@pixie

Quote:

I would love to see a solution of a fpga 1200 amiga so that you still was able to connect past accelerator boards, clock port devices, (in the same line you today have some consoles being emulated which accept cartridges), and an integrated flicker fixer for running on a crt because I am feeling lucky.

Minimig aimed for games only, and this is sadly the case for most of its most successors.

accelerator ? No, but a CPU softcore with the speed of 25MHz 68030 and some MB of RAM
clock port devices ? No, not even a parallel or serial adapter, maybe Joystick ports (no paddles)
integrated flicker fixer ? Yes, some version have it and some even have an integrated RTG.
harddisk ? No, but a harddrive on SD-Card
floppy ? No, but it can read ADFs from SD-Card

Any further development of this hardware ? No, because most people don't care for stand alone Amiga replacements, they prefer expensive accelerators for their dying A500/A1200 systems.

bhabbott Quote:

The second paragraph is important because it stresses the desire for an authentic retro experience. So a modern reproduction of a 68k CPU might be considered 'retro' if it performs like they did in the 1990's, but one that runs at the speed of modern systems - and therefore attracts software that needs it - is stretching the definition too far.

Many modern CPU cores are fully static designs which means they can operate at any clock speed between zero and the max clock speed (68060, 68040V, mP6/Vortex86 core).

https://www.vortex86.com/products/Vortex86DX2 Quote:

Vortex86DX2 is a high performance and fully static 32-bit X86 processor with the compatibility of Windows based, Linux and most popular 32-bit RTOS.

Some Amiga 68060 accelerators already have adjustable clock rates controlled by software selection. The software clock selection is shown at the beginning of the following video and then RetroCengo plays games with 68060@105MHz which benefit from the speed.

ZeeWolf 1 and 2 turbotest on Amiga 1200 68060 105MHz Warp1260 system
https://www.youtube.com/watch?v=e8VFWiPuj-g

The 68060 also has a half cache, instruction cache off, data cache off, etc. options too. Newer CPU cores don't need to be cycle exact anymore. There are other RetroCengo videos showing him playing a wide selection of Amiga games on the 68060 and a YouTube video of a guy with a CD32 who selected a 68LC060 without FPU for his CD32 Terrible Fire because it offered more value than a 68040 and the FPU is not used for many games.

The 68060 is comparable to the mP6/Vortex86 core.

spec | 68060 | Vortex86
bits 32-bit 32-bit
logic fully_static fully_static
type superscalar superscalar
design in-order in-order
int-stages 8-stage 8-stage
int-pipes 2+branch 3
FPU yes fully-pipelined
SIMD no yes
transistors 2,530,000 3,600,000 (both with 8kiB I+D caches)

https://en.wikipedia.org/wiki/MP6
https://www.cpu-collection.de/?tn=&l0=co&l1=Rise&l2=mP6

The Vortex86 core is more advanced as it was released in 1998 instead of 1994 like the 68060. They both have low power features targeting embedded use which gave the mP6/Vortex86 core a 2nd life. The 68060 in-order design was much lower power, had superior integer performance and used fewer transistors (less area) than the Pentium in-order design giving it a clean sweep in PPA for embedded use. It's too bad the 68060 was hidden away and forgotten like the Amiga chipset and AmigaOS when it is a great design for embedded use as demonstrated by ex-Amiga engineers trying to obtain what the Amiga had and having to build similar but often inferior systems instead. Only Amiga makes it possible.

bhabbott Quote:

There's a lot more to a PC than just a CPU that runs x86 code. Some things you won't find on a modern motherboard:- - ISA slots. - parallel printer port. - RS232 serial ports. - Floppy drive port. - IDE hard drive port. - CPU with performance identical to an 8088, 8086 or 80386SX.

You mean old PC I/O like in the following pics?





https://linuxgizmos.com/vortex86-cpus-gain-new-life-with-linux-kernel-detection-and-a-3-5-inch-sbc/

It's missing IDE because it has SATA instead and an IDE connector is as useless as a SCSI connector today because the drives aren't made anymore. There is no floppy port either but missing I/O may be available through SBC headers and GPIO pins. This VDX3-6726 is already expensive at $342 USD where more of the less used I/O could be available through GPIO pins to reduce the cost (it's not expensive for the SoC to support old I/O). This "SBC is designed to replace boards based on aging AMD Geode LX800 processors, which are being discontinued." It was designed for a niche embedded market use which wanted particular I/O options and not for the retro PC market. The article talks about 3rd party Linux support in 2022 for this ancient/retro SBC using mostly 20+ year old tech. This is real hardware and gains developer support even though it has ancient/retro x86 and I/O support!

Actual 8088, 8086 or 80386SX cycle exact and clock exact performance is only needed for the earliest games. The same is true for the 68000 and after that cycle exact and clock exact wasn't as important. Clock speed selection and cache options may still be useful though. In the end, a vortex86 core can play most x86 games. A 68k SoC with a 68000 core, 68060+ cores and more clock speed control could play most of the huge 68k game library. Whether this meets your idea of retro doesn't matter if it sells to retro fans. There will always be a few pain in the asses that complain it doesn't have this or that and don't want to create it with the headers/GPIO so you create a nice high margin retro I/O board for them.

bhabbott Quote:

For that last one you might say, so what? It's very important if you want to know what it was really like to eg. play Doom or run Windows 95 on a 16MHz 386SX. That's why I bought two 386SX motherboards and an Amstrad PC2086 (and ISA video and sound cards, hard drives etc.) even though I already had several older Pentium motherboards and systems. The same thing applies to the Amiga too. Some people are happy enough with emulation, but I need a stock A500 to get the true experience of one. I never finished the TF530 card I started building because there was little point when I already have a 50MHz 030 in my A1200 (with totally period hardware including the accelerator card, hard drive and PCMCIA network card that I have had since the 90's). It's also why I bought several sealed boxes of 'new old stock' Fuji DD disks, so I could continue to enjoy using high quality floppy disks (that aren't moldy and full of crap that will destroy your drive) like I did back in the day.

