A whole CPU, designed by Fable 5.2 from 147 lines of code.
@irinatoxi3.7K followers
52,567 views
28 Comments
Comments from the original post on X (Twitter)

@TylerGiacoma
What’s crazy is how quickly hardware advancements are about to be…. People will legitimately be 3D printing their entire robots and using open source models to run them locally 🤯 Things are about to get wild
10 likes

@irinatoxi
Still early though, this one is a tiny CPU that only runs in simulation. But a year ago I couldn’t have made even that.
4 likes

@phobiaphilia
Calling both an in-order scalar processor and a modern out-of-order superscalar processor just a "CPU" is misleading... Actually this one is closer to a microcontroller than a typical CPU :(
9 likes

@I1nurSu
>It started with 147 lines of code that describe a CPU, and those turned into 6,258 logic gates and 1,023 memory cells. Verilog has been around for decades already.
5 likes

@stevencheng
Zero mismatches on 242k instructions is wild.
4 likes

@alfcnz
Cool! I may try to get it running on my FPGA! At the moment I was planning to just try an 8080. But, hey, Fable's design is more than welcome! 😀 Not sure why you added music to your video. It's distracting.
3 likes

@irinatoxi
Thanks! It was built and tested only in simulation, and it assumes memory that answers instantly, which most FPGAs don’t have without changes. The 8080 is probably the more fun project for now. Would love to see that one!
1 like

@alfcnz
You can see a few tweets on my profile and the repo with all lessons is on GitHub! 🤓🤓🤓

@buildsmrtr
Takes me back to my VLSI labs. A full semester to get a single-cycle core through synthesis, and timing closure failed half the class. A model going straight to a clean netlist saves a semester. Would love to see this on an open PDK shuttle.
3 likes

@bar96169
Are the reference model and test harness public? I’d love to see the edge cases alongside the random instructions. Tiny CPU, potentially very creative bugs 😅
3 likes

@virajitgp
you generated 147 lines of hdl code, the rest was done by compilers and interpreters, the pnr is also not a big deal there is enough oss tooling alr
2 likes

@AIQuanting
Gates plus memory cells add up to the 7,281, nice. The number I would push on is wiring 5.5 times shorter than random. Random placement is about the easiest baseline there is. Did they compare it to an actual place and route tool?
2 likes

@nker212223
想问一下你这个是什么软件?
2 likes

@pokygonguy
In digital land yes , in real world with heat and entropy , no.
1 like

@MerlincsZeroYin
The chip design field will gradually be cracked by AI too. The more chip engineers use AI, the stronger AI’s chip-related capabilities become. The formerly closed-source chip world is slowly following the path that software development took.
1 like

@zbt3sup1
6k gates is essentially non existent. This is not the accomplishment your first sentence misleadingly conveys.
1 like

@DBru42
Makes me wonder if in the future there will be software designed CPUs like there is software designed radio.
1 like

@1day1money
@grok intel이 커스텀 칩보다 나은점은? 혹시 없다고해도 특허로 보호받을수 있나?

@Compilethings
He cooks hard, I will let him check some pcb, tomorrow.

@stas_sorokin_
147 lines turning into 6,258 gates and getting checked against a reference model is the part that sells it, a lot of generated hardware never gets that far. Do you run the reference model in your own flow, or only for demos like this one?








