exxos's Atari, Amiga & retro forum

But does it also run new, natively built software which can harness the full capability of the machine?

terriblefire wrote: 15 Jun 2022 11:17

sporniket wrote: 15 Jun 2022 10:28 About the FPGA, are we talking about entry level ones (like e.g. the one on the DE-10 nano, spartan6, ecp5,... with boards around 50-150$€) or also up to the high-level ones (1500-2000$€ -around the price of a Falcon030 nowadays- and beyond) ? (Honestly curious about that)

The ones i'm most familiar with is the big end Virtex 5/6 chips. They are massively impressive bits of kit but they're designed to make compute *wide*, e.g. single cycle DSP type operations, not flat out clock speed. I'd need to check what the current FPGAs are up to but these guys were expensive and awesome but they cannot clock like an ASIC can.

This was the reason that bitcoin mining on FPGAs only lasted a year or two before it moved over to ASICs.

The spartan 6 FPGA could only do 80Mhz or so with a fully pipelined and wishbone compatible bus ao68000. That design was very clean but the S6 couldnt go faster.

You can simply go look this up because FPGAs have pin to pin and gate to gate specs that basically limit your clock frequency. I've not seen one better than 5ns but they may exist. 5ns is roughly 200Mhz max. Beyond that you end up with specific FPGA fabric designed for the job. E.g. the Spartan 6 "T" edition has parts that can do PCIe.

Thanks for the hindsight

sporniket wrote: 15 Jun 2022 15:57 Thanks for the hindsight

Things may have moved on a bit since i last used them but from what i could see FPGA type stuff was moving to cpus with synthable FPGA fabric attached like the Zync cores etc. It didnt seem to be looking to go for faster and faster clocks. I think there are fundamental physical limits to the clock routing network that limit overall speed.

Thats not to say you cant do awesome stuff with them but because they use look up tables to emulate logic gates the speed of those lookups is finite.

stephen_usher wrote: 15 Jun 2022 12:05 Well, what made the Amiga special in the first place was that it was at least a generation ahead when it came to the graphics capability, both hardware and software support in the OS. The CPU was no faster than the other machines on the market at the time.

If you were to "bring it back" you really wouldn't be looking at an m68k machine (the later AmigaDOS machines weren't based upon this either) and really you would have to look at basing it upon commodity hardware, either PC based or, probably better for future-proofing, 64 bit ARM and then building the OS and graphics manipulation support on top. If you wanted it to be an "Amiga", as in running the old software then you would need to have an integrated m68k VM which would seamlessly run the old code and simulate the old hardware.

All this work though would be for a tiny niche market where it would never be profitable.

While I agree that the CPU of the Amiga wasnt anything amazing it was on a par with things of the time. You didnt notice it was slow. So that would at least need to be true of a "return" type system.

But the rest of your argument makes total sense. Back in 1985 there wasnt a computer that could do all the things an Amiga could. So people bought into it. Thesedays computers all do everything. Which is boring... dull.. Choosing a computer is like choosing a washing powder now. Its all about marketing and packaging.

So basically there will be no more complete ecosystems like Amiga again. MacOS, Linux and Windows are converging together ... they'll all run on most hardware and they do mostly the same stuff. Computing is as dull as ditchwater thesedays.

mgi wrote: 15 Jun 2022 12:10

stephen_usher wrote: 15 Jun 2022 12:05 If you were to "bring it back" you really wouldn't be looking at an m68k machine (the later AmigaDOS machines weren't based upon this either) and really you would have to look at basing it upon commodity hardware, either PC based or, probably better for future-proofing, 64 bit ARM and then building the OS and graphics manipulation support on top. If you wanted it to be an "Amiga", as in running the old software then you would need to have an integrated m68k VM which would seamlessly run the old code and simulate the old hardware.

I think this is exactly what the Pimiga is - running on ARM64 within the Pi4 and executing the original code.

... or ARAnyM (in Atari case).

terriblefire wrote: 15 Jun 2022 19:14 So basically there will be no more complete ecosystems like Amiga again. MacOS, Linux and Windows are converging together ... they'll all run on most hardware and they do mostly the same stuff. Computing is as dull as ditchwater thesedays.

This is why I have zero interest in modern PC systems, and many other gadgets in my hobby. The frequencies even the most basic systems works at is Ghz or many hundreds on Mhz. These basic systems connect to anything with all protocols available on chip. Now Atari's and Amiga's interest me due to their limitations, and the coding to push the systems to the limits. Also as stated these systems are not useful now, but they are are DAMN fun!

terriblefire wrote: 15 Jun 2022 11:17 The ones i'm most familiar with is the big end Virtex 5/6 chips. They are massively impressive bits of kit but they're designed to make compute *wide*, e.g. single cycle DSP type operations, not flat out clock speed. I'd need to check what the current FPGAs are up to but these guys were expensive and awesome but they cannot clock like an ASIC can.

High end FPGAs aren't just bigger than low cost ones, they are much faster. Current generation models can run internally at the 1 GHz level. For most tasks they would easily outperform processors like the RPi. The RPi might have a faster raw clock, but it must process sequentially, while a FPGA can process in parallel. A processor would need many cycles to accomplish the same task that an FPGA can do in a single one. Obviously that for sequential processing a fast CPU would still be faster, but this is not our typical task. Of course the whole thing is probably just academic as high end FPGAs prices start at 4 digits!

