As (one of) the alleged anti PPC villain of the site, I have a serious question.

My objection to PPC is simple. Excluding second hand Mac (which I have no problem with, beyond their fundamentally irreplaceable long term nature), it's the cost and hardware fragility of the solutions that are available.

So, how practical is a softcore PPC, one designed specifically for the fundamental limitations of 32-bit addressing and single core? I know there are numerous FPGA that come with an embedded PPC in them, but I'm thinking a pure soft core solution. Remove any complications to performance that attempting to support more than the basic 32-bit feature set would add.

Is it feasible at all, if so, what would be the sort of upper limits on performance, etc. Is there a middle ground somewhere between the 60x class and G3/G4 that os within reach, or is synthesized hardware just too slow?

_________________
Doing stupid things for fun...

Given it's supposed to be a RISC architecture, you might hope for some simplifications relative to say the 68080 design. You have two pretty big register files (IU/FPU) but equally you don't have effective addresses to worry about (beyond load/store operations).

_________________
Doing stupid things for fun...

@Karlos

Quote:

Karlos wrote: Given it's supposed to be a RISC architecture, you might hope for some simplifications relative to say the 68080 design.

Depending on whom you ask the CPU performance of the "68080" is somewhere around Phase5 PPC (with it's issues introduced by having the 68k on the same RAM bus) and clearly below an EFIKA.

So even if a fake PPC would be 2x as fast as the fake 68k it would still barely touch the bottom end of NG PPC HW.

So unless they get to a much faster/bigger FPGA (without breaking the bank) going that route is pointless.

Really digging into why emulating PPC on ARM or x86 is so relatively slow and fixing that might get useable results.

_________________
- We don't need good ideas, we haven't run out on bad ones yet
- blame Canada

@Karlos

Well, why slow FPGA, when you can use existing x86 hardware and run OS4/MorphOS via QEMU?

@pavlor

Well you know how it is. People seem to want hardware solutions. I don't have a problem with that per se, it's just that the solutions are ridiculous - essentially you spend thousands on a system you can't use properly unless you run Linux and then discover the PC you already had does that job immeasurably better.

So I'm thinking try coming at the problem from the other direction. Can cheaper, more compatible PPC hardware be implemented using tried and tested FPGA methods?

_________________
Doing stupid things for fun...

@Kronos

Quote:

So even if a fake PPC would be 2x as fast as the fake 68k it would still barely touch the bottom end of NG PPC HW.

I suppose that's what I'm curious to understand. Do the proposed architectural simplification benefits of RISC play out compared to the CISC model that has been implemented already? A whole FPGA dedicated to the task, nothing peripheral. Or at least nothing like custom chipset emulation.

Last edited by Karlos on 10-Feb-2024 at 01:49 PM.

_________________
Doing stupid things for fun...

@Karlos

Quote:

Karlos wrote: Can cheaper, more compatible PPC hardware be implemented using tried and tested FPGA methods?

Short answer: No

Long answer still no

If such a project would aim at MorphOS (or LinuxPPC) users, won't be even close in price or performance to anything Mac ranging from a 450MHz G4 PowerMac G4 to the G5_Quad.

For OS4 you have 2 options:

1. the official route, your 300$ HW project will be a 2000$ product and still not ship a decade later
2. hacking and patching some version of OS4 to make it run on it. Better spend that
energy and skill to "merge" Moana with the Peg2 version to get it running on at least some of the dirt cheap and fast (by comparison) old Macs.

Fake 68k only made "sense" because the 68060 is the fraction of the performance of a G5 despite only 10 years between them. That 10 gap is much bigger than the one between 2004 and 2024 HW.


Edith:
I would suspect that doing CISC in FPGA might actually be easier than RISC as these are more suitable to do complicated function at one speed compared to doing simple functions faster.

Back in the days with making real chips it was the other way round, hence everybody seeing RISC as the future.

Last edited by Kronos on 10-Feb-2024 at 02:17 PM.

_________________
- We don't need good ideas, we haven't run out on bad ones yet
- blame Canada

@Kronos

Quote:

Edith: I would suspect that doing CISC in FPGA might actually be easier than RISC as these are more suitable to do complicated function at one speed compared to doing simple functions faster.

That's her told.

_________________
Doing stupid things for fun...

@geen_naam

Quote:

geen_naam wrote: @Karlos LOL!!! We should have an annual vote on the dumbest topics. This one will surely win. The reason for the cost of AmigaONEs like the X5000 and A1222 is not the PPC. The reason is that they're designed by a professional company (Varisys). Cost of development and production of a small batch will easily add up to a 6 digit figure. This cost will have to be amortised over expected sales.

While that for sure is a contributing factor it simple can't be the only one.

Plenty of other small run commercial HW that don't cost a fortune.

To many entities wanting a cut to many useless c##p the to big boards running up cost and a ton of interest and other costs running up in the decade it took (and still takes) to bring them market.

Or in short the greedy and incompetent selling into a market of gullibly fanboys.

_________________
- We don't need good ideas, we haven't run out on bad ones yet
- blame Canada

@geen_naam

Quote:

geen_naam wrote: AmigaOS4 will also doable. Hyperion would be open to bundle OS4 when you provide drivers.

