SHADES Quote:

I'd like someone to make a PCIe backplane mainboard, that has a CPU slot expansion again. It's one thing that allowed those, to remain useable for so long. Even DRAM could be on them like the zorram cards or a legacy chipset fpga card

Kronos Quote:

So basically a MiniITX motherboard with a really weird form factor which has 1 male instead of 1 female PCIe connector.....

That is a strange interpretation of what SHADES wrote. I understood it to mean something different that likely has existed before. Early, the C= Gemini board experiment allowed multiple CPUs and memory on the Zorro III bus.

Dave Haynie Quote:

One of the curious things I made was the Gemini board. This was a Zorro III board with two 68030 processors and FPUs, each with 4MB of DRAM, which could talk to the main system via DMA and interrupts. This was a bit of a clumsy way of doing it, but the idea was to work on a loosely coupled multiprocessing system. Randell Jessup, which had also worked on the DSP software, was working on this concept on the software side.

Multiple Gemini boards could work in Amigas with Zorro III and bug fixed Buster chips which C= never finished. The AmigaOS doesn't support SMP and memory is no longer a single flat model with the Gemini board but it should be possible to do distributive type parallel processing. The CPUs are more like being over a fast network on the same computer hardware rather than even AMP. Memory on Zorro III boards is used on the Amiga but it is not as fast as local memory on accelerators in CPU slots (closest memory to CPU) or the motherboard (no bus overhead).

The C= Amiga 5000 and 6000 were heading to more modular hardware and Dave Haynie worked on the PiOS hardware which was also very modular.

Dave Haynie Quote:

That is, in fact, what I kept doing for the PiOS One in 1996/1997, which together with Andy Finkel, Stephan Domeyer, and Geerd Ebeling, started PiOS AG with the intent of making personal computers. In this design, CPU and memory were on a card that connected via PCI to a CPU-less main board. I think an A5000 or A6000 would have been made something like this.

https://www.quora.com/Had-Commodore-not-gone-out-of-business-could-the-Amiga-range-of-computers-gone-on-to-rival-PCs

There are disadvantages of having modular hardware though. Connections through contacts are not as reliable (contact cleaner exists for a reason) or close (tighter integration is faster and potentially cheaper). PCIe may use the same connector on older boards but newer revisions every 2-7 years of PCIe have about double the throughput. Memory is usually found in dedicated memory slots so at least the memory controller can be on the CPU/SoC which is better for performance today but the CPU/SoC will not know how to use newer higher performance memory standards in the future. Of course the highest performance memory is integrated on the CPU/SoC chip itself (SRAM, eDRAM). Likewise, Integrating CPU cores together on the same chip has major performance and cost advantages. An integrated GPU on a SoC chip has major performance and cost advantages even though the best GPUs usually come on a modular card. With integration, a Raspberry Pi 4 SBC can be sold for $35 which is likely cheaper than a universal modular bare motherboard with a bunch of connectors and the RPi 4 offers a lot more value. I expect the majority of the expense of AmigaNOne hardware is to provide a relatively large motherboard with modularity, especially for a gfx card.

Last edited by matthey on 03-Sep-2023 at 06:14 PM.

@matthey

Quote:

matthey wrote: SHADES Quote: I'd like someone to make a PCIe backplane mainboard, that has a CPU slot expansion again. It's one thing that allowed those, to remain useable for so long. Even DRAM could be on them like the zorram cards or a legacy chipset fpga card Kronos Quote: So basically a MiniITX motherboard with a really weird form factor which has 1 male instead of 1 female PCIe connector..... That is a strange interpretation of what SHADES wrote. I understood it to mean something different that likely has existed before. Early, the C= Gemini board experiment allowed multiple CPUs and memory on the Zorro III bus.

Yeah yeah, this might have made sense for some alternative history multi CPU capable 68k AmigaOS, but this is a thread about an NG PPC "Amiga".

