@Kronos

there were millions of amiga users only in germany. If only a small percentage of them interested you could sell much more than 10.000 units

Traditional amiga is a market but of course only a niche. You will not sell millions of devices

@BigD

Quote:

Well if it makes A-EON's custom hardware pointless when Trevor's invested thousands of his own money into this new hardware and software amalgam then I get where he's coming from. Also, if Hyperion's AmigaOS ends up on x86 (unlikely) then it likely to go the same way as BeOS IMHO.

That's interesting your example of an amalgam. It's almost like an anagram. Oh, speaking of, AmigaOne is an anagram of egomania.

But, I see no need for custom hardware, except for an OS4 laptop. The AmigaOne XE was a reference board made by MAI. Bad pick maybe. But these days there are companies like Raptor with the Blackbird. They are using real POWER CPUs with real power. When I checked it was cheaper than an X5000 and way more powerful. It kicks an X5000 in the nuts and kicks a PowerMac G5 in the dust!

Now it's only suitable for high end OS4 market. But, if you are going to charge high end prices for PowerPC hardware that isn't even desktop G5 class, why wouldn't you provide the best available hardware based on that CPU arch? I just found this! I am way late in catching up with Trevor's blog or my memory is failing. Trevor bought a pimped out Talos II in serial #1!

https://www.talospace.com/2018/12/who-got-talos-ii-order-1.html

But check this out if you haven't read it. This is a guy who bought a Talos II so he's running a POWER system and is criticising the X5000! When it's not the average x64 user but an actual POWER user, who has spent the money to specifically run Linux on a POWER9 system that is criticising an AmigaOne model, you know there is something wrong!

https://www.talospace.com/2020/01/another-amiga-you-dont-want.html

@cdimauro

Quote:

cdimauro wrote: @pixie Quote: pixie wrote: @agami We now have #emu68k for arm and we have AROS for arm also, wouldn't it be nice having AROS on arm using emu68k as a virtual processor for AROS 68k, it could be used like petunia/trance are used on AmigaOS4/MorphOS Not possible. AROS on ARM isn't running in big-endian mode, so the transparent execution of 68k applications doesn't work.

I was under the impression that arm could be mad to work in either mode

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

@Kronos

Quote:

Sticking to obscure overpriced special sauce HW that is lagging 15-20 years behind in performance (and won't see any improvement ever again) is a sure way to make user numbers dwindle.

Oh I don't know. The Vampire is exactly that. The Apollo core is said to be clocked at only 100 Mhz and the Vampire SA would be way more popular than an X5000! It's also more expensive than a cheap PC mobo and way less powerful. How on earth did they get any sales?

@Hypex

Quote:

Hypex wrote: The Vampire is exactly that.

... and thats exactly why it will fail to grow into a viable eco system in the long run.

Sure initial hype is big and so is demand, but everybody who might be a costumer either already has one or is in line.

Once that demand is satisfied sales will dwindle, new SW won't get released as frequent and in few years it will be at the same point as NG. Only with much weaker HW and a userbase even more disconnected from reality.

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

@Hypex

It's more for just a Blackbird plus CPU and no heatsink than it is for a complete X5000 with OS4 and driver licences.

@Hans

Quote:

I personally prefer the idea of ARM over x86 for a different reason. Going x86 when everyone else (Incl. Apple & Microsoft) are transitioning to ARM sounds like a great way to choose yet another soon to be on the decline CPU architecture.

I think this brings up an interesting point we need to consider. When PowerPC was chosen it was the new kid on the block and going places. Apple helped to cement it as a desktop CPU and embedded designs suitable for use in various devices like washing machines and TV were created. It even ended up in space! And had a stint in game consoles for one generation. With the 88000 in decline the PPC was a suitable 68K replacement. What could go wrong? Apple abandoned it just after the XE came out and then it disappeared from the desktops.

