ENBSeries

The Elder Scrolls Skyrim (TES Skyrim, Скурим Свитки)

Graphic mod ENBSeries / patch ENBoost 0.244

Optimized ambient occlusions for UseOld=false mode. Precision parameter have less impact on performance than before.

Nice!
Measurable performance increase of about 2 fps for me with SamplingPrecision on 'Medium'.. 'High' and 'Extreme' take less of a toll as well, so I'll stick with 'High' now.

Performance of individual effect measured by profiler, fps is useless, especially difference.

Before:
SSAO ~0.0165

After:
SSAO ~0.0072

With SamplingPrecision on High.
It definitely is faster.

Boris
Are you kidding me ?
Using extreme parameters with scales ramped up all the way and beyond, the performance improvement (w/ SamplingPrecision on extreme) is what I call stellar !
Even if FPS isn't correct measure, it's +3 to +5 in Whiterun, +2 to 3 average in tundra... and up to +5 in some interiors...
Also, is it an idea, or is there even less "noise" generated ? It's almost totally unnoticeable now....
Well done !!

kaffeekranz
Okay, these are proper result, ssao/ssil work twice faster.

Oyama
Noise decreased a little for more or less planar areas, because new code using adaptive level of detail, but not a big difference. I'll try to optimize a little more, but this is not that optimization we spoke about, don't want to implement it, better to move to gi and keep current ssao/ssil as old effects.

Boris
Yeah, I got that was not what we talked about
But this is already a real nice step for users Really faster, it's obvious.

EDIT : also, bare with my ignorance plz , but GI takes care of both occlusion and illumination, right ?
If it's even more accurate than your actual algorythm, I can' t wait for this ! And I personally don't give a single fuck about the performance in early stages of development.
As you said, actual AO-IL code is already fantastic, performance wise, most users don't use over 1.0 scale and over 0 or 1 qualities @1080p, and performance with these values is excellent.

I've never been so excited about an optimization before! I've been pushing my scales a bit lately and have been noticing small performance hits, but this has pretty much alleviated it Thanks Boris!

Btw, when you say GI, I'm assuming you mean Global Illumination? Just checking

Jafin16 wrote: Btw, when you say GI, I'm assuming you mean Global Illumination? Just checking

I wouldn't speak in Boris's place, but I think that's it.
That's what I understood, but I may be wrong...

Version updated, download again
Made workaround against some compiler ugliness, instead of proper optimizing ssao code, it added a lot of useless instructions. Part of them i don't know how to remove without writing in assembly, but this update is better than previous and noticable for videocards with small amount of shader units.


Gi is global illumination, code which i'd like to implement have better quality than current ssao/ssil, it almost don't have noise, but have higher minimal performance penalty than current ssao/ssil. Probably i'll try first partially compute ssao/ssil based on that technique, in case of success by performance vs quality - green light. But this probably not next in plans, want to finish clouds, it's more complex and creative task, so more interesting.