Something that is just 1 SoC with some minor support components and RAM that needs to be kept close to ensure reasonable latency.
So in the end it would be everything on the PCIe card and with noone (expect IBM and POWER) doing any real PPC chip development for the past 20 years there simply is no upgrade path.

-> solution in search for a problem


And even with 68k, PCI (let alone PCIe) barely makes sense for even an overclocked 060 and once you go past that you might as well get standalone Vampire which has no need for such expansions.

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

Kronos Quote:

Something that is just 1 SoC with some minor support components and RAM that needs to be kept close to ensure reasonable latency. So in the end it would be everything on the PCIe card and with noone (expect IBM and POWER) doing any real PPC chip development for the past 20 years there simply is no upgrade path.

It sounds like you want a PPC Amiga on a PCIe card to slot in a x86-64 motherboard hoping it will reduce the cost. There have been similar Amiga hardware attempts.

Inside Out was a 68040 or 68060 Amiga on a PCI card but never released.
https://bigbookofamigahardware.com/bboah/product.aspx?id=41

The Siamese EZ-PC Tower housed Amiga and x86 motherboards in the same case with interoperability between systems.
https://bigbookofamigahardware.com/bboah/product.aspx?id=1601

Sonnet made PPC accelerators on PCI cards for Macs which have been used on Amiga from PCI bus boards like the Mediator.
https://everymac.com/upgrade_cards/sonnettech/crescendo_g4_pci/crescendo_g4_7200_400.html

I'm not sure how much could be saved in cost as it is mostly leaving off the cheaper I/O part of the motherboard. Raspberry Pi has compute modules (SoC+memory) and separate I/O boards.

https://www.raspberrypi.com/products/compute-module-4-io-board/
https://www.amazon.com/Compute-Module-IO-Board-Compatible/dp/B08PVLH3PZ

If all the I/O is needed, it is cheaper to buy a RPi SBC than to buy a RPi CM and separate IO board. A small SBC where everything is integrated is the cheapest hardware option.

Kronos Quote:

And even with 68k, PCI (let alone PCIe) barely makes sense for even an overclocked 060 and once you go past that you might as well get standalone Vampire which has no need for such expansions.

The Vamp/AC FPGA doesn't have room for a 3D GPU. All but low performance CPU and GPU cores are too big to fit in an affordable FPGA Amiga. The Vamp/AC hardware could have added PCI support easily if there were enough free pins on the FPGA while PCIe would have required a moderately more expensive FPGA with SerDes support. With SerDes, SATA could have been supported too. HDMI and high speed USB may be more reliable also. SerDes lanes don't come free though. An integrated 3D GPU is the way to go where the Amiga chipset already has integrated 2D graphics but an ASIC is required for acceptable performance.

Last edited by matthey on 03-Sep-2023 at 08:23 PM.

@matthey

I certainly don't want a PPC Amiga on a PCIe card and I think I have made that perfectly clear in my 1st response to that side topic.


As vor Vampire + 3D card.

So you want an FPGA system emulate an Amiga on an HW level to then run badly ported PC games that would run much better on a potato?

Solution ...... searching..... problem......

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

@AmigaMac

Quote:

AmigaMac wrote: Will the A1222+ support MorphOS? How about a dual boot between MorphOS and AmigaOS4?

Highly unlikely.
I'd say the MorphOS team is more interested in expending its limited resources in porting AWAY from PPC.

_________________
All the way, with 68k

@SHADES

Quote:

SHADES wrote: @NutsAboutAmiga I'd like someone to make a PCIe backplane mainboard, that has a CPU slot expansion again.

I'm with you on this one. Sort of like a 21C S100 bus.

Through the focus on decreasing production costs, the industry as a whole has created a significant e-waste problem.

I know that it's capitalism driving renewed purchasing of widgets to replace barely outdated components. Shareholders demand an increase in valuation, dividends, or both.

