VIC 20 Video Chess

Released on 20 October 2025

Download VIC 20 and C-64 versions at:

bit.ly/videochess-dropbox

Tip jar:
ko-fi.com/aleksieeben

Video Chess manual at archive.org:

archive.org/details/Video_Chess_1978_Atari

Original Video Chess disassembly:

nanochess.org/video_chess.html

Video Chess: Atari Archive Episode 31:

www.youtube.com/watch?v=ERdpCddJ3S8

Other VIC 20 ports by Aleksi Eeben

Elite: vic20elite.wordpress.com

Ultima III: vic20ultima3.wordpress.com

Ultima II: vic20ultima2.wordpress.com

Ultima I: vic20ultima1.wordpress.com

About Video Chess

Video Chess is a chess game released by Atari in November 1979 for the Atari VCS / 2600, programmed by Larry Wagner and Bob Whitehead. Despite the severe hardware limits (only 128 bytes of RAM and a 4 KB ROM cartridge), the game implements a functional chess engine with multiple difficulty levels, en passant, castling, and a full-depth (albeit shallow) search. The achievement is historically significant as one of the earliest microcomputer chess programs squeezed into minimal memory, showing what was possible with extreme optimization.

The disassembly and analysis by Oscar Toledo G. in 2023 reveals that the game keeps the board in memory from addresses $80 to $BF (one byte per square), storing piece type in the lower nibble and reusing bits in upper nibble as scratch / side / flags. Move generation is table-driven using displacement tables (e.g. queen moves in rays) and BCD arithmetic for position handling. The program simulates a stack (within the same memory region) to push and pop board states when exploring moves. It implements quiescence search logic, king opposition heuristics, and dynamic depth increases in low-material endgames.

There is also code for castling (although the computer itself can only castle on the king side) and en passant. Some players have speculated a bug on levels 6–7 (apparently “moving two pieces at once”), but the reverse-engineering did not conclusively validate it, treating it more as legend than confirmed flaw.

In the landscape of early microcomputer chess programs, Video Chess stands out as a demonstration of how far you can push severe constraints. It may not match the playing strength of Sargon II (released around the same era, which was more ambitious and respected as a chess engine) — especially in deeper play or sophisticated positional understanding — but Video Chess is remarkable for fitting a working chess engine (with special rules) into just 4 KB plus 128 bytes of RAM. Many chess programs on microcomputers used much more resources. Dedicated chess computers, such as the Fidelity Chess Challenger, had four times more ROM and RAM to work with, and only used a simple digital display to show the moves. The design is a testament to extreme optimization, code reuse, and clever tricks. It also served to show that even “toy” game platforms could host nontrivial AI.

VIC 20 Port

This 2025 adaptation by Aleksi Eeben, based on Oscar Toledo G.’s disassembly, brings Atari’s classic Video Chess to Commodore’s beloved VIC 20. On the VIC 20, the program retains the same core move logic—supporting castling, en passant, and eight difficulty levels—while reworking the display code to suit the VIC’s graphics capabilities. What makes this port special is that, unlike most chess programs which require RAM expansion, it runs on an unexpanded VIC 20, making it one of the very few, if not the only, chess games playable on a stock machine without additional RAM.

In the VIC 20 version, the screen stays on while the computer thinks, showing lively “turbo loader”–style sparkles instead of the Atari’s blank, flashing screen. On the Atari, a physical switch indicated when the board was in edit mode; on the VIC 20, a purple background is used instead. The classic Atari color-cycling screen saver has been removed, pixel lines in pieces are doubled, and the screen ratio is slightly flatter. While the same sound triggers are used in the code, the effects sound subtly different.

At 46 years old, this game wears its history on its sleeve. By modern standards, the controls feel slow and unresponsive, but they were well-suited to the Atari 2600’s plastic joystick and the expectations of players used to digital clocks, black-and-white LCDs, early TVs, and VCRs. For historical accuracy, none of the user interface code has been modernized, so each button press and movement requires a bit of patience—just as it did in the late 1970s and early 1980s—making the game both a chess challenge and a small time capsule experience.

Estimated Elo for Atari Video Chess & Sargon II

Estimating Elo for early microcomputer chess programs is necessarily approximate, as strength depends on hardware and test conditions. Sargon II was one of the strongest engines of its era, with top tournament-level play around 1500 Elo and casual play in the 1150–1250 range. Atari 2600 Video Chess, by comparison, likely plays around 1200 Elo on average, with intermediate levels ranging roughly from 1100 to 1300. While Sargon II holds a modest edge, both programs showcase impressive chess capabilities given the severe memory and processing constraints of their respective systems.

The “Battle”: ChatGPT vs Atari 2600 Video Chess

In mid-2025 some hobbyists staged a playful “match” between a running emulation of Atari 2600 Video Chess (the original 1979 game) and ChatGPT. The setup: ChatGPT was given board states and asked to propose the next move; the Atari engine simply responded with its own move as in normal play. To many people’s surprise, Video Chess won.

The media and tech-reporting around this “upset” emphasized that, while ChatGPT is powerful at natural language, it is not a dedicated chess engine — it lacks deep tactical calculation, memory of full game state across turns, and search heuristics. Meanwhile, the Atari engine, though primitive, still mechanistically applied alpha-beta search (shallow), move generation, and evaluation based on piece counts and simple heuristics — enough to best ChatGPT when the latter mis-tracked the board or made blunders in move planning.

The result is more symbolic than scientific, but it caught people’s attention as a charming example: a 4 KB vintage chess engine beating a modern language model in their shared domain of chess. It underscores that “knowing chess theory” and “playing chess via search” are different modes of reasoning — and that general models like ChatGPT still struggle when asked to maintain long chains of precise, rigid logic under constrained memory.

ChatGPT did moderately well in summarizing all of this information though, effectively condensing the history, technical achievements, and gameplay details of Atari 2600 Video Chess into a clear overview.