Quote:

BrianHoskins wrote: My choice was 68k. A very close second choice, if I could have made it, would have been PPC. With respect to 68k, this is by far the most compatible CPU platform from which to enjoy our AMIGA computers. However, the lack of availability (especially for 060 but also even for 030) is a significant problem in our community. The ideal scenario here, from my perspective, would be an open source effort to implement 68k in a modern FPGA, whilst providing it in packages which match the existing pinout of the real CPUs so that they can be used as drop in replacements.

There are weak performance open source 68k cores available like the TG68. MikeJ has said he will open the FPGA Arcade 68k core which is a little more powerful. Most small production FPGA hardware designers want a small core like these as it keeps the price and size down (all board logic fits in one FPGA). If more performance is needed, an embedded hard CPU is added. Most embedded companies who would want a powerful FPGA core want the option to eventually create an ASIC and want a core which is prepared for an ASIC. I wanted to have a CPU designer who created ASICs take a look at the Apollo Core to give advice on preparing for it (we also talked about marketing a 68k ASIC as an off the shelf part) but Gunnar went the other direction in optimizing for performance in an FPGA and holding on tight. The original Natami 68k programmer, JensK, wanted to open the sources while retaining copyrights. The CEO I talked to about very high production embedded CPUs wanted open hardware. My preference is fairly open sources, cores and hardware but not free. Being more open encourages proliferation of products but some protection to make money may be appropriate.

It is more difficult to make 68040/68060 pin compatible CPUs. Many FPGAs are moving to lower voltage than even the 3.3V of the 68060. An ex-tractable package is not cheap and most CPUs would be soldered on today. It is not easy to simulate the exact behavior at a low level without detailed knowledge of their logic. We need to move forward. There would be new affordable accelerators available for original hardware if there were cheap 68k CPUs available.

Quote:

I am very surprised to see that an overwhelming majority voted x86. Are these genuine votes? It seems to me that there is very little possibility of enjoying the AMIGA platform on the x86 platform. It would mean porting over the OS, and I don't think that's ever likely to happen. Even if it did, this would be a choice that effectively discards the 'old' AMIGA experience and embraces a purely 'new' experience. This, for me, crosses x86 off the list because the old AMIGA computers are an important part of my AMIGA experience as a whole. I suppose these votes (if genuine) could be driven be a desire to see AROS lead the future direction of the AMIGA platform. I do respect AROS and it's an interesting project. It doesn't capture everything that I personally want from my AMIGA experience, though.

There were some irregularities in the poll votes. Many x86_64 votes came in very quickly. I think there are quite a few legitimate x86_64 votes though. Many Amiga users want more powerful hardware but I see expansion of the user base from an entry level product as higher priority for survival. AROS is already available for x86_64 and MorphOS also has a head start. AmigaOS 4 might be a little prettier but would be late to the game and likely arrive behind on features. I don't think AROS x86_64 has as many users as UAE because of lack of compatibility. It's also very difficult to support all the different hardware variations. Competition and lack of drivers are more worrisome than dealing with the ugly x86_64 ISA.

Quote:

davidf215 wrote: I have read some of their website. Unless things have changed, ARM is a different implementation of the PPC/RISC technology. Yes, Motorola did good with the 68k. It’s a good processor. However, looking forward, 64 bit is in while 32 bit is fading. A 64 bit chip and arch would be better for Amiga NG.

The largest CPU market in the world is embedded where the 32 bit CPU is the most popular by far. Not that polls are the most accurate, but the AspenCore gives the following poll data.

2017 embedded CPU main processor market share
8 bit - 12% (5 year trend is holding)
16 bit - 9% (5 year trend is declining)
32 bit - 63% (5 year trend is holding)
64 bit - 13% (5 year trend is increasing)

https://m.eet.com/media/1246048/2017-embedded-market-study.pdf

32 bit CPUs in embedded are healthy and dominating 64 bit CPUs by nearly 5 to 1 as 64 bit CPUs are often overkill and wasteful. The 64 bit embedded market nearly doubled from 2012 to 2017 though. The 8 bit CPU is holding strong because sometimes it is adequate and often most energy efficient and cheapest. The Adafruit’s Trinket board with Microchip ATtiny85 (8 bit AVR) MCU is only 1.2" x 0.6" x 0.2" (31mm x 15.5mm x 5mm). The following is a pic.

http://eecatalog.com/8bit/wp-content/blogs.dir/6/files/2017/07/Figure3-holdingtinyPCB.jpg

How many 64 bit CPUs have that small of foot print?

