New Genres

[NotAI]

Thinking that people are not using modern hardware as much as they could ... and new genres of games are possible when even ordinary consumers have like ... 24 cores to play with.

Even my work *laptop* has 16 desktop cores at this point.

Hardware materialism ... the theory that imagination concerning mechanics depends primarily on what can implemented on a typical machine at the moment.

I'm pretty sure indie can make all sorts of new genres of rpg that rely mechanically on being able to waste a good 10 cores at least.

Edit. Like spell creation mechanics that could NOT have been put in until 2023-2024? (Something that might require a lot of concurrent brute force checking of optional conditions while activated?)

Edit. Some kind of totally different XP system or making levels procedural and dependent on the state of the game world?

Why isn't stuff going wild?

[KnightoftheWind]

CPUs are so good nowadays that you wouldn't even need a Graphics card to render great visuals. The issue is that most new titles don't have a software renderer whatsoever, save for a select few obscure indies that I am aware of. Not that any of the diversity hire "programmers" are talented enough to implement one. You could code a game in pure C and port it to any platform you can think of, but devs and publishers never even consider that.

Thus the CPU sits mostly unused, with awful enemy A.I and dull mechanics.

[Klerik]

I can run 99% of new games out now with 4 threads at around 3.6 ghz and most of the slowdown will be because some art student took a shader brush and tessellated the **** out of a tree behind a wall somewhere so they could get that NVIDIA sponsorship to sell new graphics cards.

And by game I mean mental illness tech-demo.

[Humbaba]

Is there such a thing as "coding talent"? Are there clever tricks a talented code monkey can employ to improve performance?

[Klerik]

Here's an idea.

Diablo 2: Part 2.

[rusty_shackleford]

CPU performance isn't significantly better now than it was 15 years ago, we just have more of them.
You can see this in action by playing some games made in the mid-to-late 00s which still tend to run like ****.

Many problems in gaming aren't trivially parallelizable, and those that are, are usually offloaded to the GPU already.

But to answer your question(?), simulation games are an obvious choice. I mean real simulation games like Dwarf Fortress or Noita.

[J1M]

Is there such a thing as "coding talent"? Are there clever tricks a talented code monkey can employ to improve performance?

Yes. TLDR: Ignore everything taught in university and use arrays for everything.

[rusty_shackleford]

You can see this in action by playing some games made in the mid-to-late 00s which still tend to run like ****.

A good example is Gothic 3, by the way. Even though you can turn the simulation radius up further than 5 feet away from you now, it will bring any PC to its knees even at medium distance. Makes the game really feel alive though, surprisingly fun with the right mods.

Yes. TLDR: Ignore everything taught in university and use arrays for everything.

Cache is king.

[Ranselknulf]

The only effective way to increase cpu performance at this point is with more cores.

The diameter of an atom is in the range of 0.3 Nanometers (NM). Current best cpus use 3 NM transistors and I think there are 2 NM architectures in the pipeline.

Some ******* jack chinese physicists have made transistors on the size of 0.35 NM already, but making cpus and **** with that is years away still. Things can't really get much smaller.

At a certain point, computers either become the quantum meme, or you just start adding more cores. We are basically at the point where just adding more cores is the easiest way to increase performance, and AMD seems to be ahead of the game in this regards compared to Intel.

[GhostCow]

Single core performance is still going up plenty. Zen 2 to zen 3 was a 20% uplift in single core

[rusty_shackleford]

Single core performance is still going up plenty. Zen 2 to zen 3 was a 20% uplift in single core

It was previously doubling about every 18 months, this is what's implied when people say "Moore's law is dead", but they're actually referring to Koomey's law & Dennard scaling.

[2factorauth]

Thinking that people are not using modern hardware as much as they could ... and new genres of games are possible when even ordinary consumers have like ... 24 cores to play with.

Why isn't stuff going wild?

Why try interesting stuff if you can be a pixel graphics indie making a walking sim?

Also, there are limits to parallelization. The interesting and important stuff is all all single thread or otherwise requires substantial architecture.

[NotAI]

The correct model can make software a lot more concurrent than most people think. For graphics the bucket brigade plus purely functional programming and arrays was the main thing.

For gameplay, the thing is to make designs that are more loosely coupled. Even though that means a lot more overhead. With today's compute that overhead is okay. For gameplay that is even more kitchen sink and emergent. Simulation, exactly.

Specifically, the less exact timing matters, the easier it is to make semidecomposable. A fighting game is harder to make concurrent than a big, complex RPG. Turn-based easier than real-time. Real-time easy too, with a trick, like if instead of dying, when HP goes to 0, a last chance 3 second shield pops up and some animation and HP reverts to 1, once per fight.

Agree that single core performance is also increasing, because designs are still improving. Too much of chip design is just due to whatever automated design software was used previously. So almost always plenty of room left to improve. Beside that, giving every core its own cache is the next big thing to speed up per core.

[KnightoftheWind]

I think the desire for computer hardware to increase in power year-over-year is just another form of consoomerism. And this is helped by incompetent devs who code bloated and unplayable games that require the latest and greatest to even run. And assuming that CPUs never increased in performance from this day and onward, it would still be more than enough for a wide variety of use cases. If a modern software renderer was made that took advantage of all the cores we have now, the games that used it would be more than sufficent for the vast majority of people. No GPUs required.

Computers really were faster, when they were slower.

[rusty_shackleford]

functional programming is a meme pushed by people who end up cutting their penises off

[J1M]

When denuvo eats 40% of your CPU...

[2factorauth]

functional programming is a meme pushed by people who end up cutting their penises off

monads gonads

[2factorauth]

The correct model can make software a lot more concurrent than most people think. For graphics the bucket brigade plus purely functional programming and arrays was the main thing.

For gameplay, the thing is to make designs that are more loosely coupled. Even though that means a lot more overhead. With today's compute that overhead is okay. For gameplay that is even more kitchen sink and emergent. Simulation, exactly.

Specifically, the less exact timing matters, the easier it is to make semidecomposable. A fighting game is harder to make concurrent than a big, complex RPG. Turn-based easier than real-time. Real-time easy too, with a trick, like if instead of dying, when HP goes to 0, a last chance 3 second shield pops up and some animation and HP reverts to 1, once per fight.

Agree that single core performance is also increasing, because designs are still improving. Too much of chip design is just due to whatever automated design software was used previously. So almost always plenty of room left to improve. Beside that, giving every core its own cache is the next big thing to speed up per core.

None of this is easy though.

When you look at Dwarf Fortress, single thread bottleneck.
CDDA? Single thread bottleneck.
And they dont even try to make something interesting and those are zero gfx pure algo devs.
Factorio? A little bit

[NotAI]

It's much easier to remove the single thread bottlenecks than people think. In those kinds of pure algo games for sure - they don't have to be single threaded.

Like adding a little bit of randomness and then making approximately the same code as before tolerate that...also usually allows decoupling most things from most things.

Yet another way is refactoring the core loop into many processes that have certain conditions upon which they send a message to each other and get updated in light of the state of the rest.

Message passing is not the only way of implementing for large core counts. Nor is avoiding side effects except at invariant. There are a dozen ways.

Only catch is they are not the default styles of most developers. It's easy but expensive because people switch styles en masse if and only if they are paid to do it by their place of employment or else the other method is 100x more productive. Well, most methods are more productive. But not 100x more. AI prompting is the recent example of something being 100x and people switching to that. But nobody switched yet to Smalltalk-likes which is 10x or around.

[somerandomdude]

The number of cores on a CPU doesn't matter if games are unoptimized **** that won't utilize them properly. And this is more often the case than not.