Sholud something like this have been the Amiga's gaming future?

[Hattig]

As for the 8mhz cpu in early archimedes, its probably closer to 2x raw performance of the 68000 cpu in an a500, not to mention the fact that risc architectures often require more instructions for same end result as a cisc cpu.

The 10 MIPS 12.5MHz ARM60 in the 3D0 console was computationally around the same performance as a 25MHz+ 68030. Almost certainly the ARM2 in the Archimedes was at least 4x faster than the 68000 (4 MIPS vs 0.7 MIPS), if only because it was pipelined and had a 32-bit ALU. ARM of course has free predication and barrel shifting in its instruction set, so simple anti-RISC arguments aren't always applicable.

The GBA used an ARM7 chip at 16MHz. This has higher IPC (nearly 1DMIPS/MHz) as well as a higher clock speed. It's probably equivalent to a 50MHz+ 68030 - and only has to do 50% of the pixel rendering, so we'd need a 25MHz 68040 in an Amiga.

[Hattig]

Alternatively......maybe someone can tell us about the NatAmi 3D performance, there are some still shots but no videos of the screenshots shown on the website!

Gunnar just posted some N68050 FPU stats - apparently a 100MHz N68050 will perform like a 300 to 800 MHz 68040 in FPU performance depending on how optimised for the N68050 the FPU code is, or a 1.5GHz 486. But that's just based on instruction throughput and latency figures.

[Hattig]

@Hattig The ARM 2 was  very powerfull chip. Clocking in at 4 MIPS easy, VS the 68k 1MIP. But these are RISC MIPS, ARM2 had a much simpler instruction set (roughly one third the 68k), where you need to combine some instructions to equal one 68k CISC one. So some your throughput on MIPS are combined instructions. Rough comparison might x2 as powerfull. I would not say any more than that.

Yep, ARM2 appears to have around 35 instructions, so less than the 68000. Of course with predication you don't need some branches in the code, and you also avoid separate shift instructions as they're built in (and that also reduces the instruction set). As an aside, the 68000 code will be more compact because ARM uses 32-bits per instruction, and 68000 starts at 16. It took until Thumb2 for code density on ARM to match 68000.

We could use benchmark MIPS - like Dhrystone, although of course these have flaws too) ARM2 could get 0.33 DMIPS/MHz (so 2.66 DMIPS for an 8MHz ARM2). I'm having difficulty finding 68000 figures though, although I did see 0.2 DMIPS/MHz for the 68030 mentioned, which could let us assume ~0.1 DMIPS/MHz for the 68000.

[Hattig]

That's the only valid mips based post in the entire thread.
And, no, early (lower clocked) ARM processors did not outperform Motorola's later (higher clocked) 68K models.
Instead of raw benchmarks or mips figures you need to compare comparable software products running under an OS.

Nobody has claimed that early low clocked ARM CPUs would outperform later 68000 CPUs! ARM2 came out in 1986, around the same time as the 68030, so the fact that they perform similarly isn't shocking (although the ARM does it in a fraction of the transistor count).

There's 68000 MIPS and there's DMIPS which allows better comparison between architectures. I guess there's CoreMark too.

Not that it matters today.

[Hattig]

course it matters, most of us now own machines that are over 20 years old, but still love talking about this stuff

Anyway, i don't think the original archie ARM2 was a match for an 030, if thats what your getting at

Heh, nope, not if it took a 12.5MHz ARM60 to match a ~25MHz 68030 ...

Maybe the 8MHz ARM2 would have matched the 16MHz 68030 crippled with 16-bit bus in the Atari Falcon! But really it was more like a ~20Mhz 68000.