Quote:

Of course, that’s why I mentioned that ARM could be considered once the PPC has run its course (A-Eon X5000 and A1222 products no longer produced and stocks gone) for the X5000 and A1222 product lines. But the money has already been spent for PPC on X5000 and A1222. The market will decided once the A1222 arrives, and by my understanding there is a market base prepared to buy.

We can already see the sales potential of the x5000 at its current price and it is unimpressive. It's nice enough hardware but too expensive. Tabor will likely sell a little better but it is cheapened hardware that is not cheap enough. Total sales of both will likely be less than 2000 units. Maybe Hyperion would come close to break even after development costs and overhead but Tabor is likely to have higher development costs. Trevor would probably like to get his investment back on the Tabor despite the nightmare it has turned into. There is no game changer for the Amiga here.

Quote:

Are you suggesting 68k in a multi-core framework? Like an m3, m5, or m7 (similar to i3, i5, i7) but Moto instead of Intel? Aros has SMP.

The 68k could and should go multi-core. The reason the Apollo Core doesn't have multiple cores is limited space in an affordable FPGA. The logic is copy and paste. It did add multi-threading which uses the CPU for the blitter in SAGA. The Apollo Core is also 64 bit although it does little to take advantage of the most valuable feature which is 64 bit addressing (lacking a virtual MMU and having a small memory). An FPGA offers the ultimate in customization but bigger ones get expensive.

Quote:

I think Trevor has recently mentioned that there has already been a revision to the A1222/Tabor motherboard due to components not being available. It could be that a new core may also be included before the A1222 ships. An owner could easily sell an older Tabor motherboard to acquire a newer model (happens on eBay frequently for Amiga Classics, so Tabor would probably be no different). Others may simply add the 1st generation Tabor to their collection and buy a newer version later.

Trevor likely already bought the poor compatibility e500v2 core CPUs and it was likely a sizable investment. There may be other changes necessary to switch to the e500mc core as well. The 1st gen Tabors would lose a lot of value when the 2nd gen Tabors with e500mc cores were released.

Quote:

This is why I also think that ARM would be a better choice than 68k once the PPC in the X5000 and A1222 have run their course. An AmigaOS 4.x that would run natively on a RP3+ would gain potential users since it’s cheap, and cheap is practically king in American consumerism.

I realize how important cheap hardware is to the Amiga. It is possible for 68k hardware to be in the same ballpark for price and performance as Raspberry Pi 3 hardware. Raspberry Pi owners feel restricted by 1GiB of memory but that is a huge amount of memory for a 68k Amiga.

Quote:

I have read where some think open sourcing AmigaOS is a good pathway. However, I think open sourcing AmigaOS would lead nowhere. Aros is already open source and it hasn’t developed as quickly as AmigaOS 4. I’m not saying it’s a bad product. It is a good alternative. But because Aros is open source and it has developed and spread slower than AmigaOS 4, then it can be concluded that open sourcing AmigaOS would be no better.

It is more important to have the source available for embedded as the results of polling from the link above show. Also note that the Raspberry Pi form factor is 23% of the embedded market in the 2017 poll. See why I switch between embedded and entry level hardware?

@matthey

Quote:

There were some irregularities in the poll votes

Just repeating what I've said many times before...

Standard forum polls can be and -are- manipulated with ease.

When someone wants to gather information for whatever purpose, the policy is to inform an AW staff member of your desire to run a protected xoops poll.
Submit the poll with the request.
If approved it will appear:
https://amigaworld.net/modules/xoopspoll/

#6

_________________
This posting, in its entirety, represents solely the perspective of the author.
*Secrecy has served us so well*

@number6

Or one could use a free polling site.

Which CPU architecture are you most interested in for AmigaOS in the future?

I tried voting twice in this poll, and it would not allow me to do it.