Yes, of course that ASIC is faster. But unless you are going to fab chips, we aren't really talking about ASIC vs FPGA.

The spartan 6 FPGA could only do 80Mhz or so with a fully pipelined and wishbone compatible bus ao68000. That design was very clean but the S6 couldnt go faster.
...
Things may have moved on a bit since i last used them but from what i could see FPGA type stuff was moving to cpus with synthable FPGA fabric attached like the Zync cores etc. It didnt seem to be looking to go for faster and faster clocks. I think there are fundamental physical limits to the clock routing network that limit overall speed.

This doesn't have much to do with technology or physics. It is just a purely business issue. There is no profitable market to produce fast low cost FPGAs. Low cost FPGAs are simply produced "slow", with about the same performance for more than a decade. The main manufacturers are not interested in making them any faster, among other things because they don't want to create cheap competition for their high end products that are much more profitable. Low cost FPGAs evolve with more features, bigger, and much more better power efficiency. That was the main goal, especially low power, for the latest generations of low cost FPGAs. You want speed? Get a high end FPGA.

Now, there is a middle range of FPGA families, between the low cost and the high end ...

Unfortunately since Intel bought Altera, and now AMD that bought XIlinx, they seem to disregard the low cost range entirely.

ijor wrote: 16 Jun 2022 15:12 High end FPGAs aren't just bigger than low cost ones, they are much faster. Current generation models can run internally at the 1 GHz level. For most tasks they would easily outperform processors like the RPi. The RPi might have a faster raw clock, but it must process sequentially, while a FPGA can process in parallel. A processor would need many cycles to accomplish the same task that an FPGA can do in a single one. Obviously that for sequential processing a fast CPU would still be faster, but this is not our typical task. Of course the whole thing is probably just academic as high end FPGAs prices start at 4 digits!

Yes, of course that ASIC is faster. But unless you are going to fab chips, we aren't really talking about ASIC vs FPGA.

I wasnt aware the latest high end FPGAs had got to or above 1Ghz. The simple physics of the clock routing network at that speed are mind boggling. My PhD is in an area related to nano technology, from a nano technology department, and it makes my brain hurt even thinking about how this is accomplished.

That said, and as you allude to, what is the point of getting a 500Mhz 680x0 CPU using a £1000 FPGA when you can get the same performance using software defined CPU technique for £100. Its also orders of magnitude easier to code and fix bugs in a SD-CPU. Software Defined CPUs are the future.

Attaching more MIPS to a slow bus only gets you so far.
This has been the story of the last 30 years of desktop computing and we’re only just turning the corner on that.

Nobody who is paying attention in hardware could think an Amiga would ever be competitive even with new internals.

OTOH higher number is important for one reason, and that’s because modern development practices are spendy! If you want more new/good Amiga development then more MIPS and ram helps.

To me who is a long time fan of "next-gen" Amiga systems (you know, the PowerPC thing we don't mention these days), this discussion just feels like groundhog day.

It was almost 20 years ago, when we had MorphOS (and a bit later OS4), running 68k software on PowerPC, with faster-than native speeds on few hundred Mhz PowerPCs. Here it is the slowest supported MorphOS hardware, a 396Mhz Efika, running an Elude '060 demo (68k binary) with RTG support, faster than any real-68k-based Amiga ever could.) And this is hardware from 2006-7, and was super low-end even back then. Not to mention all the advanced OS features that are there for more than a decade and a half now, but only just getting into 68k systems, mostly in barely usable forms.

And yeah, maybe PowerPC wasn't your thing, but long before PiMiga (or even a Raspberry Pi) was a thing, we had Amithlon, that was doing essentially the same on x86 PCs...

But then the retro wave happened, people came back to the scene, took a look at next-gen, went like "yeah, it can't run Superfrog (or insert your favorite OCS game here), so whatever", and went on with their A500 business. The only problem is of course, whoever doesn't learn from history is condemned to repeat it, so here we go again.

But OK, lets ignore all that, never happened. But even when it comes to expanding classics, based on experience gathered with the next-gens, in 2014(!) at kinda the start of the FPGA CPU mania, I actually made a forum post (in Hungarian), predicting that cheap ARM based accelerators with JIT compilers would make a takeover, and the FPGA thing is essentially a dead-end when pursuing CPU performances where one can really make a difference. (Read: allows software that's not reasonably possible on a '060 Amiga. Say, a modern browser, or video player, or semi-modern 3D stuff.) I even tried to get a few hardware guys to work together on an early proto, but everyone dismissed the idea, as something "unrealistic". You know, everything is impossible until someone proves otherwise. Kudos to Claude (et.al.) for the PiStorm. :)

But that still doesn't fill the software gap. And MorphOS has a modern browser, that lets you watch YouTube & stuff, or use Google Docs/Office365 and similar "modern" things in an Amiga(OS)-like environment. Even if I'm a fan of ARM-based accelerators for classic, I wonder when 68k software emulated on a super-fast ARM gets here. 15 more years, I suppose. But of course, I also agree that modern computers are dull. Even modern-retro computers like the old Apple junk MorphOS runs on is a lot duller than any classic Amiga. No argument there. However, is it duller than a Raspberry Pi? I'm not so sure... I guess, everyone has their preferences when claiming things as dull.

Anyway. Never mind me. I crawl back under my wet, slimy, moldy rock with a PowerPC logo on it.