Original Amiga hardware isn't going anywhere but it is approaching 40 years old and becoming difficult and expensive to maintain in like new form. It would be great if there was more like original Amiga hardware available but it is not for everyone as it is expensive, large, lacks modern I/O, lacks features and lacks usable performance for modern general purpose computer use. The 68k Amiga needs more affordable and available hardware that has good backward compatibility but moves the 68k Amiga forward, is attractive to a wider market and is real hardware developers can target.

bhabbott Quote:

Unfortunately many YouTube videos purporting to show retro computers in action are actually done on modern machines that usually don't have identical performance. This also applies to most videos of other retro machines too. Luckily a few people like Zeusdas have stuck to using real hardware.

I've found the same issue with YouTubers using emulation or simulation and not being transparent about it. It is important that the museum pieces be maintained but people don't drive in a Model T to get groceries very often either. Technology brought improvements which should not be ignored for new products to be appealing. I'm suggesting 68k Amiga hardware that is more faithful and compatible than virtual 68k Amigas that don't use the Amiga DMA, have full time 100% CPU load and wastes 3/4 of memory. I'm suggesting 68k Amiga hardware that is more faithful and compatible than a fat PPC CPU and no Amiga chipset support. I'm not telling anyone to get rid of their old 68k Amiga hardware. I just want more affordable and practical 68k Amiga hardware so the user base can grow again.

Last edited by matthey on 10-Nov-2024 at 07:29 PM.

kolla Quote:

As I wrote - I don't care about 3D, and there is barely anything to sacrifice in the first place. MMU on the other hand _is_ here and would be a sacrifice if it by some magic disappeared, but it won't - I really don't grasp what your point is here, virtual 68k Amiga with MMU has existed for like 2 decades already and isn't going away.

MMU and chipset emulation are two of the most difficult to support and more processor intensive workloads for an emulator. If a 68k+ emulation spec is introduced, these are natural places to cut back to improve overall emulation performance. AmigaOS 4 could remove MMU support for the 68k+ spec and would likely work fine. The 3D support can be passed through to the underlying hardware so is not a problem.

kolla Quote:

Just the bare minimum - the exact same Roadshow that exists for 68k, except that updates apparently are never pushed to end users. Olsen would be the first to express that Roadshow is not by any means a full featured "modern" IP stack. ... Also bare minimum, Poseidon on Amiga 68k (and AROS) is more capable.

Ouch! I don't hear AmigaOS 4 users rejecting your assertions on what is left of a pro AmigaOS 4 forum. I wish I could say all the value is in the hardware but it is so far from competitive that production runs are down to the hundreds.

kolla Quote:

Obviously you don't have any OS4 system, do you.

Warp3D and basic drivers for Voodoo 3-5 and Permedia are not included in AmigaOS 4 for free like is available for the 68k for free?

kolla Quote:

What? You asked "what about compiling for ARM RPi OS instead and forget the 68k Amiga emulator?" - this obviously exists already, and has nothing to do with Amiga.

My point was that there is no reason for the 68k Amiga to exist anymore if all it is is a virtual machine kludge. It is kind of like the Java VM which pretty much disappeared when developers realized it was more efficient to compile Java code into native code. A virtual machine Amiga makes sense where the source code was lost or the software hits the hardware but otherwise it is just a kludge that introduced a bottleneck. Most AmigaOS 4 source code is available and it doesn't hit the hardware yet 39% of AmigaOS 4 users are using AmigaOS 4 emulation according to Hans de Ruiter poll.

https://keasigmadelta.com/store/surveys/amigaos-4-x-hardware-emulation/

(88+43)/(88+43+202)=0.39 or 39%

That was with many complaints about slow AmigaOS 4 emulation but the many complaints about the high price of AmigaOS 4 hardware leaves them with the option of a virtual Amiga kludge or leaving the AmigaOS 4 market, the latter being the logical choice. Some Amiga fans are dedicated to AmigaOS 4 beyond logic while the Amiga masses have obviously left the market as the lack of hardware sales indicate.

kolla Quote:

What? x86 has nothing to do with this at all - the question was about 68k, PPC and Amiga - the largest PPC platform hands down is old macs and not Amiga, and for 68k I suspect that's also the case. If you look at what majority of developers on PPC use, it is certainly not Amiga. For 68k the numbers are probably more in favour of Amiga, but I suspect that a majority of 68k developers are still targeting old classic mac 68k hardware, and not Amiga. With a few exceptions, only Amiga developers care about 68k Amiga, and they aren't exactly high in numbers. Why you are dragging x86 SoCs into this I really don't know.

The x86 SoCs are currently available hardware which is being supported by developers and are not EOL. The 68k and PPC Amiga and Mac hardware is off the developer radar. GCC has already tried to kill 68k support a couple of times and PPC support has been reduced and is in danger of being terminated. Despite the older 68k architecture, there is still significant interest in the 68k where PPC died very quickly and there is very little active hardware and development. The NASA radiation hardened PPC chips may still be used and available but they are ~$100,000 USD chips that take a lot of expense and time to certify. What 68k or PPC SoC/CPU hardware is available that is not EOL (FPGA only cores are not a hardware target for compiler developers)?

kolla Quote:

Ah yes - the "gunnar argument" - let's make a superfast 68k FPGA system so we can run emulators, because that's so much better than emulating 68k!

As soon as a fast and large FPGA is needed, affordability and value decline but for slow and small logic like a low clocked 6502 CPU and small chipset, a FPGA becomes a very flexible and affordable tool.

kolla Quote:

And what "load-to-use stalls"?

