In the 80s and 90s, memory, integrated circuits and other components of game hardware were quite expensive. This expense forced hardware designers to make do with less, and they got creative with how they enabled games to display graphics on the screen.
Graphics were almost always defined using tiles, which were typically 8x8 or 16x16 chunks of graphics. A scene on the screen was built up using these tiles rather than specifying each and every pixel. If a game console has a screen resolution of 320x224 and each pixel's color can be chosen using a 16 bit color space, the system would need at least 140 kilobytes of video RAM to store a screen
<<320 x 224 yields 71,680 pixels, each pixel taking up two bytes (5 bits each for red, green and blue)>>
That might not sound like a lot today, but in the 80s it was significant. To put it in perspective, the Neo Geo has 68 kilobytes of video RAM1. To make matters worse, even if a system opted for the 140kb to enable each pixel be defined, it'd be quite challenging to update 71 thousand pixels 60 times per second with the processor and bandwidth speeds that were available.
Rather than specifying each individual pixel on the screen, early gaming hardware often used graphic tiles to build the screen. These tiles were typically 8x8 or 16x16 pixels in size. Coupled with the tiles is a tile map, which specifies which tiles go where to build up a scene on the screen.
<< graphic of tiles and a corresponding tilemap >>
Each tile only needs to be defined once in RAM (or ROM), and then it only takes one byte to specify which tile is to be used in at a specific spot in the tile map.
Rather than allowing tiles to specify exactly which color each of their pixels are, tiles are normally indexed. That means each pixel in the tile is usually stored as a number from 0 to 15. Then a corresponding 16 color palette is used to specify what each color index is.
<< graphic showing a palette, an indexed tile, and the final result >>
Palettes and tiles behave similarly to tiles and tile maps. You can think of the tiles being the "palette" for the tile map.
Most early game systems like the Sega Megadrive have very specific ways that tiles and palettes must be used. The Megadrive has two background planes and 80 sprites. Each background plane can be filled with a tile map, and each sprite is defined with tiles. The planes and sprites all share four 16 color palettes
<< blown out graphic of Sonic the Hedgehog showing the planes and sprites >>
When defining a background on the Megadrive, the developer must setup the two background planes to accomplish this. This sounds limiting, and it is, but there are advantages too. For example, each plane can be scrolled in any direction quite easily. It's even possible to scroll individual rows of a plane to accomplish 3D-type effects
<< bathhouse scene in SF2 to show line scroll effect >>
Unsurprisingly, the Neo Geo uses tiles and palettes to save memory. But unlike the Megadrive, it does not have any background planes. Instead, the Neo Geo offers the developer 381 sprites to work with to build their scene. All backgrounds, bullets, characters, enemies, etc on the screen are formed from these sprites.
This approach enables the Neo Geo to be more flexible than other systems, but with some tradeoffs. It is up to the developer to hand roll their own background scrolling system for example. As you get to know the Neo through programming it, you will likely come to find its sprites are both flexible and frustrating. Hopefully this book will help minimize the frustration at least.
A sprite on the Neo Geo is a single column of tiles. Sprites can be up to 32 tiles tall, and you are free to place them anywhere on the screen you would like. Building larger objects such as a screen filling boss or background is done by placing several sprites next to each other, forming a matrix of tiles out of their columns. The hardware offers aid here, by enabling you to "chain" sprites together and treat them as a single unit (this is also called "sticky" sprites).
A sprite can be scaled horizontally and vertically, and each individual tile specifies its palette and whether it is flipped horizontally or vertically. The graphics hardware even offers simple "auto animations" with sprite tiles where a sequence of tiles will be displayed one at a time, forming a simple animation. Whenever you see people in the backgrounds of Neo Geo fighters doing the same small movement over and over, almost certainly tile auto animations were used
The tiles for sprites are stored in the C ROMs. Sometimes a Neo Geo tile is called a "character", hence the 'C'. Each tile is 16x16 pixels, and indexed from 0 to 15
<< diagram of a tile >>
Zero always means transparent, even if the tile will be used for a background. Neo Geo games can store an immense quantity of tiles, maxing out at 1,048,576. This equates to 128 megabytes of graphic data! So even though graphics are built out of tiles, in many games so many different tiles are used at once, the resulting screen looks like a fully hand crafted scene rather than having that "repeated tile look".
To define a sprite such that it will show up on the screen, you set attributes about the sprite in video RAM. We will do this quite a bit throughout this book, so we'll ignore the details for now. The attributes about sprites are stored in VRAM one after the other, and so each sprite is assigned an index. If two sprites are positioned such that they overlap, the one with the higher index is drawn last. The sprite's index is the only way to control this z-ordering. If you need a sprite to be drawn above all other sprites, it must be located later in VRAM.
<< graphic demo'ing sprite index defining z-order >>
The Line Sprite Controller (LSPC) is responsible for reading VRAM, and rendering the sprites as specified to the screen. Modern games use a technique called double buffering, where the next frame to be displayed is assembled off screen. When it is ready to be displayed, the screen buffer and this off screen buffer are swapped. Then repeat the process for the next frame. This offers many benefits, a key one being the player always sees a complete, cohesive image.
The Neo Geo does not have enough memory to do double buffering. As the electron beam draws lines to the monitor, the LSPC is reading VRAM and telling the electron beam what colors to send on the fly in order to display the image. If you change VRAM while the beam is somewhere on screen, you can cause two different frames to be displayed at the same time. This is known as screen tear.
<< illustration of screen tear >>
Once the electron beam arrives at the bottom, it takes a bit of time to return to the top so it can start again. It is this window of time that you can change VRAM values without any screen tearing consequences. The time it takes to return to the top of the monitor is known as "vertical blank" and commonly shortened to "vblank". The Neo Geo will let your game know when this is happening, so you can coordinate your sprite changes accordingly. It can be challenging to make all the changes you need in the very short duration of vblank however, so sometimes despite best efforts screen tearing can still occur.
Due to memory constraints within the LSPC, there can only be a maximum of 96 sprites on any given scanline. This might sound like a lot but it's an easy limit to hit so you need to be careful. I recommend testing your game in an emulator that emulates this limitation, such as MAME. We'll cover this more later in the book.
It's not entirely true that all Neo Geo graphics are done with sprites. There is an additional plane for simple graphics called the fix layer. This layer is always drawn last and appears above everything else on screen. It can not be moved in anyway, and is formed of 40 columns of tiles, each 32 tiles tall. The tiles used for the fix layer are 8x8 pixels in size, and they are stored in the S ROM. Just like sprite tiles, they are indexed from 0 to 15, and use palettes to determine their ultimate color on screen. Most CRT monitors draw the image larger than the actual physical screen, so edges often get cut off. This is known as overscan. Due to this, SNK recommended developers only place tiles in the center 38x30 section of the fix layer.
<< diagram of the fix layer >>
The fix layer is almost always used for the "heads up display" in games. Things like life bars, credit counters, timers, etc, are usually found on the fix layer. It's convenient and simple to use, so it is ideal for these kinds of simple visuals. It's also useful for screen transition effects, such as this one found in ???
<< gif of screen transition >>
The fix layer's tile map is located in VRAM. To set a tile on the layer, find its location in memory and write the palette and tile index. Just like sprites, this should ideally be done during vblank. We will look at precisely how to do this in the next chapter.
There is also 8kb of palette data stored in the 68k's main memory, so you could say the Neo Geo needs 76kb of memory to handle graphics ↩