Irix 2600 by deSire -- An Atari VCS Demo on an E7 Cartridge

I recently managed to get custom code running on an E7 Cartridge for the Atari 2600 (you can read more about that adventure here).

This demo is vaguely inspired by SGI Irix, the OS for the first UNIX workstation I had hands on access to about 30 years ago.

This demo was first shown at Demosplash 2026 and came in ? place at the Retro Demo competition.


Video

Capture on an actual Atari 2600 and E7 Mapper. (This was captured in a bit of a rush just before the deadline). Note: the video glitches out in the middle on this capture card, but it played OK on various real-world devices and also played fine during the compo screening.

Pics of Running on Real Hardware

Running with chips harvested from a "Burgertime" Cartridge, with the RAM updated to 2k:

Requirements

An NTSC Atari 2600/VCS.

It works fine in the Stella emulator.

It also works fine on a Harmony cartridge.

And of course if you want to take a Burgertime cartridge and replace the ROMs you can do that too, but as linked to above, it might be a bit more effort than you'd want to put into it.

Credits

Note: no AI was used in the production of this demo.

Notes on the Code/Effects

The E7 cartridge provides 16k of ROM space (though only 15.5k is available) and 2k of RAM. The top 1.5k of ROM is always visible. The rest of the ROM is available as 7 different 2k chunks that can be bank-switched in to the bottom 2k of the cartridge address space. 1k of the RAM can be read/written in 4 banks of 256 bytes. Finally the remaining 1k of RAM can be switched in instead of the bottom RAM bank.

Since the Atari 2600 cartridge slot is read only, writes to the RAM are hacks, there are two different address ranges for the RAM, one for reading, one for writing. This does mean you can't use read-modify-write instructions like INC or DEC on these RAM locations. It turns out (from my experiments) that you *can* execute code from out of the RAM which is not always the case on other mapper types (looking at you, SARA).

I wrote this code sort of last minute and so not much size coding was done. Only about half the ROM is used.

ScreenshotEffectDescription
Title screen ROM BANK0: Various desire logos. These are using the four 256 banks. There's a common set of playfield data for the logo, but the foreground/background color variations are copied to the four RAM pages, and then banked in and out. There's a slight fade effect for the first two, I meant to have it fade all the way out but that's tricky to do in the cycle budget.
Bouncing Cube ROM BANK1: The SGI logo at various sizes and bouncing around the screen. Mostly showing off some of the sprite options. Not using any RAM. Had originally planned to animate and/or go behind the background, but ran out of time.
Teapot Desktop ROM BANK2: showing a fake IRIX desktop. At first it looks like it's static, with playfield graphics and a sprite overlay for extra color. But then the teapot starts animating. I actually have 8 frames of animation but could only fit 4. The data for that (a standard atari 2600 48-pixel setup) is compressed in ROM, ZX02 decompressed to the large 1k RAM, then copied to the four pages of 256-byte banked RAM. Ideally I'd be able to fit 8 frames here, but couldn't figure out a good way to get everything in memory. Also the 1k range is hard to use, since it replaces ROM you can only really use it from the 1.5k always-mapped region but that ROM space is limited.
Know This ROM BANK 3: Based on a scene from Jurassic Park, where the young hero realizes the park is being run on SGI Irix for some reason. Mostly static backgrounds with one set of sprite overlays for extra color. I intended to have a second sprite overlay and possibly even a smaller teapot animation on the screen but ran out of time. No RAM was used for this.
Sierpinski BANK4: This was my attempt to use the RAM to get a lot of colors on the screen at once. I managed 10, all using the playfield (I think other groups have gotten 11 by using all the various sprite/ball/playfield elements).
The unusual thing about this is the box drawing code lives in RAM, in 3 of the 256-byte banks. At startup the load constant/store out code is written to RAM, and then to change the color being drawn you self-modify this code to change the immediate values. This lets you change colors in only 2 cycles rather than 3 (ZP) or 4 (read from ROM) allowing more changes per line.
The end result is not very dense, but it is a crude fully addressable 10x15 grid. I had various ideas of how to show this off but in the end the best I managed was this which is Sierpinski but with the colors shifting a bit. I was hoping for a lot more here, maybe a plasma, or visualization but ran out of time.
Credits BANK5: credits with a mirror effect. These credits live in the 256-byte bank switched area. I have a font routine (up to 32 4x8 characters) that can print to the page. The mirror effect is just printing the same data upside down and alternately skipping lines. I had hoped to have the text scroll, but even though we're only display 42 lines (42x6-bytes fits in 256 bytes) scrolling them all up takes way too many cycles to fit in one frame and it's a hassle splitting it across multiple frames w/o tearing. The bottom half of the screen with the parallax and waves is re-used from a previous demo of mine.
n/a Music ROM BANK6 has the music, written by MikeR with TIAtracker. I used the export to generate the code and then hand-hacked it for ca65 assembly language. There's a trampoline in the common area so you can call into the music code (which takes roughly 11 scanlines) from any ROM area.

Sourcecode and Disk Image

You can get the disk image here: i2600.e7 (16k), 21 November 2026

You can get the submission here: i2600.zip (33k), 21 November 2026

You can get the sourcecode here in the demos/i2600 directory of my vcs github repository:
git clone https://github.com/deater/vcs/
Vote for this at Pouet
Back to my Demos