This scene doesn’t fit my description of a torture test. It’s a simple plane with a transparent texture. It’s not a matter of “fast vs accurate” here, but “fast is worthless.” It doesn’t matter how fast i get a shadow, if the quality is literally unusable.
And “use cycles” is not something I consider to be an acceptable alternative:
That’s nice to read; I’m looking forward to it.
Correct. Here is a comparison in the same 4.2 build, using each.
The PCF filtering (just added) does blur the shadow lines with noise, but it’s not a clean result.
Eevee at present (and going back in versions) does not have the cleanest shadow edges to begin with, but Next is making them far worse. Ideally, they would be improving in Next.
Well… they do are better than the shadow volumes (an object modeled to be the shadow) commonly used in Cell phone games… and those shadow “shrink warp” kind of thing that projects a copy of the character geometry on the ground… Shadows never were easy to make. I don’t see any solution that works good with shadow maps as they are a map that has resolution…
… probably ray tracing them but then people would probably need the RTX cards, right!?
…Stupid question for sure, but 
…can’t the shadow map be like UDIM textures!?
I have an RTX card, so that’s fine with me. 
Jokes (not really, but you know) aside, I’ll agree that perhaps raytracing may ultimately be required to achieve a clean hard shadow - and, be outside of what can be accomplished in the current phase. BUT, as the plan is to remove Eevee at some future point and only have Next in the software, the current situation with shadows is a serious regression.
So with that, Eevee shouldn’t be removed with things in this state. Perhaps that means 4.X would have both versions of Eevee. I don’t know the full implications of that, but the original goal as stated in the first post is “what is essential for Blender 4.1 to have the same feature set as EEVEE in Blender 4.0.” I believe having shadows that are at least as good as Eevee, is essential.
Or - sure. As there’s currently coding being done to try and improve Next on slow systems with slow integrated GPUs, perhaps consider also coding Next to optional raytrace the shadows if one has an RTX card.
If it was easy it would be already done for sure! 
I obviously agree that going from horse to donkey feels terrible, and I miss the times when things were invented, not just implemented, but I don’t have any solution for the problem.
There’s for sure a plan and now going back and coding everything in a radically different approach “just” to have better shadows, would slow everything down.
Perhaps this has a solution on the next Eevee re-write
It’s not about “just having better shadows”, it’s about the fact that Eevee Next has been repeatedly promised to have 100% feature parity with Eevee Legacy. If Eevee Legacy can do something Eevee Next can’t, it’s not feature parity, and Eevee Next cannot- by the developer’s own promises of feature parity- be considered complete
Can we do something to help the devs on this!?
I assume that they know all the papers and all the current methods… so what can we, normal final users people, do to help?
Keep pointing out what’s slipping through the cracks and hope, that’s about all we can do
What Joseph said, stress stress stress the heck out of every function Blender has. Believe it not the developers of software want you to find their mistakes or possibly prove a concept wrong. Top developers have really thick skin and don’t take anything thing personally, so keep the constructive criticism flowing.
On a side note: as an end user you should also have thick skin towards your use cases. There will times when other users may give you grief for using “function X” in a way that it wasn’t designed for. Unintended innovative uses cases makes software grow. And top developers will most likely try to incorporate your new use case into their code down the road. 
Regarding shadows, any reason why the Blender devs. can’t just reimplement the shadow system Eevee standard uses? It may not have the best quality in all cases, but it works (including soft shadowing) and should be seen as an option if something better cannot be done.
Thanks @josephhansen and @CookItOff! Will do my best to keep reviewing it. I download the builds every day and use them at work… although using Eevee Legacy as “Next” still has show stoppers.
@Ace_Dragon
It’s different rendering methods, right!? They were using “forward rendering” in Eevee-Legacy and now is “deferred rendering” on Eevee-Next (I think). Two very different methods to calculate the objects positions.
On “Forward” the objects are calculated one by one for each light, on “Deferred” the objects are buffered first.
As far as I understand it, deferred rendering still supports traditional shadow maps though. VSM is a relatively recent development and deferred rendering has been around for ages. The only other thing eevee legacy does is render those shadows multiple times from slightly different points, for soft shadows. Seems simple enough to someone like me with next to no knowledge on the subject.
You’re probably right about this. I doubt they’d remove the option for no reason, but it’d be nice to get some clarity or extra information on this because it doesn’t seem to make much sense at first glance.
Please keep op topic.
Material->Settings->Surface->Render Method Dithered will use a deferred render style, blended will use a forward rendering render style.I think mainly what users want to user would like to understand is why with PCF the shadow quality is still worse in some cases, and if that’s a known issue that will be addressed.
It now uses geometry boundaries!Edit:
I also easily notice that on inside shots Eevee-Next overhaul look is much better than the one in Eevee-Legacy, nice work!
Edit2:
As the scene got progressively more complete and intricate… performance again got substantially slower than on Eevee-Legacy… at least 30% less. I have a RTX 3060, Ryzen 7 5800X and 64Gb DDR4.
These are the stats for the scene:
It’s a full scene with 3 rigged characters, particles, volumes (for mist) assets, geometry nodes grass elements… GPU compositor… the full thing.
…and this without motion Blur.
We are on topic, we’re providing feedback about Eevee Next. Currently, the shadows in Eevee Next are not comparable with Eevee Legacy. Is that going to be fixed or not? That’s all we want to know, the technical details are not terribly helpful here
Volumes camera clipping is making something wrong:
When the camera is inside the volume:
And the volume settings start at almost 0:
Explaining the technical details is the only way we can help you understand the answer.
EEVEE-Next uses virtual shadow maps (which is the current state-of-the-art in real-time rasterized shadows).
For local lights (point/area/spot), they are supposed to be equivalent in resolution to 8192 px per Cube Size (explained in EEVEE-Legacy terms). So twice the max resolution of old EEVEE.
For sun lights the improvement is even better.
However, we can’t have actual 8k shadow-maps. That would require 1.6GB of VRAM just for a single point light, so it’s obviously not practical.
What virtual shadow maps do is rendering and storing only the parts of the shadow-map that are actually visible, and only at the resolution they actually require (you don’t need 8k resolution for a tiny point light far into the background, and you don’t need a texture at all for a light that it’s not even visible).
In practice, this system is quite complex, and sometimes this detection is not working as well as it should yet.
Also, a reminder that these kinds of threads are best for “subjective” feedback.
If you find an actual bug, something that it’s obviously not working as intended, then report it in the bug tracker, please.
If I understand you correctly, I would summarize this response as this: no, Eevee Next will not have shadows that have comparable visual results to Eevee Legacy. Is this a correct summary?
