Minimal 64 and Minimal Ultra enter 8-Bit Battle Royale

[@genjii931]

In the words of ProjectFarm, “Very impressive!”

[@TheWinnieston]

Thats absolutely nuts! People just casually designing ttl cpus, I want to be that good!

[@slu467]

I might have made it look too easy - as if I was designing this stuff "casually" but actually I am not 🙂 It took me many many hours to reach this goal.

[@alessiocaffi5992]

much appreciated vid. and new Minimal! yhea that was amazing thanks, knew you really rock👍.

[@Aeon_Electronics]

Always fun to see your videos! :) Entertaining as well as educational!

[@gsestream]

btw, you can run multiple 8-bit cpu's side by side like a gpu

[@djc1402]

Another great achievement, I always love your videos. My hacked CSCvon8+VGA can do the Mandelbrot in about 2 seconds at ~6.3MHz but its got 'hardware' multiply (8x8 bits). It would be interesting to see how fast a 6809 based computer can go. I would be very interested in a Ultra version on PCB, if you could make that happen then I would definitely build one.

[@slu467]

Hi David, no hardware muliplication on board here ;-) just a *very* efficient software implementation. Actually, I am currently not planning to release a PCB version of the Ultra since I fear that manufacturing tolerance of parts may come into play. This is really a "close to the limit" proof-of-concept design. But something with VGA and PS/2 may follow at a later stage.

[@andrewwatts1997]

This is fantastic. I wonder tho, Why does no one make 8 bit cycle instruction cpu's ? Seeing the CPI and CPI set to one would be satisfying. A next project perhaps ? Wink wink*

[@alexloktionoff6833]

Waw, thank you very much!

[@blinking_dodo]

2 days ago i decided on making my own CPU and yesterday i was thinking how to ensure and measure the efficiency of my instruction set. Today i stumbled across this video. Thanks YT. I suppose that the best way to measure efficiency is to have the CPU's perform a fixed task at fixed clock speeds and measure the time it takes to complete. To have a fair comparison of architectures/instruction sets, compilation should happen with the default compiler at default optimization settings. The task would probably best be a single C program with many calculations, branching, looping and function calls. The 2 important measurements would be binary file size and program duration Would you perhaps have any good resources on (amateur) CPU development? (I am going to use an FPGA for this)

[@TheSulross]

the Agon is at a disadvantage because it has to do graphics to its display over a serial link to a ESP32 microcontroller. So even though it has an enhanced Z80 compatible running at 18MHz, it's CPU prowess will be humbled by its display bottlekneck

[@MagnusWedmark]

Would be a blast to try out a really optimized 8-bit FPGA-based platform, for example, the well-known Picoblaze Xilinx/AMD 8-bit mini-controller is very nice and I did a lot of damage with that one. I don't own one of the later FPGA-development boards but I think it ran at 100Mhz even in the old ones and each instruction takes exactly 2 CC's => 50MIPS. It only holds of 1024 bytes of RAM (I think it was 512 instructions max) otherwise you to introduce bank switching, which brings down performance. I wonder if I can fit the program into that space.....

[@MagnusWedmark]

I actually started with implementing this using an Assembler/Emulator solution. Currently I'm just outputting 32x22 chars/pixels increasing the value for every line => Takes around 1ms in total. A wild guess would be like 1000CC (500 inst per pixel) which will end at around 1s for the whole picture.

[@lucas27932]

Excellent video and great achievement in optimization👍. Can the current Minimal64 PCB, be tweaked to a higher performance? With the more efficient micro code and/or higher clock rate? 🙂(Tip: if you plan on an next PCB design, I am interested. Make it so, that standard casings can be used. 😉)

[@slu467]

Hi Lucas, oh, yes, the Minimal 64 can be "tweaked" to run twice(!) as fast with what I have learned here. That is what I am currently having fun with... watch out for the... Minimal 64x2 ;-) But there is still some work to be done... Can you point me to "standard casings" you intend to use?