Could the same happen to ARM? Right now ARM is established and years have passed since it was the NKOTB. Well, reinvention really, since it started small as desktop RISC at the time of the Amiga and in particular the A3000.

What if OS4 is ported to ARM, then, just like PPC , ARM falls out of favour some time later? And x86/xxx takes it's place once again? It was a nice ride but, as Moore's unofficial law states, you simply cannot beat x86. Which has has embedded variants for years.

This doesn't seem likely since ARM is where PPC should be. And it looks like they gave up. Perhaps they could have tried harder with getting PPC RISC into markets that wasn't Apple, but they didn't and even the in the non-Apple embedded markets it died. It did drive a few cars around.

So, my point is, OS4 could be ported to ARM, then ten years later it too falls out of favour. Then we have the same people that said PPC was a mistake saying, oh porting to ARM was a mistake. Now Apple have gone back to Intel again. Why didn't they go x86 with OS4 like we said they should? They repeated the same mistake they did on PPC. They ported to a CPU that wasn't Intel which is just stupid. If they wanted to a CPU with three letters they should have gone AMD!

@OlafS25
You have to see how many of these millions used the Amiga for anything else than to play games. Linux, as good as it is, has problems penetrating the desktop market in a significantly and it is used way less than MacOS which requires expensive machines.

_________________

@Hypex

Quote:

So, my point is, OS4 could be ported to ARM, then ten years later it too falls out of favour. Then we have the same people that said PPC was a mistake saying, oh porting to ARM was a mistake. Now Apple have gone back to Intel again. Why didn't they go x86 with OS4 like we said they should?

Of course, if all your userland OS components were ported to some 68K CLR it could run hosted on any number of CPU architectures for which extremely mature 68K JIT already exists and on whatever high end real 68K still exists / FPGA. All you ever need to port for a new CPU are some bootstrapping kernel/HAL/JIT layer and keep everything else inside the JIT.

Virtualize your platform and run it on whatever you want.

_________________
Doing stupid things for fun...

@Karlos

Quote:

To be blunt, there's far more interest today in 68K than there is PowerPC. And while I'm sure that those running OS4 on gold plated niche hardware or MorphOS on the best of apples older HW have no regrets, it doesn't make it an attractive prospect except to a diminishingly small number of users. I would hazard a guess that neither platform has had any significant influx of new users in the last decade, just the same people updating whenever a new release comes out.

As a long time OS4 user I can tell you that there’s been an increase in active users over at amigans.net the past five years. I agree there’s more interest in 68k in general but all of us OS4 users are also 68k users so it’s obvious to me that the more interest in 68k Amiga users will also equal more interest in AmigaPPC users

_________________
____________________________
c64-2sids, A1000, A1200T-060@50(finally working!),A4000-CSMKIII
! My Master Miggies- Amiga 1000 & AmigaOne X1000 !
mancave-ramblings
X1000 I BELIEVE

cdimauro Quote:

A little bit off-topic, but about efficiency: Arm vs RISC-V? Which One Is The Most Efficient? Before someone is asking for RISC-V...

There is no CISC competition for mWh (power efficiency, get work done faster and sleep more) where x86(-64) usually wins but it has difficulty scaling this low and would be too high of power for most embedded uses if it did. The video didn't measure standby mW (probably practically zero for this group) or average mW across different workloads which are important for embedded use too. It would be nice to have CPUs that have better mWh than RISC CPUs but are lower power than OoO x86(-64) CPUs and scale lower like an in-order 68060 (not that the 68060 would scale to even Cortex-M4 levels which won in mWh but maybe to Cortex-M7 levels which has "almost double the power efficiency of the Cortex-M4" according to Wiki). RISC-V may look nearly competitive with ARM but it has to be clocked up more to get the same performance indicating weak performance/MHz even though the power/MHz is lower than ARM. This is not desirable for embedded use as higher clock speeds increase expense and noise.

