 3D Owner Joshua Halls (Xerves)