BMOW title
Floppy Emu banner

Nibbler Demos!

After fiddling with Nibbler’s hardware and fixing its glitches, it’s time to write some demo software. Let’s see what this little handmade 4-bit CPU can do! Sorry the quality on these videos isn’t the best – try changing the quality setting to 480P to get a bit more picture detail.



Gamers of a certain age will doubtless remember Mastermind, a code-breaking game for two players, which is based on an old pencil and paper game called Bulls and Cows. The codemaker chooses a secret code that’s four elements in length, where each element can be one of six possible colors. The codebreaker then has ten chances to guess the secret code. After each guess, the codemaker gives feedback in the form of black or white pegs: a black peg means some element is the right color and in the right position, and a white peg means some element is the right color but in the wrong position. The feedback pegs are position-independent, so a black peg doesn’t tell the codebreaker which of the four elements in the guess was correct.

Adapting Mastermind to Nibbler was relatively simple. A few weeks ago I wrote a Guess the Number program for testing purposes, and most of it was reusable for Mastermind. The final program was 2057 bytes, or just over half of Nibbler’s available program memory. Instead of using colored pegs, the code is a 4-digit number, where each digit is between 0 and 5. I added a few little bells and whistles, like button feedback sounds, and a little victory tune when you guess the secret code correctly. The result is surprisingly fun, if you like these kinds of logic puzzles. It works fine in the Nibbler simulator too, if you want to try it out.


Blue Danube

I wanted to focus on music and audio next. Nibbler doesn’t have any real audio hardware, only a speaker that’s directly connected to a digital output pin. Making sounds is as simple as toggling the output quickly between 0 and 1, but making specific sounds at just the right frequency is more complicated. It requires a lot of cycle counting math, to guarantee the speaker will be toggled at exactly the right rate for an A at 440 Hz or a piano’s middle C (261.6 Hz). Each period of the waveform needs to be exactly the same length, even as the path through the inner code loop varies in length due to carry propagation with multi-nibble counters. If one period is slightly off, your ear will hear it.

Setting the duration of each note involves more math, dividing the duration by the period length to find the number of periods to play. That means two notes at different frequencies but with the same duration will have different repeat counts (duration values) in the code, further adding to the necessary bookkeeping.

I started writing a music demo by hand, but it was such a pain that I couldn’t imagine building a whole song that way. Instead, I created a new tool called Music Maker to do the math and code creation for me. It takes song data as input, and generates a Nibbler assembly program as output. The song data is expressed in the Music Macro Language from Microsoft Basic, which you might recall if you ever used GWBASIC’s PLAY command. This simple one-voice text-based format is a perfect match for Nibbler’s limited audio capabilities. Songs are described as a series of notes, with optional length and octave modifiers:

"T180 DF#A L2 A L4 O4 AA P4 F#F# P4 O3 D"

Music Maker saved me a huge amount of effort, but I still needed to hand-edit the generated assembly code to repeat a few musical phrases in the right spots. Because Nibbler’s instruction set isn’t well-suited to storing constant data in programs, the code is fairly bloated, taking 3569 bytes (87% of memory) for a song that’s a few dozen measures long. I chose the Blue Danube for the demo song.



For the last demo I wanted to do something really different, so I created a real-time action game using custom graphics. The classic arcade game Frogger leapt to mind. The player must guide a frog across many lanes of traffic and a treacherous river, avoiding a variety of obstacles moving in different directions and at different speeds.

Bringing Frogger to Nibbler presented several challenges. First – graphics. The HD44780 chip inside the 16×2 character LCD screen does support user-defined character fonts, but only for eight characters, which doesn’t provide a lot of variety. Second – screen size. The screen only has two rows, and that’s not many lanes of traffic for our frog to dodge. How can you make a game with that?

Each character on the LCD screen is 5×8 pixels, so the total screen height is 16 pixels. My approach was to divide the screen into four virtual rows, each of which was 4 pixels tall, with two virtual rows per actual row of text characters. Then I created custom character fonts for each possible combination of contents in the cells in the upper and lower virtual rows. Each cell can contain one of three possible items (an obstacle, the frog, or nothing), and there are two cells, so there would seem to be 3×3 or 9 permutations. But because there’s only one frog in the game, the permutation with frogs in both cells isn’t needed, and the remaining eight permutations fit exactly into the HD44780’s eight user-defined character slots. The result is a 16×4 virtual playfield, where each playfield cell is 5×4 pixels. For a bit of added variety, I also made the upper and lower obstacles look visually distinct.

To make the game “real-time”, I needed a way to animate the obstacles while the player was moving. Without interrupts or any other real time-keeping mechanism, I had to add a timeout counter inside the busy loop that checks for button input. After a few thousand checks of the button state, the game jumps to a routine that moves the lanes of obstacles by one position, then returns to button checking. Fortunately it all happens fast enough that you can’t notice any hiccups.

Frogger was the only demo where the lack of an indirect addressing mode really hurt. Tasks like animating the playfield or checking for collisions are just screaming out for indirect addressing. Without it, I had to write code that explicitly copies/checks/moves each of the 64 playfield cells. The result is lots of ugly bloat, and a code size of 3279 bytes (80% of memory).

The final game turned out nicely! The only big flaw is the animation: the persistency of the LCD screen makes objects appear to fade and flicker, but there’s nothing I can do about that. Despite this, it’s a lot of fun to play. In the video, notice that the lanes of obstacles animate at different speeds, and each lane moves in the opposite direction from its neighbors, making the game more challenging than it first appears. If you can navigate the frog upwards from the bottom row, and escape out the top row, you’ll hear a little victory tune. Go Nibbler!

Read 5 comments and join the conversation 

5 Comments so far

  1. Stephen - August 13th, 2014 11:44 am

    Dear Steve,

    I am trying to build a clone based off of your schematics and have been able to get all of the parts so far however I am having trouble finding the clock. I know that it is the vishay dale XO52CTELNA2M4576 crystal oscillator but I have not been able to find it anywhere! There are none in stock and the minimum order requires a quantity of 500. Is it possible to use a different clock at a different speed? Perhaps 1mhz or higher than the 2.46mh? Thank you for posting your nibbler project online, its awesome! I’ve always wanted to know how a cpu works, and now I am building one!

    Thank you,
    Stephen Robinson

  2. Steve Chamberlin - August 13th, 2014 12:24 pm

    Any standard 2 MHz or 2.5 MHz 5V “can oscillator” should work fine. At 2 MHz, programs will run slightly slower and sounds will be lower in pitch, but the difference shouldn’t be too dramatic. With a 1 MHz oscillator it will still work too, but it will be noticeably slower, and you may need to edit the timing loops in the sample programs. Anything faster than 2.5 MHz probably won’t work reliably.

  3. Ben - June 23rd, 2015 2:55 pm

    For faster animation try a different LCD, the blue back light with white display is horribly slow compared to the black on green display.

  4. Nibbler 4 Bit CPU - My Blog - December 4th, 2022 6:33 pm

    […] and playing the Blue Danube Waltz. For lots more information about how these demos were made, see this post. Switch to the 480p quality setting for a bit more detail in the […]

  5. Nibbler 4 Bit CPU by scour - HackTech - December 5th, 2022 3:54 am

    […] and playing the Blue Danube Waltz. For lots more information about how these demos were made, see this post. Switch to the 480p quality setting for a bit more detail in the […]

Leave a reply. For customer support issues, please use the Customer Support link instead of writing comments.