OK, everybody, that's how good it has to look from now on.

That's now downloadable for the PS5. Here's a playthrough.

You must now be at least this good to launch an AAA title.

Nagle said:
You must now be at least this good to launch an AAA title.

Afaik, it goes below 30fps on PS5 pretty often. So if you need to deliver solid 60fps, it can still look less good.

Some guys catching up (hope industry follows as well):

The Nanite format is tricky, but public. We'll probably see plug-ins for Blender, etc. for creating Nanite meshes. Then Nanite-format assets for sale.

The Matrix demo is good to see, because we're seeing high-detail objects other than dirt and rocks. Everybody seems to demo their LOD algorithms on dirt and rocks, where it doesn't really matter what the details look like. Now we see it on stuff where edge flaws would be visible, and it still looks good.

Its nto clear how nanite deals with ray tracing on actual gpu and future iteration, a compoent of nanite works using a softwre rasterizer, and models must be static.

I wonder if hardware guys will catch up rethinking the current ahrdware pipeline , implementing visibility buffers rather than brute force rendering.

One of the articles about Nanite mentions that they can model a chain link fence as a 3D model and have it render efficiently as you zoom out. That had me puzzled for a bit. How do you reduce the level of detail on something with holes? Once each wire is down to a 3-sided tube, how do you go further.

I think I see how they do it. First, you've got to fill the holes with transparent triangles. “Fill”, in this sense, means that the silhouette of the object, seen from any direction, has no holes. Now there's something to merge. For an example, think of a bicycle wheel. You need to add transparent triangles to fill the open space between the spokes.

Now, as parts of the model drop to a lower level of detail and the spokes disappear, the LOD clustering algorithm can start to combine triangles from the spokes with transparent triangles, producing translucent triangles with some average of the texture values.

This isn't that valuable for a bicycle wheel, because it's not big enough. For a mile of chain-link fence, it's essential. Close up, near the camera, you can see every wire. In the distance, it's just a big greyish area. This will help for anything big with holes in it, like trees.

One implication of Nanite is that when a vertex is part of two triangles, but has different UVs and normals in each triangle, the system has to know that those are really the same vertex. In ordinary rendering, the renderer doesn't care about that.

Comments?

I am impressed with the unity demo, for the moment i am focusing in a personal project , but as soon as i find some holesin my work schedule i will start to experiment using a vulkan render.

Programmer71 said:
i will start to experiment using a vulkan render.

I successfully postpone this nightmare for 5 years now. Getting GUI and debug visualization to work was pain enough.
Actually i'm so afraid of low level gfx pipeline, i think a 100% compute based renderer would be easier for me. :D

Yes, its easier to write a software render than setup vulkan to redner a single triangle I think that also this is pushing lots of programmer to unity or ue5

Yeah, it looks like gfx is going an ‘experts only’ thing. Then no more experts from younger generations is the next problem maybe.
Well, i'm happy we have those low level APIs. I whine about complexity, but it's worth it.
But i'm not happy with lacking flexibility and progress on them.
Actually i would prefer vendor APIs. DX/VK always is compromised to the bare minimum, and even if newer HW would no longer need certain limitations, it takes too long for updates.
Mantle looked much simpler than VK and could do some things i really wanted.