FPGA based A500 accelerator

[Promilus]

Well from hobbyist project (even successfully validated) to commercial product there's a long way and not everyone has time, money, energy to do that. Vampire 1 was also open sourced long time ago (with both schematic and TG-based core) and guess what - there's not a single device based on it.

[Samurai_Crow]

Originally Posted by Promilus

Well from hobbyist project (even successfully validated) to commercial product there's a long way and not everyone has time, money, energy to do that. Vampire 1 was also open sourced long time ago (with both schematic and TG-based core) and guess what - there's not a single device based on it.

There are (at least) 2 things wrong with using old FPGAs to replicate CPUs while trying to improve upon them.

#1 is the fact that you're at the mercy of the patent producing CPU designers' patents reaching their expiration date 20 years after the fact in the U.S.

#2 is that the FPGAs used to produce a modern die are magnitudes larger than the ones used to produce a current single-core, dual-threaded CPU like the Apollo 68080.

This limits the chances of retro gear making a comeback to technological advance not taking place for 20 years or more, or being able to make a herculean leap in technology that doesn't infringe upon current patent potfolios of processors that have been in existence for those same 20 years.

I thought it might be neat to develop a more advanced 68090 core using the opcode cracking technology used on modern Intel Core i-series CPUs. That patent doesn't expire until September 29th, 2024 so the only way I could get a sufficient head-start on Gunnar to catch up is to get Intel's permission to use it on a CERN OHL open source core linked above in one of my previous posts or ask them to do it for us. Fat chance on either of those.

[Promilus]

@Samurai_Crow nobody force you to use old FPGA. You can adapt the code provided for Cyclone II (original Vampire V600) to Cyclone V or Artix 7 or even Cyclone 10. Most logic in description language is universal either way.
Now
@1 - you can use whatever opensource 68k implementation in form of softcore without looking at patent holders and improve architecture in whatever way you want (e.g. AC68080!) 68k FPGA needs only P5-like architecture improvements (which AC68080 has ... and more) for performance but even if high performance isn't the target (so TG68 with decent memory access) there are also opportunities to integrate things like RTG, IDE, USB HID (or maybe even SATA as there's LiteSATA implementation in FPGA). It's not like most amiga users really need pentium II class performance and 3D graphics which was hardly used 20 years ago and neither is now. Most users need straightforward and compact solution which addresses their needs for fast storage, high res workbench and decent performance overall and all that with a price tag much smaller than decent, modern netbook.
@2 - I'm not sure what's your point here. You take particular FPGA which will be able to host your softcore. Not the other way around. You can easily buy medium density latest tech FPGA which will fit AC68080 and it will be smaller and more efficient than Cyclone V.

[Samurai_Crow]

Re:number 2
Building an open-source CPU core for an FPGA is often the wrong tool for many jobs because it runs at a fraction of the speed of a modern CPU either in clocks or in architecture. Dr. Michal Schulz has gotten good results with AArch64 and the Emu68 JIT for a fraction of the cost of an Apollo-core based computer.

If the core isn't going into a cheap ASIC, the FPGA should focus on the chipset core where timing is more critical and the original clockspeed was lower.

[Promilus]

@Samurai_Crow - I think you didn't really think that through.
1st error - PiStorm is open initiative and not one to make (a lot of) money and except CPU interface module it uses popular (but nowadays hard to get) hardware made by 3rd party vendors. It's also pure software emulation on many levels (which means not everyone enjoys that - you can emulate whole amiga on RPi as well) so comparing it to the commercial device with their own hardware design, logic design, utility software design and hardware manufacturing is laughable. And I write this while still being rather skeptical about many of Apollo team choices. But when it comes to business - well their business model is rather well conceived.

2nd error - emu68k is JIT. AC68080 is not. emu68k won't work with plenty of demos using self modifying code where AC68080 will work with most of those. 68k I/O-wise FPGA implementation of AC68080 does work more accurate (and efficient) than half programmable logic and half bit banging of SoC. For most of advanced features (more usb devices, networking etc) you'll need linux anyway. It's not typical turbo card and not really a competitor to FPGA based solutions. FPGA based solutions are made to be as close as original 68k in code execution and also as a compact multi-tool, having connectivity, storage and graphics on-board. And while PiStorm can get the same (and better, and cheaper) for many amiga users it just won't "feel right" exactly due to the fact it is purely software emulation. And truth to be told you could've gain even more speed with x86 based SBCs - that's just a single step... is it in right direction?


Original Vampire was made on rather poor FPGA. IIRC just 8k LE. And it barely (but still) hold "phoenix core" which was way more powerful than bare TG68. And that was with SDRAM. Now ... XC6SLX16 evaluation board with equivalent of 16k LE (actually less) and DDR3 SDRAM and voltage regulators costs ~40$. Just add interface board with buffers for voltage level translation and socket to CPU - et voila, cheap and fairly powerful amiga accelerator. Add some more money and you'll get XC7A35 (artix 7 with equivalent of 35k LE, that's roughly Vampire 2 level so cheap version of AC68080). Add some more and there's XC7A100 which just might hold whole AC68080 with RTG, IDE and networking. And that's ~100$. As you can see it is fairly possible to get decent hardware in reasonable price... but it ain't that easy with softcore itself. And even with PiStorm it isn't all that easy.