Kronos Quote:

I believe I can fly.....

Could you ever fly? The 68060 was "one of the highest integer performance and easiest to use in-order CPUs on the market" in the 1990s. Despite Motorola/Freescale anti-marketing for desktop use, it was successful in the high end embedded market. While introduced in 1994, it only received "MC" status in 1999, still receiving die shrinks this late and wasn't discontinued until 2014 which is a long life for a CPU. This is incredible as there were many new ARM CPUs introduced that were low power and low cost like DEC's StrongARM SA-110 in 1996 that clocked at up to 200MHz, 4 times the clock rate of the 68060 when introduced. Better performance at lower clock rates is desired for embedded use though. The 68060 was not significantly upgraded so it eventually fell behind as newer processes reduce power use for embedded use. It appears that Motorola/Freescale had no intention of increasing clock ratings even when moving to newer processes as they didn't want the 68060 competing with OoO PPC CPUs like the 603(e) which it originally outperformed at the same clock speed which is embarrassing. Looking at 68060 parts discontinued in 2014, I see only MC68060RC50 and MC68060RC60 which likely indicates it was never clock rated higher than 60MHz which is the highest I have seen.

https://www.amigawiki.org/lib/exe/fetch.php?media=de:parts:freescale_pcn15939_eol_68xx0.pdf

The rev6 68060 often runs at 100MHz and some up to 133MHz. Too bad we couldn't play Quake at 640x480 resolution on a 68060@100MHz back in the day like in the following video.

100MHz 68060 - the BFG9060
https://youtu.be/cgE1xg1xCZ0?t=2181

Maybe you can't fly but the 68060@100MHz can fly. This is an in-order CPU at only 100MHz with only 8kiB I+D L1 caches, only 4 bytes/cycle instruction fetch and only a 32 bit data bus to memory. Now try to imagine how much a 1-2GHz 68060 would fly with at least double the L1 caches, a dual ported data cache, a L2 cache, an 8 byte/cycle instruction fetch, a 64 bit data bus and a hardware return/link stack just to get started with some basic performance upgrades.

Kronos Quote:

In reality bringing 68k to a level where it would be a viable alternative in any market would cost far more than anybody sensible would spend in fixing a problem that does not exist.

The basic performance upgrades I mentioned above would retain the 68060 core design. These are individually basic bolt-ons (HDL code from existing CPU designs could be reused) and the other changes are likely not major changes. All together it would not be a small job but experienced CPU architects should have no problem doing this work which is not particularly difficult. It would likely take some time to familiarize themselves with the deign and verify changes. There would be expensive development costs but the production cost would be cheap. Even with all these enhancements, the cost of an ASIC should only be a few dollars per chip when mass produced and the design could be reused and could remain viable for more than a decade like the original 68060.

Kronos Quote:

The whole "68k is booming" is that same delusion that we have seen at the start of "NG" and had been active even in the years before. Sure there is a market but it is limited and once you saturated that small vocal group sales will collapse. That is true wether you are offering a 500€ (+/-) system at the performance of a 20+ year old PC or if you offer a sub 50€ retro consoles.

When is the Raspberry Pi market going to be saturated?



Right now they can't keep them in stock. Not many dips in that sales chart. The embedded sales add defensiveness to the more cyclical PC and hobby like markets. It looks like a good market. They even told us their business strategy to stay competitive by aggressively pushing down prices. More than a few Amiga users would rather spend millions on emulation, OSs and other software that often can't be given away for free but the Raspberry Pi Foundation is focusing on hardware. Smart guys.

Last edited by matthey on 22-Oct-2022 at 06:07 PM.

@matthey

"The 68060 was "one of the highest integer performance and easiest to use in-order CPUs on the market" in the 1990s."

So? Are you trying to imply everybody stood still and just cranked up the clocks??
Even if that was true, still doesn't mean the 060 would even work at modern speeds or on modern nodes without a complete redesign.


