I often like to cover oddities here; details of computers and arcade systems people may not have heard of, that didn’t sell well but had unique or interesting characteristics. But focusing on oddities like that can disguise the fact that sometimes, even systems that were very popular can stand out for unique design. Such be it with the Nintendo Entertainment System: had Nintendo’s console flopped, it’d definitely have a post already. But it doesn’t. Let’s fix that! And answer a simple question I was asked: what’s the deal with special chip games?

The Famicom

The Family Computer was Nintendo’s first console. Releasing in 1983, just a year after 1982’s Donkey Kong Jr., the first Nintendo arcade game to be developed entirely in-house, without help from Ikegami Tsushinki. (Though they may have peeked at the prior boards just a bit, and got sued just a bit, but that’s another story) And isn’t it adorable?

The interesting thing about the Famicom is that it was released in 1983, and its last licensed game in Japan, Master Takahashi’s Adventure Island IV, came out in 1994. The last licensed game, Wario’s Woods, came out in 1995 in Europe. That’s a long lifespan, and unlike the PC Engine, whose last licensed game Dead of the Brain Vol. 1 and 2 came out after a several-year drought, the Famicom was continuously getting more and more games. (Between unlicensed titles, the NES-on-a-chip and homebrew, it has never really stopped, either)

Think about the Famicom launch title Donkey Kong. It’s a reasonable arcade conversion, with high quality graphics. But much like the ColecoVision pack-in, it only includes three of the arcade game’s four levels, and none of the interstitials (“HOW HIGH CAN YOU GET”). It has 16kiB of program ROM and 8kiB of character ROM (just like Aspect Star “N”).

The largest game, HAL Laboratory’s 1991 Metal Slader Glory, dwarfs even Adventure Island IV with 512kiB of program ROM and 512kiB of character ROM. Its MMC5 mapper chip allows for a ton of advanced features. As for the game itself, it’s a text adventure with some stunning graphics, especially when viewed on a CRT.

But most people know this. The Famicom and NES benefitted from chips– like the aforementioned MMC5– inside the cartridges that expanded the system’s capabilities. For example, even a more modest title like Championship Bowling uses a “CNROM” board, which allows for larger graphics ROMs than the 8kiB limit; in this case, 32kiB, four times the graphics space.

Have you ever wondered, though, what that means, exactly? What are these additional chips doing? And why did we see this on Nintendo’s consoles more than any other manufacturer?

Here’s the inside of the most sophisticated Sega Master System game, Phantasy Star. It’s got a larger program ROM (with a mapper to help out), and a smaller, battery-backed RAM chip. But it offers no technical enhancements to the Master System hardware; graphically, Phantasy Star is unable to do anything that launch-title Teddy Boy couldn’t have done; it can just do more of it with that big beefy ROM. Championship Bowling is digging into the deep internals of the graphics side of the system in a way Phantasy Star simply can’t. (Or, depending on your point of view, doesn’t need to)

Cartridge ports

The Fairchild Channel F used the Fairchild F8 CPU, which had kind of a strange architecture with no address bus per se. So let’s switch to the Atari 2600.

The Atari 2600’s cartridge port is a model of sheer simplicity. 13 address lines allow for 8kiB of gaming action– well, actually, A12 needs to be used as a chip select, so 4kiB of gaming action. There’s an 8-bit data bus, 5V for power, and grounds. That’s all you need, right?

Wrong. The Atari 2600’s cartridge port is actually very limited, and couldn’t even do Phantasy Star above. It’s missing a pin on the 6507 CPU called “R/W”; this means that the Atari cartridge has no idea when the game is reading or writing from memory. This is fine for a ROM chip, which can’t be written to anyway. But what if you wanted to put RAM on the cartridge? After all, the 2600 only has 128 bytes built in. You can’t do it. Not in a straightforward fashion, anyway.