Load-to-use stalls, or sometimes called load-use stalls, are stalls that occur in most RISC CPU designs after a load instruction before the data loaded is available in later instructions. A stall means the CPU pipeline is stopped until needed resources become available.

The following 68k instruction is a common instruction.

add.l mem,d0
; 1 instruction, 1 cycle, 1 register on the 68060

A JIT compiler will convert the 68k instruction to 2 RISC instructions.

load mem,r1
add r1,r0
; 2 instructions, 5 cycles, 2 registers on a Cortex-A53

The Cortex-A53 as used in the RPi 3, THEA500 Mini and the A600GS has a load-to-use latency of 3 cycles for L1 cache accesses so the add instruction must stall creating pipeline bubbles until the load data is available for the add instruction.

load mem,r1
bubble
bubble
bubble
bubble
bubble
bubble
add r1,r0
; 2 instructions, 5 cycles, 2 registers on a Cortex-A53

I add 2 bubbles per cycle because the Cortex-A53 has 2 execution pipelines with each bubble representing a missed chance to execute an instruction. OoO RISC CPUs can hide some of the latency and reduce the stalls but the Cortex-A53 is an in-order design. Compiler instruction scheduling can rearrange instructions to reduce load-to-use stalls but JIT compilers are unlikely to. A RPi 4 OoO Cortex-A72 core is roughly 10 times the size of a Cortex-A53 with increased power and cost. A 68060 core is a fraction of the size of a Cortex-A53 core.

You will probably say so what. Load-to-use stalls are just another gross emulation inefficiency like no Amiga DMA, 100% CPU load all the time, 3/4 of memory wasted, etc. The once elegant 68k Amiga becomes a pile of crap for people who know the 68k Amiga though. Actually, knowledgeable developers know how noncompetitive emulation is which is why emulation is not used for embedded use, compilers do not have targets for it or even FPGA cores and why it is EOL support.

kolla Quote:

We already have affordable FPGA systems that cover just about all other systems and people are suporhappy about them (MiSTer and clones).

FPGA simulation is good for old consoles that can not easily be upgraded. The 68k hardware like the 68k Amiga, Atari ST, Mac, X68000 and CD32 can be upgraded to a 68060 and an ASIC SoC which could be mass produced for $1 USD could upgrade the 68060@50-100MHz to a 68060@1-2GHz with larger caches for much more than a 10 to 20 times performance increase.

kolla Quote:

Why? Really - why?! Being such a strong believer, how come you barely have any 68k hardware already?

I have plenty of original 68k Amiga hardware.

4 Amiga 2000
2 Amiga 3000
1 Amiga 3000T
1 Amiga 4000
1 Amiga 4000T
1 CD32
1 Amiga 1000
1 Amiga 500

I have a dozen 68k Amigas with plenty of goodies. Many of them have aging issues and I am not motivated to fix them or upgrade them to more modern I/O. The CD32 is the only one I turned on in the last year and it has audio issues and stomping the ground will make CDs skip. My Amiga 3000T power supply blew up over a year ago. My old hardware is in decline just like the Amiga hardware market. I should sell off my old hardware before prices collapse and die with the users but I am not motivated to do anything with my Amigas including selling them. Even looking at them makes me sad.

kolla Quote:

Only thing that's embarrassing when it comes to Amiga is the number of nutcases, but hey... we love you too!

The nutcases are another reason I stopped using and developing for the Amiga. Yea, I was an Amiga nutcase for sticking around as long as I did but now I'm just a nutcase minus the Amiga part.

@OneTimer1
I guess the main point is to be able to use old hardware directly so those who prefer their expensive accelerators could still be able to use those. There's obviously as you pointed out nice solutions, I myself would like to have the flicker fixer already implemented on the board. The point being having the flexibility to articulate with old hardware, I am talking on A1200 grade hardware, but having the ability to use Zorro boards would also be sweet. You could have interesting new cases. And instead of having a pistorm on cpu extension port have the connections already ready on board would also be neat.

_________________
Indigo 3D Lounge, my second home.
The Illusion of Choice | Am*ga

@matthey

Quote:

FPGA simulation is good for old consoles that can not easily be upgraded. The 68k hardware like the 68k Amiga, Atari ST, Mac, X68000 and CD32 can be upgraded to a 68060 and an ASIC SoC which could be mass produced for $1 USD could upgrade the 68060@50-100MHz to a 68060@1-2GHz with larger caches for much more than a 10 to 20 times performance increase.

What software is needing that kind of power a pistorm doesn't provide it already? I would think that first one need to to introduce a solution in big numbers that allow software to be developed, if it was a 68060 2ghz even better, but a draconized raspberry pi 5 would also be interesting and perhaps faster for that goal alone.

_________________
Indigo 3D Lounge, my second home.
The Illusion of Choice | Am*ga

@pixie

I think Matthey still dreams of a cpmpetitive 68k platform on level of mainstream hardware. I more see 68k as retro. For that PiStorm or Apollo are powerful enough

@matthey

You are making a case for the presumed (in your case) inefficiencies of the emulation in Emu68 as a case to not want to use it. That's fine, if that's your preference, but what really matters to most people are, in no particular order:

1. How fast is it?
2. How compatible is it?
3. How affordable/available is it?
4. Bonuses

Judge these against existing hardware solutions.

1. Stamps all over the fastest overclocked 68060 Rev6. Faster than Vampire in many cases (maybe there are some memory bound tasks the vampire cold perform better at). Faster than many PPC, for that matter. Just look at real world benchmarks.

2. Better than average compatibility for most 040 class accelerators and always able to improve thanks to being software upgradable.

3. Much more reasonably priced than any 060 board, particularly when factoring through cost of a high end 060. And much more available too.

4. Additional hardware: RTG, networking, etc.

Now you can take your OCD gripes at to how it works internally or you can take the tangible benefits above. I'd take the win, personally.

