lopasSSR v.0.0.a trial release
It works on upbge v0.2.4 render attachemnt branch
all download link are under the video description
6 Likes
it looks better than eevee
1 Like
thank you, many features have not yet been implemented…but they will be soon (free time allowing)
What’s the impact on GPU performance once this is used?
Fred/K.S
unfortunately in 0.2.4b render attachment branch does not work the unlocking of the frames, so I can hardly tell you a precise value, in 2.3 render attachment branch the same algorithm works in fullHD around 200 fps on the GTX 1050Ti of my notebook
Looks already amazing! Any chance for Eevee?
Mkay.
sounds fair enough!
Fred/K.S
i tested it with ryzen 5 2400g in fedora linux using wine and i get 60fps in the small window and 20 to 30 fps in “fullscreen” (large window). in theory the fps should be higher on a native windows system or using linux binaries. some optimization and a faster upbge should increase it even further, as well as lower quality.
but that’s with the test file which is two planes and four spheres, maybe the filter is a bit heavy.
the effect is completely on the gpu (not the cpu) and is independent of the complexity of the scene.
what’s your gpu?
Eevee already has a SSR filter with a progressive temporal refinement that makes it very good at rough surfaces rendering but in a non-in-game situation. At the moment I’m adding support to the rough surfaces even in my SSR but I have to use a different approach just because i’m interested in gaming so my results will be faster than eevee (the result must be completed in 1 frame only) but probably it will be not so precise. A game engine and blender-eevee have different purposes so something ideal for one could not be a good choice for the other and that’s why I think it’s not a good idea to try to bring eevee’s filters / shaders into upbge
2 Likes
ryzen 5 2400g…
the gpu is in the cpu.
sorry, I was ignoring the fact that you were referring to an integrated gpu … .probably emulation and low level gpu (i say low level because mine is considered mid-level and i’m seeing have highter benchmark results then RX vega 11) are not yet a good combination for this kind of effects using fullHD sampling, I have to introduce, in the next release, the half resolution sampling mode and this should brings a 4x increase in terms the speed … hopefully go better, thank you for having tested and posted your results, this helps me!