Digital Foundry Article Technical Discussion [2019]

Status
Not open for further replies.

This is the guy that made the PS2 Duck demo!
Kutaragi wanted him to work on the graphics that dont use polygons. And he managed to do it! And the rubber duck was the result!
Disappointingly enough this technology was never used in actual games, but it was probably a gate to the PS2's untapped powers.
It makes me wonder what we would have got if his work was used further.
The water effects of the tech demo are beyond anything that was done in real time graphics and it probably challenges present implementations.

Yes, that was all impressive stuff. Unfortunately, sometimes impressive tech just isn't feasible to pull off in anything other than a demo for a variety of reasons.

At least the PS2 didn't have the weight of the industry (polygon based rendering) going against it like NVidia's NV1 which used hardware accelerated quadratic surfaces instead of industry standard polygons.

So, it's likely that while non-polygonal rendering was possible on PS2, it just wasn't feasible at a game level as being a game console it didn't matter as much whether you used industry standard polygons or not. Especially when 3D rendering was still relatively new in games and the industry standard polygonal rendering was mostly used in non-gaming sectors.

IE - If it was possible to do non-polygonal rendered games on PS2, I'm sure Sony would have done it, especially with Kutaragi at the helm.

That said, regardless of its suitability for games at the time, the tech was definitely impressive!

Regards,
SB
 
Yes, that was all impressive stuff. Unfortunately, sometimes impressive tech just isn't feasible to pull off in anything other than a demo for a variety of reasons.

At least the PS2 didn't have the weight of the industry (polygon based rendering) going against it like NVidia's NV1 which used hardware accelerated quadratic surfaces instead of industry standard polygons.

So, it's likely that while non-polygonal rendering was possible on PS2, it just wasn't feasible at a game level as being a game console it didn't matter as much whether you used industry standard polygons or not. Especially when 3D rendering was still relatively new in games and the industry standard polygonal rendering was mostly used in non-gaming sectors.

IE - If it was possible to do non-polygonal rendered games on PS2, I'm sure Sony would have done it, especially with Kutaragi at the helm.

That said, regardless of its suitability for games at the time, the tech was definitely impressive!

Regards,
SB
Or maybe it was feasible, just not the industry standard for the developers to have to learn, use and design games using this method when the PS2 could handle polygons just fine.

edit: btw didnt SSX use NURBS for the terrain?
 
Last edited:
IE - If it was possible to do non-polygonal rendered games on PS2, I'm sure Sony would have done it, especially with Kutaragi at the helm.
It was possible. There are plenty of artistic reasons not to use NURBS etc though as Laa-Yosh went into in detail. There are realistically only so many objects that could really benefit. Given the need to integrate those objects into a triangle-based renderer as well, it's no wonder the tech remained niche.

I guess it's akin to the tessellation unit in XB360 - a tech that sounds great on paper but didn't really have value in the real world given it's practical limitations.
 
edit: btw didnt SSX use NURBS for the terrain?

Think i posted it here before but, DF said something about bezier rendering was used for SSX/tricky.


Seems like the premium consoles where prioritezed for this title, it's good that were getting new consoles next year.
 
Last edited:
Wait, no polygons? Instalove, for me! 3 I'm OBSESSED with alternatives to polygons.

There are plenty of artistic reasons not to use NURBS etc though as Laa-Yosh went into in detail.

Yeah, personally i experimented with bezier patches for characters before the year 2000. I was pretty excited when NV even added HW acceleration for this with GeForce3 - this was fast! But they abandoned the feature with the next generation because nobody used it.
Some time later i realized it's indeed pointless. For characters Catmull Clark subdivision is so much better. Modeling with bezier patches is true pain.

Recently there was one guy at the gamedev forum who worked on Nurbs for games. (Nurbs is basically stitched bezier patches, but for continuous curvature we need an extended form with weighted control points, called rational bezier patch IIRC.)
He listed several advantages for Nurbs:

Can be implemented efficiently with HW tessellation. (Catmull Clark is difficult, although some recent CoD game did it for characters.)
Modeling is much faster (but only for experienced CAD people, not game artists).
UV Parametrization is no extra work.
Dynamic LOD. (I said this can only increase detail, but does not help with the harder and more important problem of reducing them. He replied there would be options to remove patches, closing holes etc, but i think that's to complex to be practical.)
Less data and RAM.
Unlike Catmull Clark, NURBS are C2 continuous across irregular vertices.

But other people at the forum did not agree. Counter arguments:
Complexity of implementation.
Not artist friendly.

Personally i'm not convinced either because it's just another system to handle a single special case (likely human made stuff like vehicles or buildings. Nobody would want to model and animate Nurbs characters).
It sounds more reasonable to convert your Nurbs models to triangle meshes and work on a general LOD system for this instead. But i also think it's good some people still explore those alternatives...