As for the rPI, the chips used in these were commonly used and on more or less current nodes BEFORE the rPIs were developed, hence they already had economics of scale.

If there was a massive demand for modern 68k chips anything "Amiga" could piggy back on such chips would existed all along.

They don't hence noone to shoulder the gigantic upfront cost of leapfrogging a 25 year old CPU into the modern age.


All of it ignoring that any new 68k on that scale would not fly under the radar of whatever name the cooperate shell owing the IP has today.

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

@matthey

Quote:

The rev6 68060 often runs at 100MHz and some up to 133MHz. Too bad we couldn't play Quake at 640x480 resolution on a 68060@100MHz back in the day like in the following video.

It's GL quake running on MiniGL. All the really heavy work, the texturing, lighting, blending, depth testing and basic rasrerisation is happening in hardware. The CPU is doing all the executive stuff, sound mixing, PVS and geometry transformation.

_________________
Doing stupid things for fun...

Kronos Quote:

So? Are you trying to imply everybody stood still and just cranked up the clocks?? Even if that was true, still doesn't mean the 060 would even work at modern speeds or on modern nodes without a complete redesign.

It's not just the clock they cranked up. They increased the caches too. This is where most of the modern ARM in-order CPU performance comes from. There is no x86(-64) competition for in-order ARM and their reputation for embedded use allowed these designs to not get as much attention. The in-order designs are more focused on power reduction than performance too. ARM has recently been aggressively bringing out new high performance OoO 64 bit AArch64 designs to attack x86-64 desktop and server markets while they have been mostly ignoring their lower end embedded designs where fat AArch64 does not scale well. Their good code density Thumb and Thumb2 ISAs, licensed from SuperH which is derived from the 68k, are the secrets to their success in the embedded market. Nobody seems willing to challenge them in their bread and butter lower end embedded markets they are now ignoring.

Kronos Quote:

As for the rPI, the chips used in these were commonly used and on more or less current nodes BEFORE the rPIs were developed, hence they already had economics of scale. If there was a massive demand for modern 68k chips anything "Amiga" could piggy back on such chips would existed all along. They don't hence noone to shoulder the gigantic upfront cost of leapfrogging a 25 year old CPU into the modern age.

Sure, ARM has enough customers that their CPU cores are available on a variety of processes. They didn't used to be anywhere near as optimized as x86-64 cores for their process (mostly synthesizable cores that could be used for any process). Newer AArch64 cores are likely more optimized for the process and are much improved designs which is why they are competing better.

Kronos Quote:

All of it ignoring that any new 68k on that scale would not fly under the radar of whatever name the cooperate shell owing the IP has today.

Except that NXP/Freescale likely doesn't have any use for the 68k as they generate little to no profit from it and the designs are ancient. They prefer to pay license fees to ARM for a la cart ARM designs rather than create new CPU designs and ISAs. The 68k is off their radar and anything they can get from old designs is a bonus. It may be possible to license or acquire 68k core designs for retro use and a museum where it belongs due to the historical significance. This may present an opportunity.

Karlos Quote:

It's GL quake running on MiniGL. All the really heavy work, the texturing, lighting, blending, depth testing and basic rasterisation is happening in hardware. The CPU is doing all the executive stuff, sound mixing, PVS and geometry transformation.

It is still 1990s GPU hardware without T&L through a slow PCI to Zorro III gfx bus. This requires some performance for 640x480. I used 512x384 for GLQuake on my CSMKIII with 68060@75MHz and it wasn't quite as smooth as the video. The GLQuake FAQ says a "Pentium 90 or higher CPU" is required for GLQuake and we don't have as good of compilers for the 68060.

http://asmodius.fortunecity.ws/glquake_faq.htm

Last edited by matthey on 22-Oct-2022 at 08:17 PM.
Last edited by matthey on 22-Oct-2022 at 08:09 PM.