Sorry, missed that one.

It is quite clear that Ben cares far more about anybody else NOT winning than about loosing himself.
Might also be true for other "players".

So OS4 would be a problem unless you put $$$ in hand to calm down the egos.

_________________
- We don't need good ideas, we haven't run out on bad ones yet
- blame Canada

@geen_naam

Who gives a shit why they are so expensive? The fact is they are and they are failure prone with basically zero hope of repair. Not a great advertisement for your "professionally designed hardware".

If you want new users, you need to lower the bar to entry or change the platform. Preferably both at once.

The question around a soft core is to understand what is possible as I haven't kept up to date with the progress in FPGA capabilities in recent years. And if it isn't possible, that's fine.

One thing is certain. Stay on PPC for maximum irrelevance.

_________________
Doing stupid things for fun...

@geen_naam

If you'd asked him something similar bout Amiga inc in 2002-2006 you would have gotten the same reply despite being plenty of blood in the water at that time.

"open to idea" means nothing from a men whose profession is twisting words beyond their common sense meaning.

Hence it make no differences if you do a product based fake PPC in FPGA or real legacy/NOS chips from whatever FreeScale is called this week, once you get to the SW/license side things will slow down and cost a fortune.

Hence it took 8+ years (might still get to the full decade) for from the 299$ "nah that missing/broken FPU won't be a problem" Tabor to the still not shipping 1699$ A1222.

Last edited by Kronos on 10-Feb-2024 at 04:16 PM.
Last edited by Kronos on 10-Feb-2024 at 04:14 PM.

_________________
- We don't need good ideas, we haven't run out on bad ones yet
- blame Canada

@Karlos

We amiga users use ppc for 27 years since 1997. accept that.
leave ppc as it is. start working on something that will be worth use on arm.

@ppcamiga1

Quote:

We amiga users...

You're not an Amiga user. Prove me wrong.

Last edited by Karlos on 10-Feb-2024 at 05:37 PM.

_________________
Doing stupid things for fun...

@Kronos

Quote:

Hence it took 8+ years (might still get to the full decade) for from the 299$ "nah that missing/broken FPU won't be a problem" Tabor to the still not shipping 1699$ A1222

This exact thing. Farces like this are why I ask if there's an alternative route for "new" hardware better suited to the limitations of the platform.

How many Vampire / PiStorm sales versus all X1000+ machines put together?

The potential user base exists it's just not an attractive prospect primarily due to the expense and questionable compatibility.

green_naam can deride and mock all he likes, but the facts are simple: PPC just doesn't excite people like it may have done back when it was first suggested. If it were available at a much lower price then people might be tempted.

_________________
Doing stupid things for fun...

@Karlos

Quote:

Karlos wrote: @Kronos This exact thing. Farces like this are why I ask if there's an alternative route for "new" hardware better suited to the limitations of the platform. How many Vampire / PiStorm sales versus all X1000+ machines put together?

If Cloanto/Hyperion could gatekeep 68k AmigaOS as good as OS4 that number would matter, but it doesn't.

If someone hacks OS4/MorphOS to run on alien PPC or if someone make a HW that is indistinguishable from supported HW ......
.... legal s##t would fly.

Using old Macs or building a sensible board with NOS chips will have a far better bang/buck ratio than anything PPC not just today but at least for another decade.

That HW alone is useless and that where the pain starts regardless of the route you went.

FPGA PPC is a solution trying to fix an issue it clearly didn't understand.

_________________
- We don't need good ideas, we haven't run out on bad ones yet
- blame Canada

@Kronos

Quote:

FPGA PPC is a solution trying to fix an issue it clearly didn't understand.

It's not even a solution. The top clock speeds are still hundreds of MHz. I'd hoped they'd advanced a bit in the last few years but not by that much.

Oh well. Back to being robbed blind for a half working system.

_________________
Doing stupid things for fun...

@Karlos

Quote:

. The top clock speeds are still hundreds of MHz. I'd hoped they'd advanced a bit in the last few years but not by that much.

It seems they get bigger (on smaller nodes), but not that much faster, just like CPUs and GPUs.

Which is nice if you can do stuff in parallel but mostly useless if you want "single core" style applications/design to go faster.

_________________
- We don't need good ideas, we haven't run out on bad ones yet
- blame Canada

@Karlos

Quote:

So, how practical is a softcore PPC, one designed specifically for the fundamental limitations of 32-bit addressing and single core?

It could be possible but ... I don't think anyone would like to buy an 600$ system with a 300MHz processor (except some Vampire4 customers). And even AOS4 fans would reject it without proper AOS4 support.

And just because this is called an 'Amiga Forum' I would suggest to got to a cloned AmigaOS supporting more performant hardware, as long as AOS4 isn't available.

Last edited by OneTimer1 on 10-Feb-2024 at 07:46 PM.

@Karlos

I suggested a while ago, prior to the A1222 respin, that A-Eon place a FPGA on the Tabor that would be configured to be a standard PPC FPU. Code wouldn't necessarily have to be emulated, just redirected from the SPE to the FPGU.

_________________
I sold my SAM460ex lite... waiting for money to buy a Raspberry Pi... or a Classic A1000 with Buffee... or an A1222... and OS4.3 FE update 11