Gotta have that 10% performance boost.
There have been many times when in order to get a new CPU, I had to also upgrade the motherboard, the RAM, and back in the day even the graphics card (PCI, AGP, PCIe).
Some of this ends up on the second-hand market, but a lot of it ends up in the landfill.

Over the last 20 years, how much has the digital audio of a PC improved? So many SKUs of so many motherboard OEMs included it by default. Keeps the audio chip makers in business, but basically I could've been fine with 1 to maybe 2 sound cards in all that time. Especially when I think about how I do most of my gaming with wireless headphones.

Same goes for Ethernet. It's been 1 GbE for about 15 years for most users. Most motherboards I've had since the mid 2000s have had GbE. Replace the MoBo, unnecessarily replace Ethernet, SATA 3, USB 2, USB 3.

Sure, it costs only an extra few dollars for the OEM to include these on the board, but as a card the consumer would be paying for the extra PCB, the system integration testing, support, box. But they'd be doing so less frequently, and there'd be less e-waste.

_________________
All the way, with 68k

@agami

Audio and ethernet is just some chicken feed these days while any connecror is a real component.

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

@agami
"Highly unlikely.
I'd say the MorphOS team is more interested in expending its limited resources in porting AWAY from PPC."

That would make sense and I would think that would be the case. The Mac PCIe (or the high clock (2.5/2.7 AGP ones for that matter) stuff is really about the best experience for AmigaNG for performance and cost. Amiga 4.x should have just ported over to that instead of creating and supporting hardware and chasing developers to make stuff work.

With Iris email now moving to Iris suite is yet one more step to make a comprehensive solution you can install in 5 minutes and major applications just auto update if you let them. It is just so slick. Iris saw the update and asked me if I wanted to update it just did it. The network improvements and other touches on the OS make the ISO shift a one stop move. Brilliant if you ask me.

Last edited by Matt3k on 05-Sep-2023 at 06:30 PM.
Last edited by Matt3k on 05-Sep-2023 at 06:29 PM.

@Matt3k

Quote:

I'd say the MorphOS team is more interested in expending its limited resources in porting AWAY from PPC."

Sensible IMHO! If they still have the resources to port away! Good luck to them!

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

@BigD

They already started some time ago. The difficult part is making it compatible with existing MorphOS PPC and Amiga 68k software in a transparent way. In the meantime they continue to support existing users with regular updates.

https://www.youtube.com/watch?v=oAUZivdbn-Q

@matthey

Quote:

Dave Haynie Quote: That is, in fact, what I kept doing for the PiOS One in 1996/1997, which together with Andy Finkel, Stephan Domeyer, and Geerd Ebeling, started PiOS AG with the intent of making personal computers. In this design, CPU and memory were on a card that connected via PCI to a CPU-less main board. I think an A5000 or A6000 would have been made something like this. https://www.quora.com/Had-Commodore-not-gone-out-of-business-could-the-Amiga-range-of-computers-gone-on-to-rival-PCs

Exactly. Yes.
PiOS-One. Obviously, on something like PCIe.
Good memory Matthey.
Yep, the standard constantly improves, a new backplain board every 2-7 years isn't a bad deal and PCIe is largely backward compatible. Certainly a lot less expensive than a populated mainboard that also, requires refresh.

Last edited by SHADES on 06-Sep-2023 at 02:58 AM.

_________________
It's not the question that's the problem, it's the problem that's the question.

@SHADES

Quote:

SHADES wrote: Yep, the standard constantly improves, a new backplain board every 2-7 years isn't a bad deal and PCIe is largely backward compatible. Certainly a lot less expensive than a populated mainboard that also, requires refresh.

So you either exchange it every 2 years with the motherboard (what was the point again?) or you wait 7 years and end up putting a PCIe5 compute unit into a PCIe3 backplane.

The average user is just gonna buy a new PC every 2-3 years cos the old one "feels slow" while the geeks selling and buying 2nd hand will have twice as many boxes to ship around.