However, parametric geometry is technically not an alternative to triangles (like voxels, points or SDF). Trinagles are still required, and any 'downsides' of triangles are amplified.
 
However, parametric geometry is technically not an alternative to triangles (like voxels, points or SDF). Trinagles are still required...
Not completely true. ;) You can rasterise NURBS directly. The ray-tracer RealSoft3D does not tessellate its NURBS but traces them mathematically. There may be ways to resolve NURBS geometry into something renderable without resorting to triangles. Future geometry representations might well be a development focus when ray-tracing is a primary rendering tech.
 
Seems like the premium consoles where prioritezed for this title, it's good that were getting new consoles next year.
From what hearing seems like PS was prioritized.
Benefits of being lead platform.

4pro seems to hold up very well, if anything this would show don't need new consoles than do.
 
There may be ways to resolve NURBS geometry into something renderable without resorting to triangles. Future geometry representations might well be a development focus when ray-tracing is a primary rendering tech.
Noooo! You can already do this with custom intersection shaders. Please no more extra fixed function stuff just to render curvy sports cars! :D
 
I wasn't thinking hardware. With RT, the full range of spatial representations opens up. Optimising rendering will probably become more about these software choices than anything. But that depends on how well the RT can be used to work with arbitrary models, and how well these models can be mapped to the processing capabilities.
 
From what hearing seems like PS was prioritized.
Benefits of being lead platform.

4pro seems to hold up very well, if anything this would show don't need new consoles than do.

Said premium consoles, not xbox consoles :)
It just doesn’t run well on the base ones, like many other modern graphical intense games, as DF stated in the video.

I then said it’s good we get new consoles since most own base units and don’t want to upgrade to a premium a year prior next-gen.
 
With RT, the full range of spatial representations opens up. Optimising rendering will probably become more about these software choices than anything.
What do you mean? Can you list some examples?

Parametric surfaces do not seem so attractive: There is no simple closed form solution to find intersection, so it's ALU heavy. It can not add details, only make close ups look perfectly smooth. At some distance triangles become more efficient quickly. Can not solve diffuse geometry problems like foliage.
 
Said premium consoles, not xbox consoles :)
It just doesn’t run well on the base ones, like many other modern graphical intense games, as DF stated in the video.

I then said it’s good we get new consoles since most own base units and don’t want to upgrade to a premium a year prior next-gen.
According to VG tech, PS4 and XBX versions of the game run very similarly: Pro > PS4 = XBX >> XB1

https://docs.google.com/spreadsheets/d/14lNNVFuzSHpRrTuJOj92yUC4I4JC2QhvysNQp13GU14/edit#gid=0
 
What do you mean? Can you list some examples?
Voxels and SDF for tech that presently exists. Separated from triangles, you no longer have to worry about representations that can be converted to triangles. Instead, future graphics will be about sampling a spatial position per pixel. And not just for pixels, but for traced features as well. So maybe we'll have light intensity modelled as SDF? We already have dabbling with these mixed representations with the likes of lighting volumes.

That all depends on what the RT hardware is capable of doing though. If it's built around tracing triangles, it won't be terribly helpful for exploring more efficient spatial models.
 
Voxels and SDF
Ha, ok.
With Voxels + HW RT i see one more option for the proposed traversal shaders: If BVH format would be exposed (vendor extensions on PC), one could trace voxels directly by using the bounding boxes for them.
So something like the Minecraft mod which used voxels for caching could be hardware accelerated if APIs allow so.
It's also interesting for glossy reflections: A voxel approximation would be fine at some distance, and voxel mip maps can be used to get prefiltered results.

Traversal shaders really seem the RT killer feature for me. Hopefully all vendors will support them wither their next / first RT implementations.
 
Said premium consoles, not xbox consoles :)
It just doesn’t run well on the base ones, like many other modern graphical intense games, as DF stated in the video.

I then said it’s good we get new consoles since most own base units and don’t want to upgrade to a premium a year prior next-gen.
Yea, saw what you said.
But I've been seeing that it's not doing very well on the 1X, which I've just seen this breakdown highlighting it also.

Most own base console, but can already upgrade to premium one which you said played it well. Which would mean wouldn't need new consoles.
Not that I think that's reason not to release one personally.
 
Most own base console, but can already upgrade to premium one which you said played it well. Which would mean wouldn't need new consoles.
Not that I think that's reason not to release one personally.

Yes, but like myself, i won't be buying a Pro as i'm having the base ps4 now. It's too close to next gen for me, il shell out for a PS5 by then.
 
  • Like
Reactions: Jay
Status
Not open for further replies.
Back
Top