Unless you can interest some wealthy backers with the means, any physical 68K ASIC resurrection is just a silly pipe dream.

Last edited by Karlos on 11-Nov-2024 at 10:25 AM.

_________________
Doing stupid things for fun...

@matthey

Quote:

I am not motivated to fix them or upgrade them to more modern I/O

Get some PiStorms already for the old 68000 systems, it doesn’t cost much.

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@kolla

I get the impression that for matthey, PiStorm is Kryptonite.

_________________
Doing stupid things for fun...

@kolla

Quote:

kolla wrote: Quote: I am not motivated to fix them or upgrade them to more modern I/O Get some PiStorms already for the old 68000 systems, it doesn’t cost much.

My inner selves is saying: 'Reject software emulation'
The fact say: 'emulation is faster, cheaper, available'

But there is something else about speed, availability, flicker fixers, hardware interfaces:
'We can plug an ARM into an A500/A1200 but we can use an ARM/AMD64 instead of an A500/A1200'

@OneTimer1

To be honest, because I had the 040/603e BlizzPPC combination, I often wondered what some form of bare metal 68K emulation could do for my A1200 back in a day. It was running DoomAttack (68K) at 640*400 under OS4 classic that opened my eyes to the idea that a general 68K accelerator need not use an actual 68K CPU. The PiStorm concept seems entirely logical to me.

_________________
Doing stupid things for fun...

pixie Quote:

What software is needing that kind of power a pistorm doesn't provide it already? I would think that first one need to to introduce a solution in big numbers that allow software to be developed, if it was a 68060 2ghz even better, but a draconized raspberry pi 5 would also be interesting and perhaps faster for that goal alone.

What Amiga needed more than a 68000+256kiB of memory until a 68060+128MiB Amiga? What Rpi needed more performance and memory until the RPi 2, 3, 4 and 5 came out? Why upgrade at all when there is no software that requires it yet? Shouldn't the original RPi have been enough for everyone?

The original RPi had a single scalar in-order ARM1176JZF-S@700MHz core and 256MiB of memory. The 68060 is superscalar so more advanced and has better integer performance/MHz.

1994 68060 1.8 DMIPS/MHz 8kiB-I/8kiB-D 500nm 8-stage superscalar 2-way
1999 ColdFireV4 1.54 DMIPS/MHz 16kiB-I/8kiB-D ~250nm 9-stage limited superscalar
2002 ColdFireV5 1.83 DMIPS/MHz 32kiB-I/32kiB-D 130nm 9-stage superscalar 2-way

2002 ARM11 1.25 DMIPS/MHz 16kiB-I/16kiB-D 90-40nm (RPi1=40nm) 8-stage scalar with parallelism
2005 Cortex-A8 2.0 DMIPS/MHz 32kiB-I/32kiB-D 65-45nm 13-stage superscalar 2-way
2011 Cortex-A7 1.9 DMIPS/MHz 32kiB-I/32kiB-D 40-28nm 8-stage superscalar 2-way
2012 Cortex-A53 2.3 DMIPS/MHz 32kiB-I/32kiB-D 40-10nm (RPi3=40nm) 8-stage superscalar 2-way

Year | CPU | transistors
1975 6502 3,500
1979 68000 68,000
1984 68020 190,000
1985 ARM1 25,000
1985 80386 275,000
1986 ARM2 30,000
1987 68030 273,000
1990 68040 1,170,000
1993 Pentium 3,100,000 superscalar in-order 2-way
1994 68060 2,530,000 superscalar in-order 2-way
1994 ARM7 250,000
1995 PentiumPro 5,500,000 OoO uop
2002 ARM11 7,500,000
2008 Nehalem 731,000,000 (1st gen Core i7 with 4 cores) 64 bit OoO uop
2011 Cortex-A7 10,000,000 superscalar in-order 2-way
2012 Cortex-A53 12,500,000 64-bit superscalar in-order 2-way
2012 Cortex-A57 75,000,000 64-bit OoO 3-way big.LITTLE companion of Cortex-A53

Rough ARM transistor counts come from the following link.
https://www.sciencedirect.com/topics/computer-science/stage-pipeline

The transistor budgets of in-order ARM cores was increasing linearly but OoO ARM cores are increasing exponentially. The RPi 4 Cortex-A72 is two generations past the Cortex-A57. I can no longer find the transistor counts for newer ARM cores but I will guess the trend below for the RPi 4 and RPi 5.

RPi 1 ARM11 core - ~3 times 68060 transistors
RPi 2 Cortex-A7 - ~4 times 68060 transistors times 4 cores is ~16 times 68060 transistors
RPi 3 Cortex-A53 - ~5 times 68060 transistors times 4 cores is ~20 times 68060 transistors
??? Cortex-A57 - ~30 times 68060 transistors times 4 cores is ~119 times 68060 transistors
RPi 4 Cortex-A72 - ~40? times 68060 transistors times 4 cores is ~160? times 68060 transistors
RPi 5 Cortex-A76 - ~60? times 68060 transistors times 4 cores is ~240? times 68060 transistors

Transistors are very cheap but they still add up with massive increases and active transistors increase the power. Remember when x86 CPUs doubled as space heaters and ARM CPUs were weak low power cores?

https://www.raspberrypi.com/news/introducing-raspberry-pi-5/ Quote:

27W USB-C Power Supply Raspberry Pi 5 consumes significantly less power, and runs significantly cooler, than Raspberry Pi 4 when running an identical workload. However, the much higher performance ceiling means that for the most intensive workloads, and in particular for pathological “power virus� workloads, peak power consumption increases to around 12W, versus 8W for Raspberry Pi 4.