[@lucas27932]

Hello Slu467, The pcb can be conform mini-ITX, flex-ATX or micro-ATX, to be used in standard pc casings. An other (a more DIY) option is, to make it fit in widely available containers, that can be fitted in to a casing for the Minimal. For example: the 212x212x35mm (transparent) FerreroRocker Giftbox; Mepal 195x125x20/52mm Lunchbox "Take a break" flat/large; IKEA 240x350x60mm HARVMATTA, 310x230x70mm HÄSTVISKARE or 460x250x30mm ELLOVEN (this last one, has room for a keyboard underneath). Or something similar. (Ikea, measurements are estimated inner dimentions. The FR Giftbox and Mepal are exact inner dimentions).😀@@slu467

[@LostieTrekieTechie]

Do the rules prohibit using a multi-processor or ultra parallel GPGPU-like system where every pixel is calculated in parallel with its own processor?

[@slu467]

I haven't stated it explicitly, but I think Matt uses his benchmark for retro-style 8-bit systems. I hijacked it for DIY TTL computers. If you can build a TTL version of what you propose above that'll count as well.

[@alexloktionoff6833]

Can you make your new versions modular 10x10cm for easy/cheaper reproduction? Or 1:1 replacement for C64, XZSpectrum or mITX format?

[@slu467]

Hm, I am not a fan of modular design of that (rather small) size. I drives the cost up, actually. And why make it fit into a C64? You would need a different keyboard anyway...

[@alexloktionoff6833]

Good modular design is easy to troubleshoot for beginners, and 10x10 is cheaper to order. Yep, keyboard is different, just an idea to make childhood dream come true - understand the computer, make the same. But what about mITX format? People could simply put them in available PC case - complete device. @@slu467

[@alexk24240]

I see in 2.0 version memory-toPC logic changed. can you describe changes? maybe another video about Minimal evolution? thanks for your job.

[@slu467]

Thanks for following along. Soon there will be a new Minimal 64x4 coming out displaying the full feature set (4x processing power of a C64). Some explanation will follow...

[@foo0815]

A comparison to a 68000 or 68008 (to stay 8 bit) based system would be interesting...

[@slu467]

... but I will leave that to the "owner" of the "8-Bit Battle Royal".

[@alexloktionoff6833]

So, even using not so powerful SAM CPU h/w architecture you've managed to develop very efficient CPU instructions set? Did you perform some simulations/benchmarks while designing the CPU? Please make a video comparing instruction set comparison, why your Mandelbrot implementation outperforms 6502 & Z80!!! please please

[@slu467]

Although technically my design is of the SAP type it took much cumulative work to get it to this level. "Minor" details become quite important at 10MHz with constant simulations/benchmarks for many months 🙂Maybe more about it in the future but I guess most people will find this boring...

[@alexloktionoff6833]

Guys let's vote for instruction set explanatory/comparison vision! Of course your DIY h/w is brilliant. But instruction set is different and very interesting topic. Probably you've invented very powerful 8bit instruction set which is even more important than just one working and fast home computer.@@slu467

[@zombiigames4151]

I wonder what happens if i run Minimal 64 on scratch

[@slu467]

😁

[@stephenwhite506]

I wonder how George Foot's 35MHz 6502 would rate? Extrapolating from the other 6502 based figures, I guess that they would be roughly 0.74s with 14.8Mips!

[@slu467]

I think it is one thing to have a fast CPU and another to have a fast CPU interacting in a meaningful way with RAM and peripherals and stuff. For the Minimal, the bottleneck is RAM/FLASH access times.

[@robicjedi]

The I/O must play a significant role here. The UART cpu has instructions just for the the UART, while that must take many cycles for other computers. Is there a way to do all the calculation in memory only, so that the I/O is not included?

[@slu467]

Actually the graphical output requires only a very small fraction of compute time here and does not play a significant role.

[@StevenSherman-kr4xm]

Nery nice