@Hypex

Quote:

Hypex wrote: @agami Now strangely, Apple have gone from RISC to CISC and are migrating back to RISC again. They already moved on from PPC but then they are now moving back to a PPC replacement. If they already solved the CPU problem by going x86/64 why would they repeat the same path they took earlier by going back to RISC than can never beat x64?

Because they can do it from the performance/drawn power PoV.
Quote:

But, the reason OS4 shouldn't run on a PC, is because it technically cannot as it is. OS4 is a big endian OS so it cannot naturally run natively as is.

I don't see why the endianess is a problem with any o.s., OS4 included.

Unless Petunia is mandatory to have it ported as well.
Quote:

It also has dependencies on PPC and some of these are lower level interrupt contexts that are PPC specific.

Same as above: I don't see why this could block a port to any other architecture. It should be possible to rewrite it.
Quote:

However, like I said in another thread, my work around would be a hybrid solution, with a custom compiler that only emitted big endian x86 codes.

Here is it: Efficiently Migrate Byte-Order-Sensitive Applications
The bi-endian Intel® C++ Compiler helps bypass the struggle, enabling application migration from big-endian computer systems to little-endian systems based on Intel® architecture.

BTW, this compiler was developed by the general site manager were I was employed.
Quote:

Aside from that, going x86 or x64 now means that people expect to run it on their home PC and any old PC lying around. Most likely they would still need a custom PC build. I wouldn't expect any hardware to work as they would need a mobo with support for all drivers on board.

Supported components would be needed, like with AROS.

@OlafS25

Quote:

OlafS25 wrote: @cdimauro no it is not. Amithlon has no chipset emulation either as far as I know. So what Michal wants to do is running AmigaOS or Aros 68k on RPi without needing any host system or integration in original amigas.

Actually Emu68 depends on the Amiga chipset, so that's why PiStorm is needed.

Rewriting it using only the host system would require a lot of work, because you have to write/rewrite a lot of stuff cheating the original ROMs to let the 68k code think that the chipset is still there (partially: enough for the scope).

@pixie

Quote:

pixie wrote: @cdimauro Quote: cdimauro wrote: Not possible. AROS on ARM isn't running in big-endian mode, so the transparent execution of 68k applications doesn't work. I was under the impression that arm could be mad to work in either mode

It was deprecated by ARM and it's not mandatory anymore to be implemented by ARM chip vendors.

So, you have ARM chips that support it and others which don't.

Specifically, it looks like that more modern the ARM chip don't support it. Which means the in future you should expect that only little-endian is supported.

@matthey

Quote:

matthey wrote: Except that NXP/Freescale likely doesn't have any use for the 68k as they generate little to no profit from it and the designs are ancient.

So what is it? Ancient designs or design that were so good that you just have C&P them onto a modern node to leapfrog 25 years?

Sure 68k generates no profit, but the very moment someone else makes it profitable it is a given that they want their share.

In reality there is no demand for fast/modern 68k today. What little there was got served with ColdFire for a while but ran out of costumers long ago.

So unless you find someone willing to invest $$$$$$$$$$ in bringing it up to speed (whatever is needed for the application at hand) AND clear up licensing the whole discussion of an non FPGA 68k CPU is moot.

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

@matthey

Quote:

matthey wrote: cdimauro Quote: A little bit off-topic, but about efficiency: Arm vs RISC-V? Which One Is The Most Efficient? Before someone is asking for RISC-V... There is no CISC competition for mWh (power efficiency, get work done faster and sleep more) where x86(-64) usually wins but it has difficulty scaling this low and would be too high of power for most embedded uses if it did. The video didn't measure standby mW (probably practically zero for this group) or average mW across different workloads which are important for embedded use too. It would be nice to have CPUs that have better mWh than RISC CPUs but are lower power than OoO x86(-64) CPUs and scale lower like an in-order 68060 (not that the 68060 would scale to even Cortex-M4 levels which won in mWh but maybe to Cortex-M7 levels which has "almost double the power efficiency of the Cortex-M4" according to Wiki). RISC-V may look nearly competitive with ARM but it has to be clocked up more to get the same performance indicating weak performance/MHz even though the power/MHz is lower than ARM. This is not desirable for embedded use as higher clock speeds increase expense and noise.