JIT compilation is a high CPU load workload all the time to keep latency to a minimum. The RPi 4 SoC chip is using a 28nm process and the RPi 5 SoC chip is using a 16nm process yet the power consumption is higher for both than the Vortex86 SoC chips using a 40nm process.

https://www.vortex86.com/products Quote:

The Vortex86 family is x86 System-on-Chip which highly integrate both legacy and modern I/O, with the custom BIOS support, 2 to 6 watt low power consumption, -40 to 85℃ wide working temperature and 10 years long life cycle, it's an ideal and reliable CPU for industrial/automation applications.

The 64-bit superscalar in-order 4xCortex-A53 core RPi 3 SoC chip is about 6W max, 32-bit superscalar in-order 4xCortex-A7 core RPi 2 SoC chip is about 4.1W and scalar in-order RPi 1 1xARM11 core SoC is 1.5W all using a 40nm process. In-order CPU cores are lower power and cheaper than massive OoO cores. Passive cooling saves money and is more practical to embed too. Waiting for more die shrinks to drop power consumption may not be an option beyond where we are now as performance and power gains below a 28nm process are decreasing while costs increase. The RPi 5 SoC 16nm process is already an expensive process for the embedded market and the high power use limits embedded use. The RPi 4 is more practical but could in-order CPU cores using a fraction of the transistor budget for a lower cost and 50% to 75% of the power reach similar performance?

OlafS25 Quote:

I think Matthey still dreams of a competitive 68k platform on level of mainstream hardware. I more see 68k as retro. For that PiStorm or Apollo are powerful enough

The Amiga went extinct because a 68EC020&AGA@14MHz and 2MiB of memory was not enough to compete against high clocked 386 and 486 PCs with SVGA and 4-8MiB of memory. Maybe a 68000&OCS@7MHz and 256kiB of memory was enough and it should have just been cost reduced into a C64 replacement? Would that have been more competitive or just obsolete? Is the 68k Amiga obsolete right now because the hardware lacks value and is not competitive?

Karlos Quote:

You are making a case for the presumed (in your case) inefficiencies of the emulation in Emu68 as a case to not want to use it. That's fine, if that's your preference, but what really matters to most people are, in no particular order: 1. How fast is it? 2. How compatible is it? 3. How affordable/available is it? 4. Bonuses Judge these against existing hardware solutions.

Use value to judge against 30-40 year old hardware but ignore value to compare against possible new 68k Amiga hardware.

Karlos Quote:

1. Stamps all over the fastest overclocked 68060 Rev6. Faster than Vampire in many cases (maybe there are some memory bound tasks the vampire cold perform better at). Faster than many PPC, for that matter. Just look at real world benchmarks. 2. Better than average compatibility for most 040 class accelerators and always able to improve thanks to being software upgradable. 3. Much more reasonably priced than any 060 board, particularly when factoring through cost of a high end 060. And much more available too. 4. Additional hardware: RTG, networking, etc. Now you can take your OCD gripes at to how it works internally or you can take the tangible benefits above. I'd take the win, personally.

Low end semi-modern 68k Amiga hardware would crush the ARM virtual 68k Amiga more than ARM 68k virtual Amigas are crushing the 30-40 year old original Amiga hardware.

Karlos Quote:

Unless you can interest some wealthy backers with the means, any physical 68K ASIC resurrection is just a silly pipe dream.

Jay Miner could have said a 6502@1MHz and 64kiB of memory was enough for the Amiga. Instead, he created the 68000 Amiga which RJ Mical estimated to cost $49 million to take the Amiga from design to market if starting over which is something like $110 million today adjusted for inflation. The 68000 and Amiga chipset designs were difficult lacking computer tools back then (the 68000 created the workstation market that many chip design tools later used) so they were laid out by hand. Today, there are existing modular 68k designs considered obsolete like the 68060, many FPGA Amiga chipset cores already created and more modern modular I/O can be licensed like SiFive IP. It wouldn't cost anywhere close to $110 million to design and bring a new low end 68k Amiga to market. If licensing could be obtained, it would likely cost less than the original Amiga Corporation investment by 3 dentists, and perhaps an oil man, of $7 million. This is not a lot of money anymore. Trevor has likely thrown away half this amount on Amiga1 (I can't call it investing). Crowd funding could put a pretty good dent in the amount required but it would be better as a backup plan when ready for production. A few wealthy investors may be able to finance it. I know older people who have a million dollars in their checking account and middle age people who are millionaires but much of their money is in retirement so they may need a pink sheet business listing to allow easier online investing. The RPi startup was funded by 5 investors although the 68k Amiga is more difficult to bring to market but nothing like it originally was.

kolla Quote:

Get some PiStorms already for the old 68000 systems, it doesn’t cost much.

I have a CSMK3 and two rev 6 68060s. The drive access speed of the CSMK3 is awesome which makes up for any lack of CPU performance. I would buy a more modern 68060 accelerator before a PiStorm but the problem is my original hardware is failing.

4000T loses board configuration after heated up (hair line fracture on board?)
3000T power supply blew up
4000 cheap SIMM sockets are broken and can't use 3000T/4000T Mediator
CD32 audio problems and skips if stomping floor

I no longer have a working Amiga that can use my Elbox 3000T/4000T Mediator. Fixing the original hardware has a much higher cost than a PiStorm. We are talking about 30-40 year old hardware which are practically museum pieces at this point. Hardcore Amiga fans persevere long after the fair weather masses have vacated the Amiga market.

Last edited by matthey on 11-Nov-2024 at 09:54 PM.
Last edited by matthey on 11-Nov-2024 at 08:58 PM.

@Karlos

Quote:

It was running DoomAttack (68K) at 640*400 under OS4 classic that opened my eyes to the idea that a general 68K accelerator need not use an actual 68K CPU. The PiStorm concept seems entirely logical to me.

The PiStorm needs adapter hardware and at least a working Amiga.

If you Amiga died or if you don't want to change your collector item, you can use a RasPi without the adapter.

