Most of us will be familiar with the concept of the speedrun, an attempt to complete a game (or levels of a game) in the fastest time possible. However, this is the first time I’ve seen a CPU build get the same treatment, as Hackaday writer Julian Scheffers has managed to simulate a functional CPU from scratch in Logisim over the course of a mere six hours.
The project is called Stovepipe, by virtue of the fact that it needed a name quickly, rather than reference to a popular 19th century style of top hat. Scheffers says that Stovepipe’s hardware was made in under four hours, with the extra two dedicated to building the assembler afterwards.
The ISA (instruction set architecture) was created by a process of removing things that weren’t strictly necessary, resulting in eight major opcodes represented over 512 bits, far less than one of Sceffers previous 8192 bit processors, the GR8CPU.
Described as “an exercise in minimalism”, the end result features a mere 256 bytes of RAM, zero I/O ports, and an accumulator functioning as the only user-accessible register. Instructions are described as taking one cycle to fetch and between one and three cycles to run.
Compared to Boa³², Scheffers most recent previous build, the pared-down chip is significantly slower thanks to that chip’s 32 registers. However, it did outperform GR8CPU by virtue of the fact that it only needs one cycle to load an instruction compared to GR8CPU’s three. Boa³², however, looks like a Ryzen 7 9800X3D by comparison thanks to its pipelined architecture and separate address and data busses.
Still, there’s the time factor to consider. Scheffers estimates that Boa³² was completed over the course of two months, while this build appears to have been thrown together in one six hour session.
While it’s not quite a record holder (to its creator’s knowledge) in terms of performance in relation to size, it’s still a massively impressive achievement—particularly to someone like me who’s eyes begin to cross simply explaining the build, rather than making one for myself.
Scheffer says that if they attempt to make a Stovepipe 2 they’ll record the time taken with an actual speedrun timer, but even taking their word for it, it’s a pretty great example of processor design pushed to its limits. The practical usage of designing and simulating a working CPU in such a short space of time is dubious at best, but the fact that it can be done? Pretty mind-boggling, if you ask me.