Last edited by bison on 02-Feb-2019 at 05:44 PM.

_________________
"Unix is supposed to fix that." -- Jay Miner

@bison

Right.

The point of the protected xoops polls is that they are restricted to community members.
That would represent one point of difference, if deemed important by anyone trying to gather information.

#6

_________________
This posting, in its entirety, represents solely the perspective of the author.
*Secrecy has served us so well*

I voted for POWER (of course) because it is still in the realm of ppc and has built-in CAPI, which with a proper PCIe card may offer a super Amiga. At least it has possibilities.

However, I would go for x86_64, but only if it was the one under my table.

_________________
Tinkering with computers.

@number6

Quote:

The point of the protected xoops polls is that they are restricted to community members.

That's an advantage, but probably not a big one. One of the great things about the Amiga community is that it is wonderfully obscure.

_________________
"Unix is supposed to fix that." -- Jay Miner

@bison

Quote:

bison wrote: @number6 Quote: The point of the protected xoops polls is that they are restricted to community members. That's an advantage, but probably not a big one. One of the great things about the Amiga community is that it is wonderfully obscure.

It's an advantage that should absolutely without a shadow of a doubt be taken without any hesitation. It is imperative that mob rule be allowed to win out and at this very moment Hyperion and A-EON are taking notes as to how they should proceed based on the findings of these AmigaWorld polls.

If the x86-64 option wins then ALL of the multi-million dollar AmigaONE R&D fund will be redirected to x86-64 ISA support and a new PeeCeeONE machine delivered within a year of the poll finishing. The stakes are very high and if only the poll was secure the 'CORRECT' mob rule outcome would present itself.

Even the Apollo Team have agreed to throw all Apollo core code and Vampire plans in the bin and drink the x86-64 'kool-aid' if only the users speak out as one single voice of majority and dispel the foolishness of 68k and PPC forever!


"Keep drinking the x86-64 Kool-Aid, also available in ARM flavour!"

Last edited by BigD on 02-Feb-2019 at 10:25 PM.
Last edited by BigD on 02-Feb-2019 at 10:16 PM.

_________________
"Art challenges technology. Technology inspires the art."
John Lasseter, Co-Founder of Pixar Animation Studios

@Thread

A poll for a CPU platform is symptom for the typical error I have experienced on many Amiga boards.


You should not ask for a CPU, you should ask for available systems or at least for motherboards.

You might find new POWER systems but you will never find cheap ones, you can find cheap ARM systems but the CPU will never compete with a x86_64.

If you look for speed, availability or price of a system you will end with a x86_64 system.

And you should generate a new kind of AmigaOS for this hardware, something with memory protection, SMP and different user levels using UAE for backwards compatibility.

@OneTimer1

But it's also talks about future, we don't know the future, but its assumed that number of core be more important the specialized instructions, this is way ARM is on the horizon, it might be that ARM can challenge x86 systems, if becomes factor of number of cores per watt.

The future of AmigaOS is a bit more unclear, it already runs on many platforms on emulation like WinUAE / e-uae. And there is the clones like AROS and MorphOS, will AmigaOS forever be Power or PowerPC we don't know, is near future or 1000 years from now. Time is not specified, maybe find a way to make quantum computing possible in the home, or cpu's be designed more like neural networks, allowing the CPU intelligently use the resources, and CPU power more efficiently. Anticipate cache demands.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@NutsAboutAmiga

Predicting the future is always difficult, but with some help of Laplace series it iseems to be a rather safe choice to concentrate on x64. It was leading all the recent years and even if ARM will catch up or even overtake, x64 will not vanish or move to low end within the next decades.
That said ARM is also a viable future.
But Power rather not.

Or to make a stupid car analogy (Hooray!). Amiga runs on a stirling motor which is a nice concept but dead except from some niche solutions. Now, what's the future engine for Amiga? A fuel cell driven e-engine or a petrol engine?¹
I think fuel cell may have a great future (but fuel cell was hyped in books 20-30 yeasr ago already.!), but I am more than sure that petrol engines will stil be there in 20 years. Maybe optimized for a higher ethanol share and probably lesser spread, but they _will_ stay. Not that I like that fact . I drive cars as little as possible (about once a month), but am cycling or use trains for ecological reasons - which I encourage to do so as well. I'd be more than happy if petrol driven cars would have vanished in 20 years, but I don't see tat coming...
x64 is similar. Maybe other solutions are even better or hold higher potential, but it is just there and will stay.