@matthey

One of the thing more interesting on Amiga OS is it's snappiness, and that isn't due only to CPU. If you get the benefit of hw accelerated gfx all works as intended, if there's an issue with it you could have a fast CPU and still lag behind. Imagine you have AROS on arm, do you think it would feel that much faster if it has hw accelerated graphics on both? AmigaOS on WinUAE already feels a magnitude faster than my windows, there's some point of diminishing returns...

The only thing I would like from AmigaOS was that it would allow better rendering engine, like anti-aliasing fonts, the use of alpha channel on windows, like composition engine on PS4, but all that could and should be done by gfx and not CPU.

_________________
Indigo 3D Lounge, my second home.
The Illusion of Choice | Am*ga

@matthey

Quote:

Low end semi-modern 68k Amiga hardware would crush the ARM virtual 68k Amiga more than ARM...

Oh please, this is just pure fantasy. It's fun to wallow in lala land for a while, but occasionally you need to get real. Emu68 on Pi exists. Your imaginary semi-modern 68K does not. It almost certainly never will exist either. How many years have you rambled on about it? In the meantime FPGA and emulation based solutions, not to mention a number of new 040 / 060 boards. Nobody with the interest has the capital and resources / facilities to produce an ASIC implementation or I'm sure it would've happened by now.

For those of us that want to enjoy a high end 68K on an original Amiga machine, before it, we or both die of old age, we can't wait for some eccentric billionaire to invent and mass produce your CPU as a gift to the community.

I'm sorry. Emu68 on PiStorm utterly trashed your 68K CPU because it was the only one that turned up for the fight (by virtue of existing) and thus won by default.

Last edited by Karlos on 11-Nov-2024 at 10:10 PM.
Last edited by Karlos on 11-Nov-2024 at 10:02 PM.

_________________
Doing stupid things for fun...

pixie Quote:

One of the thing more interesting on Amiga OS is it's snappiness, and that isn't due only to CPU. If you get the benefit of hw accelerated gfx all works as intended, if there's an issue with it you could have a fast CPU and still lag behind. Imagine you have AROS on arm, do you think it would feel that much faster if it has hw accelerated graphics on both? AmigaOS on WinUAE already feels a magnitude faster than my windows, there's some point of diminishing returns...

Commodore exploited the advantage of the 68k Amiga to offload the CPU by using the cheapest 68k embedded CPUs possible. It allowed very cheap hardware on the low end but the Amiga platform missed out on some CPU intensive software including software rendered 3D games that eventually killed the Amiga. The Amiga is responsive but more performance and memory opens up the rest of the world after the mid-1990s. Modern web browsers, office software, games, etc. all need a large user base to have several times the current 68k Amiga performance and memory. It's great that the 68k Amiga can scale so low which is an advantage that should be retained but small embedded systems often have several times the performance and memory of a 68k Amiga. Take the original RPi for example with ARM11 CPU and 256MiB of memory that Linux with a GUI has trouble with. The scalar ARM11 CPU is weaker than a superscalar 68060 but it is clocked at 700MHz so overall performance is several times higher and 256MiB of memory is more than most original Amiga hardware users have ever seen. Virtual Amigas are high performance and have plenty of memory in comparison to original Amiga hardware but they waste 3/4 of resources on the host system which is not good value compared to running native software on the same hardware. The 68k Amiga is lightweight which makes up some of the difference but there is a missed opportunity for the 68k Amiga to shine on similar spec low end hardware.

pixie Quote:

The only thing I would like from AmigaOS was that it would allow better rendering engine, like anti-aliasing fonts, the use of alpha channel on windows, like composition engine on PS4, but all that could and should be done by gfx and not CPU.

CPU performance may benefit anti-aliased fonts and certainly benefits scalable fonts. Alpha channel tricks are more in the realm of chipsets.

Karlos Quote:

Oh please, this is just pure fantasy. It's fun to wallow in lala land for a while, but occasionally you need to get real. Emu68 on Pi exists. Your imaginary semi-modern 68K does not. It almost certainly never will exist either. How many years have you rambled on about it? In the meantime FPGA and emulation based solutions, not to mention a number of new 040 / 060 boards. Nobody with the interest has the capital and resources / facilities to produce an ASIC implementation or I'm sure it would've happened by now.

Some Amiga users once thought recreating an Amiga in a FPGA was a hoax too. Now you believe in FPGA development at least. FPGAs are one step on the way to an ASIC for serious developers creating competitive hardware. I don't see the cost as the major hurdle bringing everything together. Amiga IP uncertainty is perhaps the biggest problem. Trevor should do what is in the best interest of the Amiga and let all Amiga IP ownership go back to Amiga Corporation under one ownership. He is more of a road block than an investor.

Karlos Quote:

For those of us that want to enjoy a high end 68K on an original Amiga machine, before it, we or both die of old age, we can't wait for some eccentric billionaire to invent and mass produce your CPU as a gift to the community.

Billionaires aren't required. Large businesses like Commodore aren't required anymore. Small businesses could pull off a low end ASIC if competitive hardware was a priority. With Amiga hardware becoming extinct, I would think competitive hardware would be a priority.

@matthey

Developing a ASIC might cost a couple of million Dollar. But then you just have the chip in you hand. You must still market it and bring it in products. You must prefinance a mass production to get prices down.

And you must persuade people to buy it or find companies thst license it. Today you have basically INTEL or AMD in PCs and Notebooks, ARM and cutsom chips based on ARM in devices that need something cheap and not needing too much POWER. Outside from that you certainly have POWER-chips in IBM-Servers and a bunch of different more exotic chips in embedded market. But even there I guess is ARM becoming dominant.

Where would your 68k dream ASIC fit in there? If you are really in contact with investors you certainly already made business plans. And there you must show where the special customer benefit is and how it differentiates from bigger competitors. The heritage from 80s and early 90s with Amiga and ST is certainly not enough.

