[tralarpa]

I wonder how much faster the ARM2 would have been compared to the 68k in a first-generation Amiga. The Amiga's chip memory only delivered 7 MBytes/s, shared between the CPU and the chipset! With its 32-bit instruction words, the ARM2 would have been very far from its theoretical performance.

[lproven]

> I wonder how much faster the ARM2 would have been compared to the 68k in a first-generation Amiga.

Considerably faster. I looked at both (and the ST) and bought an Archimedes.

ARM chips benchmarked from the ARM2 up to the RasPi 3B+:

https://stardot.org.uk/forums/viewtopic.php?t=20379

68000 benchmarks around that time:

http://www.faqs.org/faqs/motorola/68k-chips-faq/

ARM2:

Dhrystone/sec 5463

68000 @ 8MHz:

Dhrystones

68000 2100

MIPS

https://en.wikichip.org/wiki/acorn/microarchitectures/arm2

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

ARM2: from 6 to 10 million instructions per second, depending on instruction mix

68000: 1.4 MIPS typical.

(For comparison: Intel 8086 at the same speed, something like 300 Whetstones, 0.5 MIPS. So either of them stomped all over a comparable x86 machine from that time.)

So, very roughly, ARM2 was between 2-3x faster in typical use.

Note:

. Neither CPU could do FP in hardware.

. Neither had cache memory.

. The Amiga had a lot of complex hardware acceleration for graphics; the original ARM2 machines from Acorn (Archimedes A2305, A310, A400) had essentially none.

So, Amiga games could do things that on the Arc required raw CPU, typically done careful hand-coded assembler.

[tralarpa]

Yes, but that's my point: the ARM2 cannot get faster than 1.25 MIPS in an Amiga because of the memory bandwidth (assuming that the CPU uses 5 MBytes/s of the available 7 MBytes/s that it has to share with the graphics chipset).

[panick21]

That is defently the limitation that you need to improve over time. And eventually you likely split the graphcis memory off from the main memory.

To improve performance you need to start to add cache in or around your CPU/MMU and you need an increasingly clever memory chip to arbitrate memory access.

But you are right, you always going to run into limitations. As RAM is the most expensive part of the BOM, literally anything you can do to get maximum utilily of out of the memory bandwith is where you need to spend design time.

Like ARM did a 16 bit encoding for your 32 bit RISC would have been a great design feature to further reduce the how intensive it is to keep the CPU active. Of course nobody had thought of that combination, that was an early 90s thing.

RAM is just so expensive back then, you build your machine around sharing it as well as you can and living with the downsides.

[lproven]

I do not follow.

There was never any ARM-based Amiga from Commodore or any Commodore partner.

Any CPU performance is 100% theoretical because there was no such hardware.

The 68000 performance numbers I cited are from contemporary benchmarks and they _favour_ the 68K. The real chip in real Amigas ran slower.

The Acorn Archimedes used the ARM2; the ARM was developed for the Archimedes range. Its display, sound, memory controller, etc. are all pretty much unaccelerated.

Now there are Arm-based Amigas but they run the Amiberry emulator on top of Linux.

[rasz]

Whole idea behind ARM was running at full speed of ram. Amiga had very little Ram BW to spare for CPU so ARM2 would be throttled by narrow slow ram bus in the Amiga.

[lproven]

Well, yes, but that is only half the story, so on its own, it does not add up.

The whole idea of ARM2 (that specific version, not ARM1, not ARM3) was that clever use of the RAM bus meant it could run at full DRAM speed.

But the other side of the coin comparing with the Amiga is that the whole idea of the Amiga was that ~5 years earlier, it was possible to do amazing unprecedented general stuff for gaming purposes in sound and video chipsets that no other computer at the time could do...

Leaving the CPU almost irrelevant. Which is in part why The Next Amiga, the Hombre, simply discarded the 68000 altogether and switched to PA-RISC.

Which infuriated the fans even back then because they wanted their backwards compatibility.

The core concept of the Amiga and Hombre is "in 1982, what amazing non-dedicated sound and video can we do in hardware for a mass market price"?

By the time it was a product, this was contaminated with "it's also a general-purpose computer" and "look at us, we have multitasking" and "look, it's a 680x0 machine, but with multimedia!" and "look, it can emulate a Mac!" and "look, it can run Unix!"