Exactly, and this is why I've posted this video: to show that weaker (simpler) architectures aren't efficient both from performance and (global) efficiency.

ARM looks better because it's more "CISCy": it can do much more "useful work" compared to RISC-V. That's why the latter needs to work at higher frequencies, so ruining the overall efficiency.

This means that a CISC has chances to do better, since it can do even more "useful work".

Which looks incredible, since we're talking about the embedded market.

This is because even embedded chips nowadays have at least L1 caches integrated, and often L2 caches as well. Which means that the "CISC tax" (the more complexity AKA more transistors required to handle the fatter decoder plus the backend that supports memory operations) is much less important or even became irrelevant.

Of course this is NOT true if we talk about SoCs with just the processor cores and no cache at all, and where the smallest die counts. Then RISCs win hands down on CISCs.

But those are very small market niches and I wouldn't care: CISCs haven't to compete on all markets.

@matthey

Quote:

Karlos Quote: Quote: It's GL quake running on MiniGL. All the really heavy work, the texturing, lighting, blending, depth testing and basic rasterisation is happening in hardware. The CPU is doing all the executive stuff, sound mixing, PVS and geometry transformation. It is still 1990s GPU hardware without T&L through a slow PCI to Zorro III gfx bus. This requires some performance for 640x480. I used 512x384 for GLQuake on my CSMKIII with 68060@75MHz and it wasn't quite as smooth as the video. The GLQuake FAQ says a "Pentium 90 or higher CPU" is required for GLQuake and we don't have as good of compilers for the 68060.

Unless your gfx card was very low on memory, resolution should've made no difference at those screen sizes. For example, glquake ran around the same speed on my BVPPC/BVision whether it was at 320*240, 400*300, 512*384 or 640*480. Even 800*600 was playable, but above that the memory and fillrate limitations really kicked in.

The amount of CPU bound geometry calculation needed doesn't change as a function of the resolution.

Last edited by Karlos on 22-Oct-2022 at 11:26 PM.

_________________
Doing stupid things for fun...

@matthey

Quote:

matthey wrote: When is the Raspberry Pi market going to be saturated? Right now they can't keep them in stock. Not many dips in that sales chart.

How many of those are RP240's? At this point the designation 'Raspberry Pi' is meaningless. The Pi's we want have become unobtanium now, right when they started to become useful. I hope the situation gets better....

Quote:

The embedded sales add defensiveness to the more cyclical PC and hobby like markets. It looks like a good market. They even told us their business strategy to stay competitive by aggressively pushing down prices. More than a few Amiga users would rather spend millions on emulation, OSs and other software that often can't be given away for free but the Raspberry Pi Foundation is focusing on hardware. Smart guys.

It is a good market, and they weren't the first. The Arduino was looked down upon by professional embedded developers, now they are being used everywhere. ESP16/32 was just a toy RF module until people discovered it had a powerful CPU inside. Hobbyist electronics is roaring back into fashion with cheap Chinese modules and open source projects doing everything from autonomous drones to 3d printers to IOT. It's almost like back in the 1970's when hobbyists were building their own hifi amps and TVs, and amateur radio was pushing the boundaries of rf design.

I don't know where the Amiga would fit in that market though. I am quite happy keeping it retro, which itself is a growing market these days. I don't care about ASIC CPUs etc. either, as I am quite happy playing around with old 68k chips. I would actually be happy if commercial Amiga interest finally died and removed the last shackles stopping us from doing whatever we want with our 30 year old tech that nobody else gives a damn about.