You sound a little like repeating success of RPi by doing similar. But why do you think there is room for anither lowcost board based on a new 68k processor? Besides that you do not have one.

Last edited by OlafS25 on 12-Nov-2024 at 12:30 AM.

@matthey

Quote:

Some Amiga users once thought recreating an Amiga in a FPGA was a hoax too. Now you believe in FPGA development at least. FPGAs are one step on the way to an ASIC for serious developers creating competitive hardware.

FPGA is an accessible technology for individual users, so of course that was always a realistic proposal. But I don't think the Vampire / 68080 is going to become an ASIC solution in my lifetime. It's already fast enough for most users that invested and frankly just because it works well enough in current FPGA doesn't mean it will jist scale right up to GHz or even hundreds of MHz without some amount redesign. Right now, any deficiencies in the fundemantal design are occluded from view thanks to the fact the relatively low clocks hide latencies all over the place. That just doesn't cut it as you ramp the clock up.


Quote:

Billionaires aren't required. Large businesses like Commodore aren't required anymore. Small businesses could pull off a low end ASIC if competitive hardware was a priority. With Amiga hardware becoming extinct, I would think competitive hardware would be a priority.

Yes they are. Do you know why? Because literally nobody, and I mean nobody, other than excentric billionaires have the millions it would take to waste on a pure vanity project like this and think nothing whatsoever of the fact that they'll never make a penny on it as there isn't a market ready for it to exploit. It's just not a product anyone is interested in. In 2024, if your CPU isn't x64 or ARM or (at a push) RISC-V compatible, it's as good as useless.

The only people that want a fast 68K ASIC are here, in the retro community - and not even all of us at that. And, to be blunt, we're a dying breed to say the least.

Last edited by Karlos on 12-Nov-2024 at 01:02 AM.

_________________
Doing stupid things for fun...

OlafS25 Quote:

Developing a ASIC might cost a couple of million Dollar. But then you just have the chip in you hand. You must still market it and bring it in products. You must prefinance a mass production to get prices down.

Right. A few chip architects would need to be hired and their work could easily cost over a million USD. Ideally, much of the work would be licensed which could cost millions but could be used for multiple SoCs depending on the license, save work and saves time. The original 68k cores are considered obsolete so shouldn't cost too much to license if they can be licensed at all. New SoC I/O capabilities could be licensed from SiFive and their license looks very reasonable and without royalties. The 68060 design and SiFive SoC IP are both modular, written in Verilog and I believe use AMBA.

https://en.wikipedia.org/wiki/Advanced_Microcontroller_Bus_Architecture

The IP is tested and the 68060 was MC certified which is highly respected. There are multiple chipsets which could be used. SAGA was originally written in AHDL and converted to VHDL so may not be the best option even though Thomas may bypass Gunnar if there was a real attempt to make an ASIC. There are at least 2 other AGA capable Amiga chipset cores so there is no shortage of choices for Amiga chipsets. Worst case, the Amiga chipset could go into a low cost FPGA if it is not mature enough for an ASIC. The ASIC fab expenses and early production would likely be at least a million more. It may be possible to sell ASIC chips by themselves to Individual Computers and other accelerator producers. Low cost SBCs are fairly cheap to produce but mass producing them would add some cost. The SBCs could be sold directly to 68k fans at a fast pace at least at first to get some cash flow back in. There are several options including finding business partners to produce cases. Maybe partner with RGL for marketing and distribution too. I'm not in a position to do more investigation but my guesstimate would be $7 million USD should go further than it did for Amiga Corporation developing and producing the original Amiga.

OlafS25 Quote:

And you must persuade people to buy it or find companies that license it. Today you have basically INTEL or AMD in PCs and Notebooks, ARM and custom chips based on ARM in devices that need something cheap and not needing too much POWER. Outside from that you certainly have POWER-chips in IBM-Servers and a bunch of different more exotic chips in embedded market. But even there I guess is ARM becoming dominant.

ARM upscaled with AArch64 in search of performance but it has limited ability to scale down much like PPC. They still have Thumb(-2) for the small Cortex-M cores but you choose either small cores, small footprint, low power and low performance or large cores, large footprint and mid to high performance. There is a gap in the middle, ARM has been focusing on scaling up while paying less attention to the low end of the embedded market and the embedded market is huge anyway. It makes sense to share retro, hobby and embedded markets to improve economies of scale and the diversification is defensive. RPi has shown the way that Commodore didn't quite make it to.

OlafS25 Quote:

Where would your 68k dream ASIC fit in there? If you are really in contact with investors you certainly already made business plans. And there you must show where the special customer benefit is and how it differentiates from bigger competitors. The heritage from 80s and early 90s with Amiga and ST is certainly not enough.

I did some investigation of business opportunities as part of the Apollo team. I found potential business partners to make an ASIC possible but Gunnar was oblivious and ignored what I was trying to do. Gunnar ignored me when I talked about creating a business. I never tried to raise money but there was no business and I wasn't able to get far in my exploration. Worse, I lacked confidence in Gunnar myself and wouldn't invest in his toy myself. I won't try to sell something I don't believe in. I have done some investigation of licensing but stopped short of getting quotes as an individual.

I am part of a small middle aged investor group. We have 4 members who think favorably of the Amiga and several millionaires. That doesn't mean our investment group would have million dollar investors. The majority of our money is in retirement accounts which we use to invest online. The person I know for sure that has over a million in his checking account is not in our investment group. I know others who could and are worth many millions but most wealthy people don't have money sitting around as they are investing it to make more money.

OlafS25 Quote:

You sound a little like repeating success of RPi by doing similar. But why do you think there is room for another lowcost board based on a new 68k processor? Besides that you do not have one.

I look for cost advantages and watch for business opportunities which is what good investors do. I have explained some of them already, even repetitiously.

