The Agon light is the fastest 8-bit microcomputer
thebyteattic.com
thebyteattic.com
My understanding is that, from 1MHz or so (like a 6502) through to about 10MHz (like an 8086), it's pretty easy to prototype a complete computer — CPU wired to memory, IO, etc, and everything in tolerance to speak to one-another within the CPU's half-clock sample rate — with just discrete logic components (e.g. NOR DIPs to combine address lines to make chip-select pins, etc) on a breadboard.
Get much faster than that, though, and those discrete logic chips start to add too much delay; so you start having to prototype directly for a PCB with surface-mount components, rather than ever having a breadboarding phase. Which sucks, because now you're having to send your design out to be printed and picked, and so your iteration speed is now 1/100th of what it would have been before.
http://www.bambi.net/atari/atari_st_prototype.html
Unfortunately wire wrap sockets are very difficult and $$ to source these days. Which is too bad because it seems like a good halfway point between breadboarding and PCB.
FWIW I've managed to managed to get clean DVI/HDMI output off FPGA on a breadboard, and I'm by no means talented in electronics.
Are they really expensive? Or have the costs of chips just come down so much that wire wrap sockets _feel_ more expensive?
https://www.jameco.com/c/DIP-Wire-Wrap-IC-Sockets.html
A lot of modern chips are in the $0.50 to $3 region. So $1 or $5 sockets kinda feels expensive, but I don't think its really cost prohibitive if you're into wire wrapping.
----------
I think the real issue is that DIP parts / Through Hole parts are completely inferior to their SMT cousins today in 2023. If I want a MOSFET jellybean, I gotta use the 2N7000, which is specified as "on" at 5V (and really prefers to be at 10V Vgs).
Meanwhile, the BSS123 jellybean MOSFET is SMT only, smaller, more efficient, easier to use and easily covers the 3V Vgs common level with today's uCs (STM32, etc. etc.).
Higher performance parts (lower Rds(on), etc. etc.) are similarly SMT only.
If you do use those breakout boards, try to put on these extra-long (.31" + or longer) pins on them. The extra length allows you to wirewrap a bit easier.
All square-posts wire-wrap excellently. The standard pin length is 0.23", barely fitting two wirewraps per pin. Typical wirewrap is 3x per pin, so .31" really helps (and you probably have the extra depth since you're wirewrapping anyway. Its not like these taller pins are much more expensive than the standard pin).
------------
I think what bothers me about soldering + wire-wrap is that I end up needing to buy Wirewrap tools _AND_ soldering tools. Wirewrap isn't too expensive, but fewer people do it (even fewer than normal through-hole prototypes). There just seems to be far more tools / boards / etc. etc. available that assume soldering.
I think I prefer wirewrap over breadboards or breadboard-like prototypes though. Having the extra height and 3x wire ties per pin "vertically" saves space (at least, board-space), since you're building vertically. Things are definitely more compact than breadboards / breadboard-like protoboards.
But you'll absolutely have to whip out the soldering iron and even use a few specialty tools for a good wirewrapping experience.
T46 pins from Vector Electronics (https://www.vectorelect.com/terminals-wire-wrap.html) and T68 forked-pins pressfit in a "standard hole" (0.042") and also provide more opportunity to wirewrap. You also want the insertion tool btw (one for the T46 pin and another for the T68 pin).
---------
I guess I've been thinking about my lab recently. I think I've decided (for now) that "wirewrap / DIP" prototyping is for simple one-off projects where 5V uC (aka: Arduino Uno or AVR / PIC chips in general) is fine.
Really, through hole is more than usable at 5V, very pleasant to do even. Yeah, you use up a lot more power than the 3V or lower power parts, but its not that big of a deal.
IMO, prototype at 5V, switch to 3.3V or lower if you need a "better 2nd pass" or something. MCP6002 OpAmps (LM358 if you're going higher voltage), AVR DD is a modern uC still in DIP-28 format, jellybeans like 2N2222 and 2N7000 are boring but they absolutely work. Besides, 5V is USB power and absurdly common in practice.
---------
If I'm trying to make some project fit inside of 5mW or 0.5mW envelope with a CR2032 battery for power? Okay, sure, I'll build a custom PCB and order those nice and fancy, superior, SMT parts.
If I'm trying to make a microamp meter, and a 1.000mA +/- 0.001mA constant current source one-off project to help me measure the above CR2032 project? Yeah, that's wirewrapping a few MCP6002 OpAmps together + ATTiny.
My damn Fluke 107 has 4.000 Amp +/- 0.001mA measurement. It can't read this other thing I'm doing, lol.
There are faster 8-bit CPUs available off the shelf, though. Hell, Silicon Labs makes some 8051 microprocessors that'll run at up to 100 MHz.
I think it is this one: https://www.tindie.com/products/circlem/minz-c-small-cased-3...
It would be nice if people would stop pretending CPUs implemented on FPGAs are somehow not real.
The whole "no FPGA" thing is kind of funny when you see people reach for microcontrollers to do their video display component, and those microcontrollers themselves often greatly outperform the main CPU. What's the major difference between dropping in an MCU vs an FPGA?
Aside: A friend of mine has reimplemented the whole C64 VIC-II VDP in Verilog ("VIC-II Kawari") complete with a test harness built around VICE, etc, amazing work; it would be so cool to see someone pony up the cash to get something like that fabbed in fresh new silicon. Then we could have a "true" retro video display controller so we can have a truly authentic neo-retro computer :-)
Also worth considering is that the CPUs themselves are often still available, so what is the incentive to stuff the CPU into an FPGA? Video chips are more difficult to come by, so many projects have to make some sort of compromise.