All to not fix something that wasn't a problem to start with.....

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

@Kronos

Quote:

So you either exchange it every 2 years with the motherboard (what was the point again?) or you wait 7 years and end up putting a PCIe5 compute unit into a PCIe3 backplane.

Cost Kronos. Cost

A fully populated mainboard is expensive to design and use.

Being CPU agnostic, allows for options to swap out and use/upgrade CPU modules, without redesiging the mainboard like you currently have to do with Intel every CPU refresh.

How often did you "upgrade" your Amiga 4000?

PCIe is backwards compatible so, you don't "have" to upgrade for each PCIe refresh if you don't want to. CPU upgrade, no problem. Plug in a Pi? no problem. It's a PCIe BUS controller on a breakout board, pretty much.
The CPU intergration board slot, is the only thing needed to design, so that you can plan for upgrades.

Last edited by SHADES on 06-Sep-2023 at 09:47 PM.

_________________
It's not the question that's the problem, it's the problem that's the question.

@Kronos

If the board consists of just the PCIe slots and the electrical connections, the slots should default to whatever version of PCI is integrated into the CPU. You'd only need to consider changing the backplane if the physical PCIe connections changed.

Kronos Quote:

So you either exchange it every 2 years with the motherboard (what was the point again?) or you wait 7 years and end up putting a PCIe5 compute unit into a PCIe3 backplane. The average user is just gonna buy a new PC every 2-3 years cos the old one "feels slow" while the geeks selling and buying 2nd hand will have twice as many boxes to ship around.

The 2-7 years I used is because that is the shortest and longest period of time between PCIe standard updates.

Year | PCIe
2003 1.0
2007 2.0
2010 3.0
2017 4.0
2019 5.0
2022 6.0
2025 7.0 as planned

More recently, the PCIe updates have been every 2 years but the next planned one will likely be 3 years and this may slowly continue to slow down judging by chip fab process improvements slowing down, getting more expensive and current leakage reducing power reductions. Moore's Law is slowly ending but it is made less obvious by nations, usually for national security reasons, subsidizing chip fabs thus making newer chip fab processes cheaper than they normally would be and sooner. The PCIe lanes are actually high speed serial lines but the serializers and deserializer of SerDes lanes are high tech blocks on chips which have to handle the serial data quickly. Each new PCIe standard roughly doubles the data throughput so the serial speed has to double and the chip has to process roughly double the amount of data.

Although PCIe roughly doubles the data throughput with each new standard, the performance of newer PCIe cards usually doesn't double so the last generation of PCIe isn't necessarily outdated. Usually the most demanding card is for a gfx card. Much of the workload of a 3D GPU is done on the GPU after textures and scenes are transferred over PCIe to the memory in the card. The PCIe bus is mostly used for new textures and scenes or old textures and scenes that did not fit in the GPU memory. Using an older PCIe standard is not as bad for a 3D gfx card as one might think. Let's take a look at an example.

The Radeon RX-6400 is one of the cheapest modern new 3D gfx cards available. The performance is considered rubbish by gamer standards but it has the advantages of being small and low power (PCIe can provide the power and no additional power connectors needed). This allows the card to be used in older non-gaming PCs like Dell and HP yet many modern games are playable at 1080P even though reduced GPU settings are sometimes necessary. Potential customers want to know how the older PCIe standard performs so there are many comparison videos. The 1st is a nice side by side visual comparison while the 2nd shows game fps graphs but compares 3 different PCIe standards.

