Redefining Console Generations

Typically, game console generations have been defined by release date, arbitrary tech specs, or marketing needs. This commonly leads to overlapping and conflicting generations, with no standard other than vibes.

I propose redefining the console generations to reflect important technical milestones.

• Generation 1 - Fixed function hardware
• Generation 2 - General purpose CPUs
• Generation 3 - Scrolling tile maps and sprites
• Generation 4 - Multiple planes
• Generation 5 - Fixed function 3D rendering
• Generation 6 - Configurable rendering
• Generation 7 - Fully programmable shaders

These not only reflect important changes in hardware, but changes in gameplay and graphics design philosophy.

Things that don't matter:

• Storage Medium - Mask ROM, discs, tape, flash, HDDs, data gets into the cpu all the same.
• Release Date - Forward thinking visionaries and laggards abound in this industry. An old console design delayed by two years doesn't enter the next generation.
• Brand Line - A new product from the same company doesn't mean a new generation, no matter how much their earnings call depends on it.
• Arbitrary Tech Specs - AKA the bit wars. A system is more that its bus width or instruction length. After the questionable returns of 64-bit, systems quietly went back to 32-bit processors while kids on the playground called them 128-bit.
• Game Quality - Good and bad games can be made under any paradigm.
• Deliberate Crippling - A late generation system with non-removable software doesn't go down to generation 1 just because they system is now 'fixed function'. Welding the doors shut on your car doesn't turn it into a horse.
• Code Ninja Shit - A clever programmer or time traveller from the future can make a system do almost anything. What we're interested in is the standard, expected mode of operation - the thought model of the system.

In the beginning, there was:

• Fixed Function Hardware

Purpose built systems that do exactly one thing. They cannot be reprogrammed. Timing is determined by capacitors charging, logic is driven directly by transistors, the physical components are the program.

This generation moves from hardware-defined logic to software-defined logic. Key features:

• Programmable CPU
• General Purpose Display

Video signals are memory driven or beam racing. Sprite and tile based accelerators may exist, but there is no scrolling or camera position. The mental model is frequently similar to a board game - there is a background with things on top of it. The core idea is that a cpu draws things to the screen.

Some of the earliest games here pre-date their "first generation" counterparts. This is because they were developed in research environments on equipment that cost millions of dollars.

This generation features hardware acceleration for:

• Sprites
• Tile Maps
• Scrolling

These are machines specifically designed to go beyond the early single-screen arcade style games. The world is now made of a pattern of tiles and we're setting a viewport into that world. As the beam scans down the screen, dedicated video hardware references the tiles and sprites as they are needed.

A refinement of the 3rd generation with the key addition of:

• Multiple Background Planes

Limited scaling and per-scanline scrolling adjustment can give a pseudo 3D effect on some systems. Some titles may repurpose a tilemap as a low resolution frame buffer for arbitrary rendering, but the systems are still built for layering up 2D planes with scrolling tiles.

This generation completely changes the rendering paradigm, adding:

• Frame Buffer
• Polygon Pipeline

Instead of referencing a tilemap as the beam scans down, we now have a list of objects that is drawn into the frame buffer every single frame. Polygons are rendered with fixed function math, sometimes with limited multitexturing capability.

This is a transitional generation where multitexturing and math options become complex enough to give shader-like outputs, but would be more accurately be described as configurable than programmable.

The era of standardization:

• Fully Programmable Shaders

You are now executing a program on every pixel of the screen. The interface model supports everything, doing more is just a matter of execution speed.

These systems also all have some kind of high speed local storage. This is more of an incidental addition than strictly necessary. There were base models of the XBox 360 with no hard drive, which an annoyance but not a game breaker. Much hype was made around the PS5's SSD, but seamless worlds can work at almost any data rate, it's a change in quantity not in kind.

Some may make a distinction between early pixel shaders and later compute shaders, where non-graphical elements can be calculated on the GPU's vector processors, but this is mainly used for special effects like physics simulation and does not seem to change the type of game being made.

That's it. We're still in generation 7. Everything is essentially the same. If you set out to make a brand new NES today on the latest 3nm process node, you could probably clock it up to 5GHz, but it could never run a modern game. But if you build a PS4 on the latest process node... well you're just building a PS5.

Attempts at new standard formfactor consoles like the Ataribox, Intellivision Amico, or Coleco Chameleon - these were all scams or delusions. We're well past the era where you can pull different chips off the shelf and get some new graphical capability. Everything is a CPU with a GPU attached that runs the shader/fragment model and the market is liquid enough that price directly correlates with performance. So when you say "we're making a new console" what you really mean is "we're purchasing an SoC, putting it in a box and artificially limiting software to run on it".

Consoles as a concept only exist because large established players can leverage their market position to artificially lock off certain games. If you're not big enough to make your own artificially walled garden, you won't be making anything.

I believe real-time generative AI could play a big role in future games.

Google has demonstrated Project Genie which can turn any image into an interactive scene. I could see this working in an open ended narrative game - a choose your own advenuture where anything could happen - but I don't think it will be consistent enough for traditional gameplay driven games (and if it was, it would be a waste of resources). It would also be great for porn but they're never going to do that.

A more robust solution would be to use traditional methods to render a basic gameplay skeleton, then use AI inferencing to produce stylized graphics on top of it. This has the potential to bypass shaders entirely and make a game in any art style imaginable. It would also be evaluated in constant time - a large constant time, but it wouldn't scale up with the complexity of the scene. A field of grass would render in the same time as a flat table.

To make this cost effective would require new hardware, and the number of companies that could do it are limited and currently unmotivated to do anything consumer facing. It might be possible on current hardware with some as-yet-undiscovered software techniqe.

Here are some things you could argue about:

• Does the PAL NES port of Elite count as generation 5 because it renders 3D graphics to a makeshift frame buffer?
• Does changing zone height on the Atari 7800 count as hardware accelerated scrolling?
• Does the 16 stage GameCube TEV count as a programmable shader?
• Are compute shaders worthy of their own generation?

Discussing either side of any of these questions would be infinitely more intelligent than arguing about whether or not two 32-bit processors add up to 64-bit, or if we should draw a line in 1987 or 1989.