Raytracing won't simplify AAA real-time rendering

Those are pretty different concerns tackled by different groups of people on AAA projects. The art requirements can be an order of magnitude larger than the gameplay side of things. That doesn’t make the gameplay side of things easy.

I've worked about half of my career in the game industry. I've practiced TDD and written automated tests (and frameworks) for desktop, web and mobile apps. Some of those have been in the medical industry where the testing is crucial. I say this to make it clear that I'm familiar with solid software engineering practices.

With that in mind, games are the hardest software I've encountered for writing automated tests. It's just notoriously difficult to do in an effective manner. It's not impossible, but it's incredibly difficult.

Also, Unity games are usually written in C# (I think it's quite safe to say that - overall - most games are not written in C++ but in C#), yet I've seen no data so far which would indicate that Unity games have any less problems than games written in C++ during production and after release (if anything, the opposite seems to be true, not for technological reasons, but because Unity is so much more beginner-friendly).

I'm no fan of C++ either, but blaming a programming language for bugs and quality problems without any counter examples at hand is a bit ridiculous.

You'd be surprised. Here's a talk from Croteam on how they test their games and engine: https://m.youtube.com/watch?v=YGIvWT-NBHk

I'd wager all major engines are exhaustively tested.

Trouble is, there's combinatorial explosion of game state, user input, assets, scripted behaviour and engine, so there's a huge area to cover.

1) you can only test-driven-develop so much in games, and that line usually stops at the engine level because the game itself is in flux so much. Automated testing is confined to making sure that checkins build on every platform. Game dev engineering is fundamentally different than programming in other fields because the goal posts move constantly.

2) If an engineer writes code expecting there to be no more than 1024 physics objects in the system, tells the designers and artists this, but then they turn around put in 2000 colliding pieces of silverware on a dinner table "because it needs to feel like a big feast" is that an engineering problem, or an art and asset management problem? Because something like 80% of my bugs are shit like this.

3) Professional game codebases use their own styles (i hesitate to say dialects) of C++ that do the things we need them to do the ways we need to do them. We don't use anybody else's framework; all of the bonus stuff we're doing lives in macros that can be inspected if an issue arises. But, please don't push your language purism on anybody else. What a tired argument to have.

Tooling for 3D modeling/texturing/rigging/etc is significantly more complex and powerful than it was 20 years ago, yet Pixar doesn’t need fewer artists for a movie today compared to Toy Story - in fact quite the opposite.

AI techniques useful to artists will get folded in the tooling and enable artists to make even more detailed/complex games & movies, but that doesn’t mean the AAA games of 2030 will require fewer artists.

However, talented small teams will likely be able to leverage them to create things that would have been inconceivable from a small team a decade ago.

0. https://en.m.wikipedia.org/wiki/Jevons_paradox

The idea was that your small indie team could keep up with the big content demand since you had algorithms generating seemingly evergreen content for your players.

What happens in practice is that the players begin playing your game at a meta level, learning how the generative process itself works; effectively removing the benefit of procgen content. As an example, consider Spelunky which claims it can generate millions of unique caverns to explore. If you assess that claim visually, it's true. But watch the streamers play and you will see that they 'speak' the algorithm. By observing the shape of a cavern in one area of a map, they know the algorithm had to make a specific concession elsewhere in the map. So this content isn't really procedural for them anymore.

Even if the "AI" content generators get more intelligent, it won't free up resources. Mechanically interesting content, as demonstrated by the indie games of the last decade, is just a different form of handmade content. A designer hand made the procgen algorithms. Players ultimately bond with the designer(s), no matter what meta level they build the content at. In a hand-built game you might start to get a sense for where the designers hide treasure chests. In a procgen game you learn the algorithms themselves and how to predict and abuse them.

There's another form of game content, story content, which remains an edge for humans. Algorithms, or "AI" if you must, can't compete here. It would be the same as waiting for AI that can write award winning movie scripts.

The marriage of story content and mechanical content is a superior game-making formula to the procgen approach. So the _some point_ you refer to is probaby far away still.

Naturally, one may wonder if this has resulted in the labor market disappearing for artists. But it hasn't. The demand for high quality art skyrocketed because the tools economically allowed for it. (In some ways the market for artists has gotten smaller, because there is less appetite for low-skill Photoshop monkeys, but highly skilled technical artists are much more in demand, because we are starting to see dramatic differences in productivity between artists just like we have seen in engineering.)