PCIE 4.0 vs 3.0 (RX 6400) - Test in 17 Games
https://www.youtube.com/watch?v=x5zJYhRorXk

RX 6400 VS GTX 1650 GAMING BENCHMARK+ DIFFERENT PCIE GENS RX 6400 PCIE 4.0 VS PCIE 3.0 VS PCIE 2.0
https://www.youtube.com/watch?v=sUDbjfRWAcQ

PCIe 3.0 isn't that much worse than PCIe 4.0 for this card but PCIe 2.0 would limit how many modern games could be played. We are talking about low end GPU gaming performance here but PPC AmigaNOne CPUs are low performance and limit gaming performance too. A decade old PC with an i5 is many times higher performance than PPC hardware and a used system with a new Radeon RX-6400 using PCIe 3.0 could probably be bought for less than $300 U.S. and still play most modern games. Well, people often upgrade when it is unnecessary thinking their decade old computer without a greater than 300W power supply could play modern games. The key is integration where the first Radeon RX-6400 video above shows a GPU 6nm chip process which was able to reduce the power enough for older computers. This shows the advantage of modularity where the customer is able to retain their investment in an older computer despite a decade old PCIe standard. As a counter argument, a customer could probably buy a Raspberry Pi 4, 5, 6, 7, 8, 9 and 10 for a similar price as a used PC and Radeon RX-6400. The RPi SBC route provides new hardware each generation with a better CPU and GPU. Granted, the RPi 4 spec is really low and there is a lack of games but there is obviously a market for the best value fanless CPU+GPU SBC. Better standard hardware with a standard OS and games like a console or the old gaming computers should be possible even though the RPi price is hard to beat. Value is as important if not more important than price as the more expensive Amiga outselling the Atari ST showed.

Rob Quote:

If the board consists of just the PCIe slots and the electrical connections, the slots should default to whatever version of PCI is integrated into the CPU. You'd only need to consider changing the backplane if the physical PCIe connections changed.

The chips on both ends of the PCIe connection would have to support the newer PCIe standard. Both the CPU and GPU of a fully modular system would need replacing to upgrade to the newer standard between them. It's possible the connecting lanes/lines need to be upgraded too. There are Cat5, Cat5e, Cat6 and Cat6a Ethernet cables for a reason and they use high speed serial lines/lanes controlled by SerDes capable chips for Gigabit Ethernet and faster.

Last edited by matthey on 06-Sep-2023 at 11:58 PM.
Last edited by matthey on 06-Sep-2023 at 11:56 PM.
Last edited by matthey on 06-Sep-2023 at 11:47 PM.

a) such system exist both for industrial uses an in small for m factor ( like some of the NUCs), hardly anybody buys that stuff cos it make no sense

b) try running a PCIe3 riser in PCIe5 and see how far you get

~> a solution that is the problem

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

@Kronos

Quote:

b) try running a PCIe3 riser in PCIe5 and see how far you get ~> a solution that is the problem

I don't see how it's a problem.
Do mainboards exist that have PCIe v3 or v4 Yes/No = Yes
Ok, great. Not a riser card.
Can you plug in PCIe cards and use them if the board exists? Yes/No = Yes
What problem.
Oh you want to use a riser card? Make it properly, good.
Oh you mean the CPU card, well a GPU is just another processing unit that runs on PCIe and is seems to do just fine. Again, what problem.

_________________
It's not the question that's the problem, it's the problem that's the question.

@SHADES

Quote:

SHADES wrote: @Kronos Quote: b) try running a PCIe3 riser in PCIe5 and see how far you get ~> a solution that is the problem I don't see how it's a problem. .

Tell me you never tried without telling me yo never tried……

Look this stuff exist, but outside of some niche cases nobody uses it or even cares.
Sometimes the unwashed masses just get it right.

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

@SHADES

It would be nice that A1200 had no CPU, and instead used the expansion board to set CPU/RAM combo, that way no CPU on motherboard would go to waste. Well we now have/will have A1200+.

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

@pixie

Quote:

It would be nice that A1200 had no CPU, and instead used the expansion board to set CPU/RAM combo, that way no CPU on motherboard would go to waste. Well we now have/will have A1200+.

Tethered to slow i/o and chipram limitations that are soldered to the board.
I'm looking to get away from that.

_________________
It's not the question that's the problem, it's the problem that's the question.