Well, of course, people found ways around these limitations. Without them, we wouldn’t have Pitfall II (Atari home computer version, not the 2600, pictured above) But it made things pretty complex to juggle when you had such a small window for your ROM to begin with.

Not only does the Master System have a larger address space for cartridges (thanks to its Zilog Z80, rather than the MOS 6507), it sent nearly all the signals from the Z80 out to the cartridge, including read and write signals. This allows for Phantasy Star’s save RAM– plus, all of this was also true of the predecessor to the Master System, the SG-1000, which had RAM expansion on at least one title, The Castle.

The NES went even further than that.

CHR

What do the ColecoVision, the Sega SG-1000, and the Master System all have in common? They all have the same amount of VRAM (Video RAM), 16kiB. This is because they have (more or less) the same video chip; and the TMS99xx video chip was very sophisticated for its time, allowing for cheaper dynamic RAM. In fact, back in 1979, that 16kiB formed the bulk of the RAM for TI’s TI-99/4A computer.

What is video RAM? It’s a solution to a serious problem. A video signal on the NES is a composite signal, ~60Hz in NTSC regions, ~15kHz line rate. Something like this.

The key thing to note about this signal, though, is that the video processor must constantly output its signal. It can’t be early, it can’t be late, it can’t take a break. This means if there’s a conflict between the processor wanting to use RAM, and the video chip wanting to use RAM, the video chip must win. The Apple ][ pulls this off with careful timing, but it only runs at 1MHz and has no sprites (or even colors, sort of). The more advanced video chips in the Commodore 64 and Atari 8-bit computers, meanwhile, must have the ability to stall the CPU to let the graphics chip through. Lines where this happens are called “badlines” on the C64.

Video RAM lets the CPU run at full speed at all times by having the graphics chip use RAM on a separate bus. The downside is that you need to go through the video chip to access this RAM, which is a lot slower. For example, writing a byte $0A to RAM address $2000 is pretty easy on the 6502:

LDA #$0A ; Load the accumulator with 0x0A
STA $2000 ; Store the accumulator at $2000

What if we want to write that byte to video RAM? On the NES, it would look like this:

LDA #$20 ; high byte of $2000
STA $2006 ; PPUADDR register
LDA #$00 ; low byte of $2000
STA $2006 ; same register
LDA #$0A
STA $2007 ; PPUDATA register

On the NES, writing an immediate value to a known arbitrary address is three times as slow as accessing regular RAM. And it’s even more limited than this. The CPU always has access to its regular RAM, but on the NES (not all systems with VRAM have this same limitation) you can not write to video RAM at all while it’s drawing the screen. Therefore, the “VBLANK”, or vertical blanking period, when the beam isn’t drawing a visible picture, is of the utmost importance.

Dare to compare

So, the TI-99/4A computers, the ColecoVision, the SG-1000, and the MSX1 standard, and more all use the TMS99xxA series video chips, and all of them had 16kiB of video RAM. Here’s what Dig Dug looks like on the MSX1.

The NES has 2kiB of video RAM, one eighth of the MSX1. It also, though, has 256 bytes of OAM (object attribute memory in NES language, or sprite data to use more common terminology), and 32 bytes of palette RAM that aren’t included in that 2kiB. So, what does Dig Dug, released by the same company in the same time period, look like on the NES?

MSX fans are probably already writing in. After all, the NES has several big advantages over the MSX. In these screenshots alone, you have:

• Three-color sprites (vs. monochrome on the MSX1)
• Up to eight sprites per scanline (vs. 4 on the MSX1)
• Higher vertical resolution
• More colorful tilemap

However, all of these should actually be taking up more VRAM than the MSX, not less. Well, of course, what’s going on is well-known. The game Dig Dug uses CHR-ROM. A Dig Dug cartridge has an 8kiB ROM inside of it that’s hooked up directly to that video bus. So to say “the MSX1 has 16kiB of VRAM, the NES has <2kiB” is really not being a fair comparison. The MSX fans have a point.

B