--
¹ I left out battery driven e-motors, to make the analogy more clear. Probably batteries will dominate the e-car market for the next few years, but that's another story.

_________________
My programs: via.bckrs.de
MorphOS user since V0.4 (2001)

@Zylesea

That is an interesting comment as far as I'm concerned (not to take anything from other interesting, non-contrarian comments).

Why? because if I see the Amiga as some easy-to-apply => easy-to-use system which resides in many devices, the concept of what connects those devices beyond the communication basis is also still here. Just because I'm against the cliche notion of power (a practical way to move forward or simply to survive is another matter entirely), doesn't mean I think there's necessarily a clear and obvious path forward as far as drawing from a consolidated and original source of energy. If you know your tech and IT history, especially the one of Amiga as a unique machine/os which used the market to transcend the market (sales are not everything) and also thinking of Commodore as the streamlined backbone for the pathways, into peeps living spaces, the concept of a more shared and also independently developed sources of power isn't far fetched. Who thought that mere users will put the humble Amiga machines to such a good use? the designers had a certain idea and not beyond it. It might be the market and the everyman and woman way of doing what the space telescopes have been doing by using many voluntary machines around the world while looking at space. Maybe we're headed this way regardless and Amiga can put its own signature on it as a community project for everybody.

Quote:

NutsAboutAmiga wrote: But it's also talks about future, we don't know the future, but its assumed that number of core be more important the specialized instructions, this is way ARM is on the horizon, it might be that ARM can challenge x86 systems, if becomes factor of number of cores per watt.

With an understanding of CPU energy efficiency, multi-core scaling and Moore's Law, we would have an idea of where the technology should go in the future. CPU energy efficiency is *not* as simple as power/MHz times the number of cores. Watch the following seminar called "CPU Subsystem Total Power Consumption".

https://youtu.be/gaC1sO0q2OI
http://www.cast-inc.com/company/tech/CAST-ProcessorPower-Webinar-Dec2012.pdf

ARM likes to advertise low active power/MHz and low idle energy use. The low power/MHz is usually a sign of weak single core performance forcing higher clock rates.

Quote:

Higher performance allows doing more in less time, and sleeping for a longer time o Requires less energy to complete a given task Higher performance allows lower clock rates o Reduces clock tree and CPU power when active o Enables use of HVT cells and allows smaller implementation, both decreasing power leakage

Idle energy use is determined by the number of active transistors so a weak core is generally better here. Lacking single core performance, ARM turns up the core clock which is less energy efficient or adds more cores which is energy efficient but only accelerates parallel workloads. ARM energy efficiency was about sitting around idle most of the time and weak performance limiting the core to not use energy too fast. Nice sales pitch to look good while giving less.

AArch64 is a move toward better single core performance. This allows "doing more in less time, and sleeping for a longer time". Better single core performance was accomplished by borrowing CISC ideas like more powerful instructions and addressing modes. This architecture is far from minimal classic RISC and is approaching the complexity of hybrid CISC while using many resources (higher energy use and number of cores reduced). Even as AArch64 borrowed ideas from CISC to improve single core performance, it does not use variable length instructions which would have used code size efficiency to improve energy efficiency. ARM can probably ride their reputation as "energy efficient" if they can significantly improve the performance though.

The slowing of Moore's Law is beginning to level the play field so performance and energy efficiency are important again. This is likely to be first visible in low end CPUs where moving to smaller die sizes becomes prohibitively expensive. A more efficient architecture may be able to come in and undercut the price of less efficient architectures which moved to more expensive processes. An architecture with good single core performance has more options and a small footprint lowers overall costs (CPU and board) while improving efficiency. Of course economies of scale are necessary to achieve this.