I'm certain that anyone capable of really understanding the operation of an 8-bit machine without an FPGA is capable of understanding the operation of one with an FPGA. There are tools for programming an FPGA that someone who previously assembled their own boards might not want to learn about, but it's not like they're going to be baffled by them.
I think what's happening here is people want to make and use something with period-correct physical parts for purely aesthetic, nostalgic reasons. I really think that is fine, it's the strangely inaccurate way some people refer to FPGAs that is irksome.
> Video chips are more difficult to come by, so many projects have to make some sort of compromise.
This use of the word "compromise" illustrates the priority aesthetics takes in these projects. The video chip based on a modern microcontroller isn't really the compromise, in engineering terms the fact that you're not using a board with fewer parts and lower power consumption that emulates a z80 on that microcontroller (or something) is the compromise.
> Also worth considering is that the CPUs themselves are often still available, so what is the incentive to stuff the CPU into an FPGA?
Potential ability to make a CPU with a higher clock rate and a board with fewer parts, and all the other reasons you'd use an FPGA.
If the cycle timing and user-facing I/O pins are the same, the main differences resulting from implementation would presumably be in areas like form factor and power dissipation.
I'm sure I've seen ESP32-based emulators running CP/M (or DOS for that matter) discussed on HN and elsewhere.
(and also better highlights why this chip is ??? ... as it has a 240MHz ESP on board. Why not forget about the 8bit microcontroller and just use the ESP? :) )
If all you want is the functionality, and don't care how it works, then why in the world even bother emulating, or implimenting, a z80 with an fpga?
Either use emulation software if you just want the experience, or just use normal contemporary hardware and software with no vintage aspect.
Are you actually running a z80 to run a spreadsheet for actual productivity? Of course not. So you must be using a z80 for some other reason than it's actual functionality. So, the works matter after all.
Pointful uses of an fpga wrt to vintage stuff is replicating parts that can't be purchased any more, or as a rapid development environment, emulating a system so you can develop some other hardware or software for that system etc. Emulating a cpu or a system which still exists, just to use it? Pointless. Any justification mentioning availability, size, cost, reliability, convenience etc, are even 10 or 100x better served by pure software. The amount of cases where the ultra cycle accuracy matters these days is actually quite small as both the emulator software and the contemporary cpus have gotten better while the target has not gotten any harder.
Yes, a cpu implemented on an fpga is somehow not real, It's more real than pure software, but it's still essentially pointless in most cases, because it serves neither the "I just want the experience" as efficiently as pure software, nor the "I want to run the original hardware" at all, not merely less efficiently.
Arguing that a z80 implemented on an FPGA is an "emulated" z80 is as nonsensical as arguing that the Agon's new ez80 running in compatibility mode is an "emulated" z80.
> a cpu implemented on an fpga is somehow not real, It's more real than pure software, but it's still essentially pointless in most cases, because it serves neither the "I just want the experience" as efficiently as pure software, nor the "I want to run the original hardware" at all
The ez80 included with the Agon light was introduced to the market during the George W Bush presidency. That board is all surface-mount electronics with nothing upgradeable or easily repairable on it. It is cool and fun but it's not "the experience" or "the original hardware." (it is its own thing, and that's cool)
> not merely less efficiently
I'm not an EE but it seems to me the "efficiency" you're talking about here does not involve dollar cost, parts count, or electrical current, so that's interesting.
If you implement all the same logic, then it's all the same logic, and by your own logic, the implementation medium doesn't matter.
Performance may differ but that's a seperate facet from "real" if you are going to allow hardware emulation to count as "real".
Even though a gate provided by an fpga is hardware, it's totally different hardware from that same gate in a native ic. You just get a very similar outward effect, IE, a very good emulation.
This conversation is stupid.
And the ez80 is totally different hardware than a classic z80. It seems like a religious argument that a z80 core implemented on an FPGA, offering exactly the same registers and features as a classic z80, is somehow less "real" than the ez80 (which doesn't offer the exact same features as a z80) simply because the FPGA might (or might not) be reprogrammable to a different set of gates.
> You just get a very similar outward effect, IE, a very good emulation.
An emulator is a thing that exists in the world, and this is not it. Someone in another thread mentioned implementing a z80 on an ESP32. That is an emulator.
> This conversation is stupid.
Insofar as it's about terminology... yeah, okay, your word.
But... The cheapest that I can see is about 50€... That doesn't sound very cheap to me :/
Am I just lacking some information or is it actually not that cheap?
Being that 8-bit processors have been gone from interactive systems for so long, anything under a couple of hundred dollars would have been “cheapest.” It would just have to come in under the price of the 80s systems. I could win the same kind of contest by making a 5-bit machine at any price!
Similar is the VGA+PS/2 equipped ESP32 board: https://www.aliexpress.us/item/2251832855287283.html at $12USD or so. But, again, not 8-bit.
I think their claim of cheapest for this Agon light is because a 8-bit home-pc style computer is such a niche/purist space. It's more expensive to build a keyboard/vga ready thing because the chip doesn't have much built-in.
https://www.olimex.com/Products/Duino/Duinomite/DUINOMITE/op...
https://www.cpushack.com/2017/11/14/cpu-of-the-day-fairchild...
It ran the Nova 4 instruction set. We were using it for a cutting-edge project: and all Digital Radio receiver, for EVLF communications to submarines in the 16 kHz range.