So, yes, we will see AI in content generation, but it isn't going to be in the form of replacing art bottlenecks. It will manifest in new tooling for artists, who will become more productive, and there will be even fewer artists who have good mastery of these tools, and the demand of quality will increase further. Which would lead to similar bottlenecks to today, though I believe artists will be paid better, and it will become even tougher to break in.

Raytracing is being used as a synecdoche for technological advances in rendering. The point is that as off-the-shelf rendering improves, the advantages of a vertically integrated engine diminish.

> Off topic, but this is not a good analogy. The "cloud" is just a network of mainframes, which means mainframes are arguably the dominant form of computing (and will become more dominant as thin clients, like Stadia, rise in popularity).

> Mainframes were not necessarily purpose-built, unique machines. They were instead defined by their sharing model, which is where the term "personal computing" comes from -- as a contrast to the mainframe/server model.

That's a very essentialist perspective; there are a lot of ways mainframe computing differs from typical personal computing and no single one of them is definitive. As a programmer, a mainframe offered you a reliable computing environment, because they're built to be highly available from the hardware up. That's very much not the case with the cloud, which follows a worse-is-better approach where your jobs can be terminated whenever and you're expected to handle it. While shared computing resources may be making a comeback, I don't think that kind of ground-up high availability will, so while the cloud has some aspects in common with mainframes, it's very much not the same thing.

If anything that tells me they may have been better off investing that developer time into the game instead of a custom engine.

p.s.: I share your opinion

Sure - those are the AAA games we're talking about (and even then, a lot of those are long-running franchises that reuse a significant chunk of engine code between multiple installments). The biggest sales numbers will keep coming from that segment for a long time. But it's a shrinking segment in terms of overall revenue (much of which is no longer up-front sales), and even more so if you look at profit; each year the line where it makes sense to use a custom engine gets higher, and a bigger share of money coming in is from games built on commodity engines.

laughs in light fields

There's an upper limit to what's useful - the real world. If our minds can't comprehend it, you've hit the upper limit of practical use. Once in game graphics become literally indistinguishable from real life then we'll start to plateau in terms of complexity of "tricks" and raw computing power will be able to catch up.

Way too slow for VR, try 5 ms just to be sure.

https://www.starwars.com/news/the-mandalorian-stagecraft-fea...

It’s very common for film CG frames to take a few hours but not usually 12 hours. You need to be able to start rendering jobs at the end of the day and have them ready for dailies on the morning. Jobs that take more than 4 hours risk clogging the queue and taking 2 days per iteration rather than 1, so people are motivated to keep it reasonable.

Another important difference between games and film is antialiasing techniques. Much of the difference in render time between them is in using fancier texture sampling and using lots more samples than games are willing to use or even need for decent quality.

Text rendering used to involve complicated hinting and then subpixel rendering.

These days on Retina screens you don't need hinting or subpixel rendering.

You just render it like any set of curves, no longer any tricks about it being a font, because it's hi-res enough.

What rendering tricks are you suggesting are being added rather than being taken away, for fonts?

But even though parallelism can be increased, the gains from them are limited as per Amdahl.

It's also possible to fix transparency ordering in rasterization using OIT and there is some hardware support for it (ROV), but it's also expensive and can be hard to do right so many games choose rough sorting

Their big problems is trees, since leaves modeled as flat squares with their the actual leaf shapes in textures (that are completely transparent around the leaves).

"Trees are our biggest problem in ray tracing. No matter what we do they're by far the slowest thing to trace, and they're much worse than on rasterizer, and if you have any ideas of how to ray trace trees efficiently, we're very happy to hear them. We've asked a lot of very smart people and no one has come up with a good answer yet. You can choose between geometry and alpha testing and both are kinda crap." - It Just Works: Ray-Traced Reflections in 'Battlefield V', GDC 2019.

Raytracing helps make realistic art possible. But it still takes a great artist to do it. There are so many subtleties to the way common surfaces reflect light that are hard to capture, even once you are given all the sliders you can dream of. My experience dabbling with Blender, was that it can be quite frustrating: you KNOW what you have set up doesn't look quite perfect yet, but you can't quite express WHY or what you need to change in the material definition to fix it. So you end up settling for "ok", meaning the resulting render looks pleasing -- but it still looks like a render.

But I guess that's why I'm not an artist.