Quote:

The future of AmigaOS is a bit more unclear, it already runs on many platforms on emulation like WinUAE / e-uae. And there is the clones like AROS and MorphOS, will AmigaOS forever be Power or PowerPC we don't know, is near future or 1000 years from now. Time is not specified, maybe find a way to make quantum computing possible in the home, or cpu's be designed more like neural networks, allowing the CPU intelligently use the resources, and CPU power more efficiently. Anticipate cache demands.

PPC is only likely to continue with customized hardware. POWER will likely continue for some time but the lack of single core performance requires more specialization for servers which is more likely to cause extinction as very specialized animal species are more likely to become extinct. Major breakthroughs in computer technology are unlikely anytime soon. Efficiency returns as it becomes more expensive to go smaller with dies sizes. There are many misconceptions about architecture efficiency but lots of opinions.

@matthey

"ARM can probably ride their reputation as "energy efficient" if they can significantly improve the performance though"

What are they energy efficient for? likewise with performances.

@matthey


1. PPC is only likely to continue with customized hardware.
2. POWER will likely continue for some time but the lack of single core performance........
3. Major breakthroughs in computer technology are unlikely anytime soon.
AND
4. There are many misconceptions about architecture efficiency but lots of opinions.

If this is another poll, I vote for number 4.


Of course x86 or ARM are the only way to go. All the experts here say so.

_________________
Tinkering with computers.

@matthey

(quoted from message 81)
Quote:

I wanted to have a CPU designer who created ASICs take a look at the Apollo Core to give advice on preparing for it (we also talked about marketing a 68k ASIC as an off the shelf part) but Gunnar went the other direction in optimizing for performance in an FPGA and holding on tight.

Gunnar did in fact make the correct decision. For all the ASIC discussions (which are fun to think about) the core is a long, long way from begin frozen in an ASIC.

I know you didn't get along with the Apollo team, but with this message and some others it appears you have some sort of grudge. After your parting of ways I haven't seen any arrows shot your way by the team. I suggest you return the favour going forward. I do like some of your postings so let's keep it civil.

Cheers!

just for evalutation ... check this arm machine specs

https://amperecomputing.com/wp-content/uploads/2019/01/eMAG8180_PB_v0.5_20180914.pdf

_________________
I love Amiga and new hope by AmigaNG
A 500 + ; CDTV; CD32;
PowerMac G5 Quad 8GB,SSD,SSHD,7800gtx,Radeon R5 230 2GB;
MacBook Pro Retina I7 2.3ghz;
#nomorea-eoninmyhome

@matthey

Thanks very much indeed for your comprehensive response. I have comments on some small parts, as follows, but I found your entire post to be of interest.

Quote:

My preference is fairly open sources, cores and hardware but not free. Being more open encourages proliferation of products but some protection to make money may be appropriate.

We are very well aligned on this point. My preference is also toward open platforms for exactly the reasons you described. But by open I also don't necessarily mean 'free' and would have no problem with the person (or people) who put in the work making money off it.

Quote:

It is more difficult to make 68040/68060 pin compatible CPUs. Many FPGAs are moving to lower voltage than even the 3.3V of the 68060. An ex-tractable package is not cheap and most CPUs would be soldered on today

I was thinking more in terms of creating a PCB which had the FPGA on it, but broken out such that it can fit as a drop-in for existing 68k CPUs. The electronics to account for IO level-shifting and power supply etc would then have to be on the PCB. Since 68k CPUs were physically large (especially 68040/060) then my intuition is that there would be plenty of room to fit such a solution. Mind you, whether it would be a good solution economically is another matter. Right now, if you want to buy a 68060 CPU, you will pay more than £100 for a second-hand unit of unknown origin on eBay. A very bad situation for enthusiasts of 68k hardware.

Quote:

It is not easy to simulate the exact behavior at a low level without detailed knowledge of their logic. We need to move forward. There would be new affordable accelerators available for original hardware if there were cheap 68k CPUs available.

On this point I will definitely defer to your better judgement. You could well be right about that. And you're definitely right that if such solutions were made available, then affordable accelerators would follow and we would simply liberate ourselves from the old hardware whilst maintaining compatibility with our existing AMIGA platform. Agreed.

I would go with Power. ARM could be a possibility but the performance is not there. x86-64 is not exciting at all and I have never even liked WinUAE. If we want to have something beyond retro we need to start somewhere and also be different/strange/crazy/cool enough to make it worth it.