Karlos Quote:

FPGA is an accessible technology for individual users, so of course that was always a realistic proposal. But I don't think the Vampire / 68080 is going to become an ASIC solution in my lifetime. It's already fast enough for most users that invested and frankly just because it works well enough in current FPGA doesn't mean it will jist scale right up to GHz or even hundreds of MHz without some amount redesign. Right now, any deficiencies in the fundemantal design are occluded from view thanks to the fact the relatively low clocks hide latencies all over the place. That just doesn't cut it as you ramp the clock up.

I would be surprised if Gunnar pulled off an ASIC for his precious too. Yes, the AC68080 would need "some redesign" to clock it up. More levels of smaller caches would likely be required for starters. I don't think it would be too much of a problem to reach 1-2 GHz in an ASIC but the design has been optimized for a low clocked FPGA and could take a considerable amount of time to tune it for a much higher clock speed. I tried to get Gunnar to work toward preparing the AC68080 for an ASIC but he continued to optimize for FPGA. I would license the 68060, if possible, instead, even though it would need work too. Caches are usually relatively easy to change although cache coherency can be tricky. There are open source Verilog multi-level cache sources available that are configurable. It is worth it to hire an experienced chief architect though.

Karlos Quote:

Yes they are. Do you know why? Because literally nobody, and I mean nobody, other than excentric billionaires have the millions it would take to waste on a pure vanity project like this and think nothing whatsoever of the fact that they'll never make a penny on it as there isn't a market ready for it to exploit. It's just not a product anyone is interested in. In 2024, if your CPU isn't x64 or ARM or (at a push) RISC-V compatible, it's as good as useless. The only people that want a fast 68K ASIC are here, in the retro community - and not even all of us at that. And, to be blunt, we're a dying breed to say the least.

I believe there are people, even developers and investors, interested in seeing the 68k revived and modern implementations in silicon for historic reasons and for posterity sake.

https://www.rocelec.com/news/mc6800-motorola-processor-family?srsltid=AfmBOoosoQ1wfy8XEXwyF0ye5ZTS4b-TtlQoMlYNmdBUEEIk3-lnezQ0 Quote:

MC6800: Exploring The First Motorola Processor Family Rochester revives this legacy product family Who would have thought that in 2023 there could still be use for an 8-bit processor, almost 50 years after introduction? The MC6800 was the iconic first processor produced by Motorola, all the way back in 1974. The parts ecosystem surrounding the MC6800, such as the MC6850 (ACIA), MC6821 (PIA), and MC6840 (PTM), created one of the very first computing systems. A unique feature of the MC6800 in 1974 was that only one 5v supply was needed in an era where -5v, +5v, and +12v were far more common. To bring the MC6800 back to the market, after being unavailable in authorized channels for decades, was an undertaking for Rochester’s Design team. Unfortunately, the original GDS2 round tapes from Motorola were unreadable. However, there was the original hand-drawn schematic. A single piece of paper in Rochester’s extensive semiconductor archives was the key in returning the product to active status. Our Design team jumped to work analyzing the schematic and replicating the functionality of every single cell in the design. The original fab process at Motorola had charge pumps to elevate on-die voltage and overcome NMOS process limitations to get to that required single 5v supply, but that would not work in any 5v process today. Predominant storage elements had cross-coupled latches with gated pass-throughs that today’s silicon design tools cannot cleanly handle. Our Design team persevered through these challenges and triumphed with the first-silicon success with working silicon in our first customer system. Rochester Electronics has a Motorola archive that includes the original physical design databases for every processor product, from the MC6800 through the MC68060, as well as many other archives. Many of these products have already been brought back to life, like the MC6800, and are available today at Rochester. The list of archives or stocked parts at Rochester is monumental, and includes MC6800, MC6802, MC6809, MC6810, MC6821, MC6840, MC6850, MC68HC000, MC68020, MC68EC030, MC68030, MC68040, and MC68060. MC68EC030s will be stocked by the middle of 2024 and have currently shipped prototypes to our first customer. Even the unique 132-pin bumpered quad-plastic package for the MC68020 was brought back by Rochester and is currently being used for assembly in our Newburyport headquarters. All of this shows the dedication and capabilities of our manufacturing and design teams. Most of these products reconstructed by the Design team at Rochester Electronics through our product replication methodologies. In the 1980s, Motorola teamed up with Thomson as a second source for many Motorola processors in the European markets. Generally, Thomson kept selling some early 8-bit products far longer than Motorola. Thomson had the fab processes of Motorola, so the products were identical in customer systems. In fact, many of the listed products above from Rochester have been sold to customers who were previously using Thomson parts. What does this mean for our long-term system customers? Rochester has the design archive or active stock for every MC68xxx processor introduced by Motorola from 1974 to 1994. The MC68060 was available from NXP/Freescale/Motorola up until 2015 and was the last of the MC68xxx line. For our customers, this means we are and can support systems where Motorola processors were active from 1974 until 2015, with over 40 years of design-ins. That’s over decades of code compatibility and support with the MC68000 code stack. Every single MC68xxx processor introduced by Motorola can be replicated or is currently stocked at Rochester Electronics.

Some people respect the 68k legacy and others don't get it like you and Trevor I guess. It's funny that they talk about the 68060 but I couldn't find any in stock. They have 68LC060s and 68EC060s but where did all the full 68060s go at their high prices? Did you notice the 68060 was active for 21 years for the embedded market despite not clocking it up greatly reducing the value and not improving it, even the small 8kiB I+D cache? Why would a business let such a successful product become obsolete without improvements? Of course, why did Commodore let the Amiga chipset become obsolete and the Amiga with it? So nobody is interested in what a different history looked like even though such a design may actually be good for low end small footprint computers today?

@matthey

Quote:

Some people respect the 68k legacy and others don't get it like you and Trevor I guess

Wait, what? I don't respect/get the 68k legacy?

Wow.

_________________
Doing stupid things for fun...