Excellent points all around. Thanks for making a great thread Zaku
- Game development cycles have become disastrously long - games that are in development for 8 years are no longer a rarity, comparing the previous output of Rockstar Games to the modern one in quantity is a sad joke
Unfortunately Games are high complexity to make. Acerola did a good video on
why game development is hard. Programming is already high complexity and many programmer jobs are really just complexity management jobs. Additionally, many games do not have good unit tests or E2E tests(often for good reason! E2E tests are long to run, and can run into issues if your game features non-determinism(many games using physics has this issue), and also for the fact some things will pass tests, but be obviously wrong when manually testing it, i.e. sounds too loud, too many overlapping, visual weirdness etc). Add in that graphics programming(shaders etc) has afaik, no good debugging approaches, and you begin to approach why the programming part is hard. Then if your engine has limitations you either need to work around it(compromising your design vision and taking a lot of time), extend the engine(which means more complexity has to be managed and you have to get comfortable with the harder work of engine development, and you have to make sure it still compiles for your target platforms. I think it's possible for Unreal as it's source available but I may be wrong) or get in touch with people who can fix these issues upstream for the engine(for the average developer... unfeasible for Unity and Unreal).
This is also in an iterative loop. Each person added into a team requires more time to communicate between them, which extends the time for development and reduces the number of iterations that can be done. This is also part of why level designs have simplified over the years, as it becomes communication between narative designers, environment artists, level designers. game designers, and developers. The more communication that needs to happen, the longer and more difficult each development iteration takes, and also the more money needed to pay for it.
That said, for many games I think it's also a case of diminishing returns. But making the most of the diminishing returns, and knowing what to leave alone and instead get their work finished is... another challenge. Like they say, the last 20% of the work, is 80% of the effort.
Additionally you're looking at things from a quantity point of view. The end consumer doesn't care about quantity... in fact some would see it as a mark of it being shovelware. This happens on steam where a lot of new games are shovelware. Consider when was the last time you looked on the playstore for a game? For me it was more than a year ago because of the sheer quantity of shovelware on it. Quantity is not quality- it's also part of why I don't really go looking very much or very hard for games on itch.io either. For good or worse, Steam is very good at curating games players are interested in.
- Budget creep essentially means that if your game fails, so does the studio
This is sadly true. The only real "polyfill" for it, is contract work, or trying to fill the gap either with early access(completely unsuitable for some game types like puzzle games), kickstarters(completely unsuitable for any developer who has no track record), or to try and get money from patreon-style things, or even just to try and make game development courses or asset packs. Arkane Studios had a huge gap between Arx Fatalis and Dishonored, that was filled in with contract work for Ubisoft and then later for Bethesda.
The other choice is to make games that are the spiritual successor to Oblivion Horse Armour DLC. I understand why they do it, but every developer who does this instead of just making good games, should quit their job and go join a Fortune 500 Company or an investment bank as they're clearly in it for money, and those are much better ways to get money. I don't think making money is bad, but you do yourself a disservice if you try to get it in bad ways by ruining things.
- More and more focus goes towards the finetuning of graphics, leading to less overall content, area size or map diversity as each individual section takes longer to create
This is true, but consider it a different way: your graphics is the first thing people see when they play your game. Sadly games with an excellent game design will be ignored if they have interesting or appealing graphics- with the exception being that the game design keeps people addicted in a "just one more" loop. Factorio does this, the graphics are a bit off putting for new gamers. Vampire Survivors also did this.
Also regarding your map diversity and area size point... it's also because of the
tooling changes. A lot has moved to mesh-based systems, which means your level designer has to know a 3D modelling tool to do his work, instead of something like a brush-based editor like Trenchbroom. This is harmful for them as it hurts their iteration speed, and harmful to the game as 3D modelling software often doesn't have ways for managing entities in the level. It's also really bad for beginner level designers as they can't dip their feet into it all without having to adopt 3D modelling, scripting and design hats. Plus their choices(see what I say about overlapping light volumes and shadows) affect performance in the game, and they're not really placed well for dealing with performance as they're not programmers.
- Game optimization goes out of the window: file sizes have become bloated and even become a selling point - a game that is 100 GB big pretends to justify as much under the illusion of "more content" and dominates the free real estate that is your memory space, becoming evergreen content and pushing out the competition
Game optimization doesn't go out of the window... The file sizes are largely down to either baked lighting for large maps, or for the various texture types that each texture has(Each of them having 2048x2048(often both high and low resultions too) albedo, roughness, AO map, emission sometimes, normal maps etc...). The reason to have lower resolution textures is to reduce VRAM costs when they're used by shaders. The reason to use baked lightmaps is to reduce GPU costs on the lighting, as Global illumination is a hard problem. Additionally a lot of the issues often come from unoptimised lighting- shadows are very expensive, overlapping lights are expensive... and so on. The 2 choices here you get are... either bundle higher resolution textures as DLC(as Monster Hunter World does), which also means you accept people are gonna make your game look like shit in youtube videos and turn people off of it... or force people to download it all at high storage cost.
Imagine if your first exposure to Skyrim was this, naturally from a youtube video or a livestream or a game news article...
You would probably be turned off if you're the average consumer...
Now if you really wanted to talk about Game Optimisation going out of the window. I partly blame Unreal, and partly blame Developers who have bought into ideology around game development rather than pragmatics.
Threat Interactive have done a bunch of videos on how Unreal is... culpable for this.