Quote:

matthey wrote: ARM can probably ride their reputation as "energy efficient" if they can significantly improve the performance though.

Quote:

Srtest wrote: What are they energy efficient for? likewise with performances.

Let's compare CPUs.

68020
Introduced: 1984
Type: CISC scalar pipelined
Pipe stages: 3
Data Bus: 32 bit
Address Bus: 32 bit (4GiB)
GP registers: 16x32 bit
Transistors: 190,000
Caches: 256B instruction direct mapped
Process: 2 um
Voltage: 5
Max Power: 1.4W@33MHz

ARM2
Introduced: 1986
Type: RISC scalar pipelined
Pipe stages: 3
Data Bus: 32 bit
Address Bus: 26 bit (64MiB)
GP registers: 15x32 bit
Transistors: 27,000
Caches: none
Process: 2 um
Voltage: 5
Max Power: 1W@8MHz (1986), 2W@12MHz (1987)

68030
Introduced: 1987
Type: CISC scalar pipelined with MMU
Pipe stages: 3
Data Bus: 32 bit
Address Bus: 32 bit (4GiB)
GP registers: 16x32 bit
Transistors: 273,000
Caches: 256B instruction direct mapped, 256B data direct mapped
Process: 1.5 um
Voltage: 5
Max Power: 2.6W@50MHz

ARM3
Introduced: 1989
Type: RISC scalar pipelined
Pipe stages: 3
Data Bus: 32 bit
Address Bus: 26 bit (64MiB)
GP registers: 15x32 bit
Transistors: 309,656 (40,229 CPU logic with the rest cache related)
Caches: 4kiB 64 way set associative unified instruction & data
Process: 1.5 um
Voltage: 5
Max Power: 2W@25MHz

The Legendary "energy efficient" ARM doesn't look so energy efficient when comparing max power to the 680x0 on the same die size. The ARM2@8MHz performance was close to the 68020@33MHz performance although they are very different animals. At the same performance level, they are close in max energy use. The ARM architecture has better code density than most old RISC architectures like Alpha, PA-RISC, MIPS, SPARC etc. and is near that of PPC but the 68k is probably about 40% better code density which saves energy and cost on caches and board memory over ARM (caches use more energy than the control logic on modern CPUs and main memory more yet). ARM2 had an advantage on the core logic area which allows a lower CPU cost when excluding caches.

ARM2
smaller CPU core logic takes less area and is cheaper (adequate when the code size is small)
better performance at a lower clock allows a cheaper board
fast interrupts
simpler and cheaper to design
likely lower idle CPU energy use due to fewer active transistors

68020
code density improves performance and saves energy especially in caches and external memory
easier to program
good interrupt performance

ARM had good performance, the 68k had good energy efficiency and the CISC 68k was higher clocking than RISC ARM. Technology didn't stop here though. Both of these CPUs were using shallow pipelines and had multi-cycle instructions although ARM often used fewer cycles. Transistors became cheaper, caches became much larger and pipelines became longer allowing most of the instructions to became single cycle for ARM and the 68k (68060). The 68k code has fewer instructions and ARM now has to execute more instructions to keep up. ARM has to add larger caches which become slower as they get larger and use more energy. ARM introduced Thumb and then Thumb2 with a 16 bit variable length instruction set to improve code density which is nearly as good as the 68k but the number of instructions increased (and memory data accesses increased). Thumb2 was a huge embedded success helping performance and efficiency on low end hardware even though the 68k has better code density than Thumb2 and fewer instructions executed than the original ARM32. AArch64 was introduced to increase single core performance and it does likely reduce the number of instructions to a little below that of the 68k but ARM gave up Thumb2 like code density and increased complexity and resource requirements to where it is not as suitable for embedded markets. The 68k has not benefited from enhancements since Motorola gave the 32 bit embedded market to ARM yet it looks like it could still be competitive judging by the performance traits. I have done comparisons of the 68060 vs RISC CPUs of its time and the 68060 had better single core performance and often used less energy given a similar number of transistors with the same die size. RISC is good for minimal embedded CPUs that sleep most of the time though.