Hacker News Reader

Comment by db48x | Hacker News Reader

He is partly correct, but forgets about the polygon budget. The GPU is fast but not infinitely so. It has lots of texture memory, but not an infinite amount. Game developers have to balance graphical fidelity against the rendering time. They actually did a really bad job of this for Cities: Skylines 2 and it was a big reason why the game was so sluggish on release. Search for the story of the teeth for a particularly salient example.

Also that grand staircase should be something the player chooses to place, rather than something you get every time a path is sloped.

View on HN

I was not familiar with the teeth thing. Link for the lazy:

https://www.ign.com/articles/cities-skylines-2-dev-addresses...

Reminds me of Dwarf Fortress. I hear the individual teeth are simulated so they can be lost in a fight and such.

Yes, except the DF doesn’t compute shadows for all the teeth. Or texture map them, or use a shader to make them look wet and shiny.

Still, if they get knocked out during a fight then your dwarves have to carry them individually to the gravesite afterwards, so it’s not like you should automatically imitate DF when you write your own game.

Will the dwarf's friends and family get upset if any tooth is left unburied? In an evil biome, will a forgotten tooth resurrect and cause trouble?

Teeth cannot rot and therefore won’t cause a negative thought if left unburied. Teeth also cannot GRASP, and therefore cannot be raised by a necromancer as an undead. A skull can be though, with or without the teeth.

^ and this is why interesting composition of systems is so critical in generating games, <3 df

Most of the changes replace one model with an equally detailed one, or add a bunch of decals that are cheap in modern engines with clustered rendering.

Adding props is a bit more expensive, but they can be very low detail and still look good from a distance. Something small like a newspaper or plastic bag can be less than 30 polygons with a 64x texture.

Nobody's asking for ray tracing and complicated shaders. Just small tweaks that make a city look more natural.

That's just demonstrably not true, the foliage alone adds a lot of polygons, so do the stairs.

Can you render low fidelity stairs as ramp with normals/shadows baked in?

When the player zooms in to street level on a building it's worth the performance hit to load in more details and do so in a better looking way. When it's zoomed out to view large portions of the city (or the whole thing) then sure, you have to trade off graphical fidelity for performance and usability as the user builds out the city and wants to watch it grow. The amount at a given zoom should be based on the graphics settings, and low should allow for low detail even when zoomed in.

LOD (level of detail) techniques are used by practically every game ever written. It doesn’t mean that the developers can ignore the poly budget.

You are correct but what the article is asking for is nothing out of the ordinary, in fact mods already do what the author is asking for, just not in an automatic way

And how are those mods on baseline performance? Are the mods appropriate for general audience mid range rigs? The tech industry has the capability of pure ray tracing. Game engines don't use it because the audience willing to shell 10k for PC to run it is exceedingly small. They have to target mid range specs to have an audience big enough to make money.

The solution is mods by modders for modders.

Do you usually interpret “nothing out of the ordinary” as “super computers only” or is that just for me?

I think you may have misunderstood me.

LOD techniques are commonplace. Cities: Skylines and the sequel both implement them. Telling us that a game developer can use LOD to fit more graphics into the same poly budget is not telling us anything we don’t already know.

And the problem still remains that all the graphics must fit within a certain poly budget, usually whatever can be rendered at 30fps on the minimum spec hardware. Adding features to every property in the city to represent weeds, fallen leaves, broken concrete, grime, wear, etc all adds to the rendering cost of the game. It all comes out of the same budget, even if those features are invisible at longer distances. Worse, that cost is multiplied by everything the user has created in their city. The more such features the developers add, the smaller a city the user can build before the game’s performance drops to unacceptable levels.

Anyway, CS and CS2 already go to great lengths to add soul to the game. Every asset has defined locations where props can be placed. One house can have a swing set while the identical house next door has a pool, or toys, or a car, or shrubbery, or any of thousands of other possibilities. The author is just complaining that these props do not make his cities look dilapidated enough.

His most valid complaint is that the games don’t do much to obscure the exact property lines, which are always perfect rectangles. Perhaps in the future one will, though it will still have a budget to fit within.

It's quite possible to do all that. Most AAA titles do by now. But creating all that detailed content at multiple levels of detail is a big job. GTA VI cost between $1.5 billion and $2 billion to produce. When you do those big zooms, vast amounts of data has to be shuttled in and out of the GPU. VRAM is too expensive to keep it all live.

There are some other current problems. GPU prices have not come down. If you want a big, high-detailed world, you need a few hundred dollars of GPU. GPUs on laptops tend to be underpowered, and if you make the GPU work to its full capacity, it runs down the battery.

One of my interests is technology for large metaverses with high detail and good performance. Everybody who made metaverse noise during the boom failed at that. Second Life, which is a large metaverse with good detail but slow performance, is about as good as it gets. (It's been improving recently; they've hired some good graphics developers.) Improbable's Morpheus/M-Squared [1] looked promising, but the site hasn't been updated in a year and the Discord community seems to have died. They usually run with "cloud rendering" (each user ties up a dedicated gamer PC in a data center), which is too expensive except for special events. Roblox is probably the only other serious US player working on this. Unreal Engine 5 can do big, static worlds that have lots of instancing, but that's achieved using large amounts of preprocessing, optimization, and performance debugging.

The underlying problem is that good simulated cities are about as interesting as real cities. Kind of boring. Second Life has a terrible problem with new users. At the start, you're in a big world the size of Los Angeles with about 40,000 people spread around. You don't know anybody. You have no idea what to do first. There is no obvious goal. There is no built in game. The system itself is completely indifferent to you, like real cities. You can explore the huge user-built world freely, but only when some users organize an activity does much happen. There's a lot going on, but it's hard to find. Those seeking passive entertainment give up quickly.

[1] https://docs.msquared.io/morpheus-platform

Culling details smaller than a pixel isn't even sacrificing quality. It's literally just what rendering is. That's why you don't render a sheet of 10^23 individual atoms.

Much of what is suggested could decrease texture memory usage. If you add a masked dirt material on top of a surface, you can often decrease the resolution of the texture of the base material underneath thanks to the added detail.

As well as the teeth, I understand they had issues along the lines of using preview rendering tech in unity that then was not further developed and dropped from the engine? Causing them to need to write a fair amount of render stuff themselves? .. something similar to that at least I'm sure.

There were, and still are, other problems as well. The teeth were just the most obvious example.

This seems like the logical explanation, it still doesn't run that great even on GeForce Now

I would say the game should generate some kind of a grand staircase every time a slope is the right shape, but let the player replace it if they wish. Would be even better if the element was chosen based on zoning policies, local ordinances, land value, etc. Soul by default, but infinitely customisable.

Reply on news.ycombinator.com