Between storage space and game performance, most people prefer game performance, but the cost in asset sizes is ridiculous. I mentioned above about the textures, often many games will channel pack a lot of textures together into one(i.e. Alpha not being used by Albedo, but instead being used as a 0-1 for the roughness map). Many game engines and developers don't do this anymore, understandably because it's time consuming to do... but at a ridiculous size like 100GB games, compression efforts should be made. Unfortunately they're not made because they know consumers will suck it up, and it lets them get a lower quality product to market quicker.
DLSS is a neural net model that make use of Nvidia's post RTX2XXX cards hardware for upscaling game graphics, achueving the goal of increasing resolution levels at a lower computational cost beyond certain levels (pretty much 4k+ at this moment)
Slightly better than TAA, but still makes everything blurry as you're upscaling from a lower resolution. DLSS uses previous frames to help in upscaling the image for the current frame, but as a result gets ghosting from previous frames and often adds detail where at native there would be none, or often smears over details where there would be. For anything at or below 1080p it's useless in my experience, and most midrange machines still target 1080p.
Over 50% use some kind of 1080p, see Steam Hardware survey. Above that you're getting into situations where it's *probably* useful but I don't use anything at or above 1080p yet... so I can't personally attest to it either way. Anyone else running with 4k monitors want to weigh in here?
Oh, and to add to the previous point, it's entirely possible with diffusion or other AI-based rendering techs, 3D modellings and shader coding will be completely unnecessary as all it would take is a basic depth-based scene render without textures or even need for hi-poly models, which is then rendering using a controlnet-like extra-model on top, for rendering the scene in any styles the devs/players desires.
If I understand correctly, you're basically saying having a bunch of rigs in a scene and then having an AI neural network plaster its image over it. What I've seen of it, works fine for video and image... but I'll believe it when I see it for games. Even still.. the idea of giving players the control of the style, is imo, in the realm of modding. Most players will not want to spend their time configuring how the game looks in something like reshade, so why should we expect it in a game? That said, some post process effects like pixelation and so on are nice to see settings to turn off(and you see it in many PS1-style games). Additionally giving this kind of control, is corrosive to the brand recognition of the game as people won't know it's the same game. Look at heavily modded Skyrim and compare it to regular Skyrim and tell me it looks like it's the same game. Not saying it's bad, but it's a hard sell. Maybe better for some kind of AI Generated Garry's Mod sandbox style game? To me, a big question is persisting details temporally and spatially(these are games, you can move in space, and most images and videos aren't concerned with this).
Oh, I almost forgot the biggest issue. Stable diffusion on my RTX 3070 takes quite a while to generate a single image from a prompt for a relatively small sized image. This would need to work in real time, at 60fps, ideally 120fps. Until you can generate images of high quality, native(or near native) resolution every 8ms, it's simply out of the question in the same way Raytracing is out of the question. This 8ms of computation has to be shared with physics systems(sometimes sharing GPUs), sound, I/O, AI, etc... All of these issues, is why it's one of those things where I'll believe in it when I see it, because I think the hardware has plateaued quite a bit.
If you're wondering what a market looks like that already has hit the performance limits of what a regular end user can notice, look at SSDs. PCIe 5.0 drives have been around for a couple years now, with very little hype around them, just because 14GB/s in sequential transfers is pointless to the average person. It's faster than almost anyone's internet connection, and few people are constantly copying huge files from one drive to another (if you did that often enough to care, you'd wear out your drive). It doesn't help that those drives are very power-hungry and run hot.
This is a good point, but I think users are noticing that games in 2017 look better than games now. DOOM 2016 looked phenomenal, while most games nowadays look like they are blurry and smeared to shit. What's worse is that UE5 being the default engine a lot of companies are using, Unreal designs their engine so many of their graphics effects rely on TAA or else you'll get graphical artifacts. The hardware improvements do exist, but they are just being wasted more and more, and due to the complexity of game development(see my first paragraph at the top) it's hard to manage that complexity let alone optimise on it. If you're using UE5, you're gonna naturally follow UE5's recommendations
I think it's also a matter of education. Many people don't know how to describe why their game's visuals look worse than the games of almost a decade ago. Of course what looks worse, is up to taste, some people like blurry shit like motion blur and all that. I utterly detest it. I also think there's a major quality issue nowadays. Too many developers say "good enough" when it's not good enough. I talked a little bit about this in my article on
reflection approaches. The most blatant example of good enough not being good enough, is in how badly behind video game sounds are. Most games do either mono sound or if they do stereo, only do the most basic panning between left and right ears. No sound reflections, no audio occlusion or absorption etc. Once you start looking at tech that's not Graphics programming you'll quickly see how woefully behind a lot of different tech is(and also how a lot of consumers don't see it or it's paved over graphically. Hell... clipping and collision issues are still very common, and they really shouldn't be so common because it's possible to find and define the
walkable environment).
Ultimately part of the issue is idiot consumers buying and not refunding shit games that have major flaws... and also games companies that become publicly traded, as the audience of their products is no longer gamers, but investors. Every publicly traded games company goes down the shitter. For Nvidia you see this already with how DLSS is rubbish at 1080p or below, and how expensive the RRP for GPUs have gotten the last 5 years and how they are marketing to AI and Cryptobros. Nvidia's audience is not gamers, but investors.
Apologies for the long post, it's something I care about, so... deal with it
I believe this is a question that has become more and more relevant in recent years as graphical prowess progresses but the overall results stagnate. We have long since reached the era of diminishing returns with increased polygons and come the PS3 era, I believe we have attained a plateau on which we can build but that we can't really elevate or surpass. PS4 is PS3+1 and PS5 is PS3+2 so to speak.