[Samurai_Crow]

@Promilus
Since the software expense is grossly higher than the cost of either an FPGA or AArch64 compute unit and the Emu68 JIT and the AC68080 are both still under development, the jury is still out on cost effectiveness. In the end all will be revealed but until then, we shall see.

[apolkosnik]

With Vampires you get black box that is essentially a door stop once those guys decide to stop doing whatever they want. At least with emu68 or Musashi you can modify the code. Also with Minimig and MiSTer you can tweak the cores to your liking, even mess with TG68K.C.

[Bruce Abbott]

Originally Posted by apolkosnik

With Vampires you get black box that is essentially a door stop once those guys decide to stop doing whatever they want.

Vampires brick themselves if 'those guys' stop doing 'whatever they want'?

At least with emu68 or Musashi you can modify the code. Also with Minimig and MiSTer you can tweak the cores to your liking, even mess with TG68K.C.

How do I do that on my Amiga?

[Samurai_Crow]

Originally Posted by apolkosnik

With Vampires you get black box that is essentially a door stop once those guys decide to stop doing whatever they want. At least with emu68 or Musashi you can modify the code. Also with Minimig and MiSTer you can tweak the cores to your liking, even mess with TG68K.C.

I agree about open source and open cores. I'm trying to figure out how to chain a non TG68k.c core with real pipelining onto a Minimig core. Unfortunately, my SiDi lacks the capacity of a MISTer and the Intel patent I want to use won't expire until September of next year.

[jbilander]

If anybody is interested, I released a FPGA-board into the public domain a week ago or so, fully OSHW (Open Source Hardware).

https://github.com/jbilander/SF2000

For now it uses a MC68SEC000 CPU onboard but I think it should be possible to fit a 68k-softcore (like TG68k or fx68k) in the FPGA, this is a bit over my head though so I would need help with this, the Gowin FPGA currently used has 8640 LE:s so a 68k core could fit I think.

With the MC68SEC000 it runs stable up to 42 MHz, yes I know it is a low-end card but the system is really responsive if KS is loaded and run from the onboard 1 MB FlashROM-chip.

If anyone wants to take a stab at this then let me know. If you want to contribute to the project I can probably provide you with a blank PCB but it is a DIY hobby project so you would have to order components and solder it yourself.

One thing left to do is to write a 4-bit SDIO controller for the MicroSD-card-slot. It would be great if Mike's previous work could be repurposed for this. If you read this Mike, please don't let your awesome work fall into oblivion Anyway, if not then I guess we reinvent the wheel.

Happy Hacking

[AJCopland]

Nice work jbilander, that's a really cool of you to release it

[mfilos]

Reminds me when Igor was making his first Vampire V1 using the TG68k core :P
Nice work man. Wishing all the best for your efforts \o/

[Promilus]

Well he did succeed. Of course there were some unresolved issues with this (along with reset not working properly by design flaw). Apollo did produce "phoenix core" being scaled down AC68080 with no FPU and obviously no SAGA or RTG as well. AFAIK after that no further support was given to those cards and they died out. To this date only Vampire and Apollo cards uses FPGA for 68k CPU. Yeah, there's unamiga, mister, DIY-A586 but those are h/w emulation of basically whole amiga hardware. And A586 which uses TG68k as well (barely) fits spartan6 slx9. Too bad no real FPGA alternative to Apollo cards was finalized.

[jbilander]

Thank you guys for the kind words.

I now have some spare SF2000 PCBs made by JLCPCB that I can let go for cost if anyone is interested in tinkering with this card. Can ship to EU countries without hassle. If you are outside the EU it's easier/cheaper to order yourself from JLCPCB (gerbers+drill zip are in the github repo). PM me if you are interested in one of these blank PCBs.

This time in blue...

[Lord Aga]

I can vouch for the awesomeness of jbilander's work

Top quality stuff!

[jbilander]

Frontier intro running on the MC68SEC000 @ 42 MHz...
https://drive.google.com/file/d/1k1B...ew?usp=sharing

booting cold and having the small A500_IDE_LED_board connected for driving the A500 drive activity LED (for both floppy and IDE).
https://drive.google.com/file/d/1fvl...ew?usp=sharing

[graffias79]

This would make a fantastic upgrade for my A2000. I don't mind keeping things 16-bit. I hope I can get my hands on one at some point. I can solder okay but I'm not great with the passive components. Hopefully someone will sell these ready-made.

[ex68k]

Originally Posted by Samurai_Crow

I agree about open source and open cores. I'm trying to figure out how to chain a non TG68k.c core with real pipelining onto a Minimig core. Unfortunately, my SiDi lacks the capacity of a MISTer and the Intel patent I want to use won't expire until September of next year.

Are you sure, you would have everything working before the patent expires?

[Samurai_Crow]

Originally Posted by ex68k

Are you sure, you would have everything working before the patent expires?

I traded off my SiDi anyway. Not developing for it anyway.

[jbilander]

Originally Posted by graffias79

I don't mind keeping things 16-bit.

Yes I agree, Keep the 16-bit machines 16-bit. Leave the 32-bit CPUs (020+) for the 32-bit machines.

The other day I tried this kind of IDE-adapter (from Ali) on the SF2000. It plugs right in and works well after removing the two transistors and the activity LED comes to life blinking after soldering in a bodge wire for the /Active (pin 39) signal. Pics below...