Programmer71 said:
I have read the presentation about nanite and it is alone a whole team effort, barely affordable for a single user

Working on similar things and having more experience on geometry processing than i wanted to get, i guess i could do Nanite tools and renderer in half a year.
I think Nanite is very good - pretty much the only true progress in games seen in a decade. I hope others invest some work into LOD too now. Nanite still does not solve all related problems.
But the rest of UE5, e.g. realtime lighting is still horribly inefficent and not future proof, imo.
If we all stop working and waiting on Epic or another big company to do all tech for us, games are already dead.
Video games are about constant progress in tech to enable new things. But this can only happen if some guys still work on such custom tech, including custom engines.
If we would not, even Epic would run out of capable engineers at some point?

“If we would not, even Epic would run out of capable engineers at some point?”

I think this is already happening, lots of people are now working on Unity or Unreal to make games and are totally unaware of the technology running under the hood, what I have noticed is that they don't even want to know. If Unity does that , why do I need to re-learn ? , I have worked for a couple of years as a tester / recruiter , I haven't found anyone who is capable of describing me a 3d rendering pipeline apart the obvious steps. When I ask how to improve things, the answer i got was : yes i have some ideas but better to stick to current tech.

Programmer71 said:
I think this is already happening, lots of people are now working on Unity or Unreal to make games and are totally unaware of the technology running under the hood, what I have noticed is that they don't even want to know.

Which may be also the reason a site like this is no longer that interesting as in the early days. At least for a low level guy like me.
Back then, we were excited to learn the basics so we can make the games of our dreams. Nowadays that's not really necessary.
Knowing how to work with Unity is enough. Knowing the inner workings of Unity in detail won't help so much, as we can not change it anyway.
But is this a true problem, or just a necessary consequence from specialization due to games became bigger on content and technical complexity?
It should be no problem in very most cases. But we get just that - a uniform standard of generic functionality. Good to make more clones and mixes of former games, but not enabling entirely new tech for new games.

Aside of engines discussion, we also have the problem of growing costs for content. Bigger games, more detail. Thus, even AAA companies developing their own engines are stuck on a current state of the art.
Potential new technology may also require new tools for content creation, changing established workflows, causing a revolution nobody can afford.

That said, maybe custom engines is indeed no great idea, admitted. At least it's a high risk to invest. But no risk, no fun. And i really miss the latter in current games ; )

This is not my first account here on gamedev , i am here since 1998 , i remember clearly those days, they were so fun , engaging and full of competent people, I have learned a lot from their advices and paper suggestions , even if i rarely posted. Probably I wouldn't have researched on my own, in areas like 3d visibility , rendering and other graph topics which were unaccessible and well kept screct by those game studios.

Yes, this sites are slowly degradating and disappearing, I only hope that one day some kind of standardizaion ( sp ?? ) will unbundle proprietary features from various engines. Some companies are alrady catching up with mesh shaders, and probably in 3-4 years, raytracing will be the dominant pipeline.

Programmer71 said:
Some companies are alrady catching up with mesh shaders, and probably in 3-4 years, raytracing will be the dominant pipeline.

Yes, but i expect more than just adoption of the latest GPU features. And i'm not sure we can count on affordable high end GPUs anymore at all. They keep Moores Law alive only with premium prices, and the new features are not all that great if looking in detail.
In contrast Nanite achieved a true visual leap only in software. While DXR is not flexible enough to support such dynamic geometry at all. To combine Nanite with raytracing, they can only use low poly proxy meshes. Which means the problem they just solved for rasterization (continuous LOD) is still present for the inferior geometry used for RT.
I have the exact same problem and thus can't use HW RT. From my perspective it's useless until the black box around BVH data structures is lifted, which i doubt will ever happen. More likely they'll make hardware BVH builders like ImgTech did long before RTX, and after that we are doomed to rebuild BVH for changing geometry every frame. And if we want continuous LOD (as we do for decades), it will never work efficiently with hardware raytracing.
To be clear: Any continuos LOD solution requires gradual changes on a mesh as distance changes, and even a minor change currently forces a complete rebuild of BVH for the whole model, which is not realtime.
Thus i'm not impressed from NVidias or Microsofts short sighted ‘solution’ of a certain problem by preventing the solution of another problem at the same time.
If BVH data structures were open, we could not only stream it from disk and save a lot redundant GPU processing, we could also modify BVH together with geometry, and no problem comes up.
On consoles that's possible in theory, but not sure if anyone invests in developing solutions restricted to certain platforms, just to demonstrate that DXR sucks.

The other point is the hardware itself. Rumors e.g. about RDNA3 say it's a 65 TF GPU, a true monster for true premium price. They do not even plan an entry level or affordable mid range. Seems they are fine with HQ visuals being a niche for the rich, as they can't produce enough GPUs for the masses anyway. And i don't think this will change so much after chip shortage is over.

So what? Pay 5000 bucks for a box just to play Call of Duty 35 and Assassins Creed 20 with build in remake of AC 1? Still the same stuff?
This does not look good. There is still growth, and with Corona even more, but at some point near in the future the ship will start to sink. Video Game Crash 2 - game over.
Then, a new generation will rise from the ashes. Maybe with more efficient HW like Apple M1 shows. And the fun may be back… :D

Isn't bvh bounding volume hierarchy ? what's so difficult to implement on our own ?

Programmer71 said:what's so difficult to implement on our own ?

Nothing. But you can not use your own BVH for HW RT. That's my problem.

Brian Karis also confirmed DXR is not compatible with Nanite and requested access to BVH.

In DXR, you can only provide a mesh and the driver builds custom and black boxed BVH for tracing (different HW vendors have different data structures).

You can also have an animated mesh (skinned characters), and you can build BVH once and then only refit it per frame.

But you can not have any changes on mesh topology. If so, a full rebuild is necessary. So if you have continuous LOD, you have to rebuild BVH for the entire scene every frame, which ofc. is not possible in real time.

It's rarely discussed because games usually have no continuous LOD. GPU power allowed us to ignore it for a long time. UE5 is the first to have this for the whole scene, and we see what it gives.

If you're interested in technical discussion, this is still a good forum: https://forum.beyond3d.com​

(bad formatting due to browser issues)