Different selling points selling in different directions.

Given the raw computer power of the Archimedes, it didn't need any fancy chipsets. They became irrelevant. Sure, yes, the Amiga could do stereo digital sound and multiple bit plane colour 2D graphics and still have enough power left over from a 7-and-a-bit-MHz 68K to run game logic.

The Archimedes looked at that, and went "how about we just do all that in software?" And bundled a solid shaded 3D demo for free. Never mind the bouncing ball, never mind the Juggler, this you could play.

The Mac was "in 1984, how much Lisa can we keep for how little money?"

The ST was "in 1985, how cheap can we make something that looks and works like a Mac but can do colour and sound, so it can play games?"

The Archimedes was "in 1986, can we do a 32-bit RISC machine so cheap and simple we can compete in the same market with Commodore, Atari and the PC compatibles?"

[actionfromafar]

To this day my pet peeve is that Commodore didn't ship Amigas with a tiny amount of scratch RAM independent of the shared bus. Would have been so useful.

[lproven]

Is that not the distinction between Chip RAM and Fast RAM in real machines?

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

[actionfromafar]

A large amount of Fast Ram would have been even more useful, but if you are cost optimizing the BOM, just a tiny sliver of it (it being Fast RAM) would have been effectively a "manual cache" managed by the developer and would have been incredibly useful for anything computation intense, like flight simulators, spreadsheets, anything not involving just blitting graphics really.

[rasz]

And it wouldnt even be that hard, just two 2kx8 SRAMs (like the ones in every NES/Famicon) and CS line going from the chipset to map it somewhere high outside the range of the chipset was very logical, easily doable in 1987 and most importantly cheap. Those 6116 srams were below $2 retail in 1987 (Byte magazine retail prices), so $4 for big performance boost.

Hell, smart Commodore would design A500 with second trapdoor near the CPU. Ship unpopulated but offer official Commodore "turbo ram" expansion carts:

- $25 4KB version. $8 BOM, 2x $1.95 6116

- $40 8KB. $12 BOM, 2x $3.5 6264

- $100 64KB. $28 BOM, 2x $12 62256

- Extreme $400 256KB. $120 BOM, 8x $12 62256. Would stick out due to big PCB so make it an attractive piece of plastic with cool logo and "EXTREME" design.

Map it at D80000 (potential 256KB of space for activities) and let software vendors auto detect fitted option by running quick memory test. Easy speed boost and easy extra money for Commodore.

[lproven]

I think I see where you are coming from now.

You are not saying "there was no uncontended RAM in the Amiga", which is what I understood you to mean. You are saying "the Amiga did have uncontended RAM but it did not ship with any as stock."

Is that right?

If so, yes, true, but the key things here IMHO are:

1. The Amiga was designed and built as a games console, one which happened to be able to do other things. It was primarily marketed and sold as a games machine.

I wrote about an Amiga-compatible OS recently:

https://www.theregister.com/2025/05/22/aros_live/

And the comments are from slightly baffled readers asking if this is any good for running their old Amiga games, or how this makes gaming any easier.

(It is not and it does not, but that is what people perceive the Amiga as being for in the 21st century.)

2. In the real 1980s market, the Amiga's glory was to a large extent stolen by the Atari ST. The ST was a much more limited machine but it was vastly better than any 8-bit machine, and an entry-level ST was more usable than an entry-level Amiga.

ST games were pretty good for the mid-1980s when a lot of people still had ZX Spectrum or Commodore 64 level kit.

Many games companies targeted the lower-end machine and ported to the higher-end one.

The Amiga was competing with the cheaper, simpler ST in the market, and keeping the costs down became imperative. That's why Commodore didn't add more hardware or more RAM to the base-level Amiga. In fact for the Amiga 600 took the spec of the A500 and cut it down.

I am not arguing that some Fast RAM as stock wouldn't have been good. It would. But probably irrelevant to most gamers, and probably would have hurt the machine's sales.

[rasz]

Afaik only A3000 and A4000 shipped with Fast ram, both computers werent really price competitive with PCs.

[memsom]

Well - I used Archimedes computers with ARM2 and owned an Amiga 500+ and honestly, I couldn't tell you the Arcie was faster. It certainly didn't have the custom chips, so it is probably not a fair comparison.