Yeah there isn’t that much optimizations that can be done to realtime dynamic shadows. UE4 already have shadow caching, resolution scaling, depth only indices, optimized depth only shaders.
So then why is it slow with dynamic shadows ? Perhaps culling of the dynamic lights and shadows isn’t optimized as it is in idTech 6 ?
Local lights in UE4 have all feasible shadow culling approaches, including shadowcaster occlusion queries.
Slow, as you define it, should be viewed in a context of particular project and scene.
Place 4 point lights with huge influence radius, operating at decent resolution and high filter settings in a heavy scene and you will toast any rig.
Lights are frustum culled, occlusion culled and distance culled. Lights that haven’t moved use cached shadows.
The issue is Crytek knows what to do to make things run better and faster, it can be “tricks” and not necessarily some regular optimization step that’s known everywhere (just guessing). Because for example, if UE4 has every optimization like other engines, Cryengine is still capable of handling like hundreds of dynamic lights in a single level like in crysis 3, while UE4 needs only 4 dynamic lights to lose half the games frame rate.
Cryengine got there because it didn’t have any static lighting solution, so they had to make it work fast with dynamic lighting otherwise Crysis series wouldn’t exist. But since UE4 has lightmass and Epic never tries to create something realistic in large scale, they never try to improve dynamic lighting as their projects can go on with static lighting. Main improvement Epic does to dynamic lighting is to say “Don’t use dynamic lights here and there”.
There are alot more games with Dynamic GI.
The division, Rise of the tomb Raider, Forza Horizon, Just cause 3, quantum break, gta v, ac sydicate, watch dogs 2. Just to name a few.
The division has precalculate probes.
Rise of the tomb Raider use VXAO at best settings.
Forza Horizon use some voxel based GI. Maybe VXGI.
Just cause 3 use VXGI at best settings.
Quantum break use precalculate probes.
Gta V use precalculated probes.
Ac sydicate use precalculated probes.
Watch dogs 2 use precalculated probes.
Horizon Zero Dawn doesn’t have dynamic GI either.
To be honest, while absence of dynamic GI is like sin, absence of IBL is like both sin and crime. Should at least have IBL so we can place local probes for each area and have ambient lighting correspond to the environment itself, like in cryengine. That’s the least Epic should do. Placing a skylight and give an entire game world the same ambient light/color was totally obsolete years ago. IBL makes environments look tons better, until maybe one day people get dynamic GI.
Probably not going to happen anytime soon sad;y
IBL would make a lot of sense. Sphere reflections are already there. Code was there. It’s simple but powerful. Proven to work on multiple games. Leaking is a problem but distance field based system could help with that just like DFAO.
Probably won’t happen anytime soon sadly 
we’ve been asking for this one for months and years 
and I keep thinking this one shouldn’t be that hard to do. in my mind the theory would be to take the capture cubemap information and lerp it with the skylight just like the reflection does, except just using the lowest mip, and pass it over to the skylight color in place of the current skylight cubemap. I’ll probably try it sometime
Talking about IBL, I remember back in 2014 everybody was like “-It’s EpicGames, they got this, just give 'em time!”…
Well, if Fortnite needs IBL, we’ll have IBL… Until then, threads like this one are just for entertainment 
What are you talking about? no one uses Nvidia’s VXGI…
what do you mean by precalculate probes?
For starters, Forza Horizon 3 uses “voxel-based global illumination system to calculate light bounces in real-time.”
QB uses screen space dynamic GI in addition to their large scale irradiance GI volume.
and VXAO is an ambient occlusion feature, has nothing to do with GI. you are just making stuff up at this point.
QB actually didn’t ship with the screenspace GI. It’s all pre calculated probes.
Yeah I made mistake with Just cause 3.
But the burden of proof is on you. You made claims that these 9 games use dynamic global illumination. Now we have concluded that two of those use some sort of dynamic GI. Forza Horizon and Just Cause 3. How about other 7?
Like I can show some links that:
Tomb Raider use pre-calculated IBL probes and use VXAO to help with the bleeding.
http://www.tombraiderhq.com/downloads/Tomb-Tech-Rise-Visual-Effects-TombRaiderHQ.pdf
https://www.geforce.com/whats-new/guides/rise-of-the-tomb-raider-graphics-and-performance-guide
Overall I agree. Not denying need for GI or improvements.
But shadowcasting performance of local lights is nearly equal or close between these two.
I am totally willing to bet, that in a given scene, there would be no more than 15% discrepancy, attributed to shadow-related GPU time spent, provided that everything else is equalized, disregarding number of shadowcasting lights, let it be 4 or 400.
https://i.imgflip.com/2awl19.jpg
^^^ Me reading GI threads ^^^