Your support is needed and is appreciated as Amigaworld.net is primarily dependent upon the support of its users.
|
|
|
|
Poster | Thread | Lou
| |
Re: One major reason why Motorola and 68k failed... Posted on 29-Jun-2024 18:41:52
| | [ #281 ] |
| |
|
Elite Member |
Joined: 2-Nov-2004 Posts: 4227
From: Rhode Island | | |
|
| I think you all keep confusing the purpose of this thread...
It's not about Commodore.
It's about Motorola.
68K lagged behind other architectures in many areas. PPC sucked.
ARM won.
Good day! |
| Status: Offline |
| | matthey
| |
Re: One major reason why Motorola and 68k failed... Posted on 29-Jun-2024 21:47:40
| | [ #282 ] |
| |
|
Elite Member |
Joined: 14-Mar-2007 Posts: 2270
From: Kansas | | |
|
| Hammer Quote:
For PS1 Geometry Transformation Engine (GTE)'s registers:
Matrix registers are 16-bit (1-bit sign, 3-bit integer, 12-bit fraction). Reading the last elements (RT33, L33, and LB3) returns the 16-bit value sign-expanded to 32-bit. This is not IEEE-754 FP16 (1-bit sign, 5-bit integer, 10-bit fraction).
Translation vector registers, each element is fixed point 32-bit (1-bit sign, 31-bit integer).
Background Color registers, each element is 32-bit (1-bit sign, 19-bit integer, 12-bit fraction). This is not IEEE-754 FP32 (1-bit sign, 8-bit integer, 23-bit fraction).
PS1's GTE's matrix feature is non-IEEE-754 FP16 math.
GTE's 16-bit matrix DSP is worse than DSP3210's IEEE-754 FP32 (GLFloat).
|
Fixed point integers with integer and fraction parts are not comparable to floating point formats with sign, exponent and fraction/mantissa parts. Floating point numbers require normalization after operations and use a wide barrel shifter (68k extended precision FPU requires 67 bit barrel shifter which was available even in the 68881). FPUs have deep pipelines as a result and are resource expensive. Fixed point has a performance advantage but floating point has a number range, quality and ease of use advantage that won out as silicon became cheaper.
Hammer Quote:
For cost reduction, design compromises have to be made while preserving fast performance.
Unlike Motorola, IBM has offered a game console low price PowerPC 602 CPU design.
|
Sure, but there are good compromises and bad ones. The PPC 602 is a PPC 603 with half of the 8kiB-I+8kiB-D caches that was on the small side giving PPC Macs a bad reputation and resulting in the PPC 603e with double the caches instead. RISC designers may not have realized it then but RISC-V research showed that fat RISC needed about 4 times the instruction cache to match the performance of a 68k instruction cache. A PPC 602 with 4kiB instruction cache would have similar performance to a 68k core with 1kiB instruction cache. The Apple Pippin console used the PPC 603 instead of the PPC 602 which was a better choice as the PPC 603 was more likely to be popular with economies of scale bringing the price down. A PPC CPU with 1/2 or 1/4 of the data cache compared to instruction cache would have offered more value for low end markets which is what some MIPS CPUs offered. The Motorola XCF5102 ColdFire 68040 hybrid had a 2kiB instruction cache (performance of 8kiB RISC instruction cache) and 1kiB data cache which was the right idea for the low end market but they dropped 68040 compatibility (also 3.3V and decoupled fetch and execution pipelines). Motorola's product planning was poor but they don't deserve all the blame for PPC mistakes as IBM was involved and made plenty of mistakes too.
https://en.wikipedia.org/wiki/PowerPC_600#PowerPC_602 Quote:
The PowerPC 602 was a stripped-down version of PowerPC 603, specially made for game consoles by Motorola and IBM, introduced in February 1995. It has smaller L1 caches (4 KB instruction and 4 KB data), a single-precision floating-point unit and a scaled back branch prediction unit. It was offered at speeds ranging from 50 to 80 MHz, and drew 1.2 W at 66 MHz. It consisted of 1 million transistors and it was 50 mm2 large manufactured in a 0.5 μm, CMOS process with four levels of interconnect.
|
Hammer Quote:
Reminder, MIPS R3000 is 1988 released hardware. R3000 attracted SGI.
80486 was released in 1989. 68040 was released in 1990.
------------
https://groups.google.com/g/comp.benchmarks/c/zvnlKLIgUDk/m/ut4VeY_t0lcJ SpecINT92 and SpecFP92 scores
SGI Indigo R3000 @ 33 32/32KB cache SpecINT: 22.4 SpecFP: 24.2 Benchmark date: Nov92
DEC 5025 R3000 @ 25 Mhz 64KB/64KB cache, SpecINT: 15.7 SpecFP: 21.7 Benchmark date: Jun93
HP 425t 68040 @ 25, 4KB/4KB cache SpecINT: 12.3 SpecFP: 10.3 Benchmark date: Jun93
Compaq Dkpro 80486DX @ 33 Mhz, 128KB L2+8KB L1 cache SpecINT: 18.2 SpecFP: 8.3 Benchmark date: Sep92
Compaq Dkpro, 80486DX2 @ 66 Mhz, 256KB L2+8KB L1 cache SpecINT: 32.2 SpecFP: 16.0 Benchmark date: Sep93
Compaq ProXL, Pentium @ 66.7 Mhz, 256KB L2+8/8KB L1 cache SpecINT: 65.1 SpecFP: 63.6 Benchmark date: Sep93
DEC 5260 R4400 @ 60 Mhz, 1M L2+16/16KB L1 cache SpecINT: 57.1 SpecFP: 54.5 Benchmark date: Sep93
DEC 2300 Alpha 21064 @ 150 Mhz, 512KB L2 + 8/8 KB L1 cache (there are faster Alpha CPUs in 1992-1993 year) SpecINT: 57.1 SpecFP: 54.5 Benchmark date: Oct93
HP 705 PA-1.1 @ 35 Mhz, PA-RISC SpecINT: 21.9 SpecFP: 33.0 Benchmark date: Nov92
HP 750 PA1.1 @ 66 Mhz, 256KB/256KB cache SpecINT: 48.1 SpecFP: 75.0 Benchmark date: Oct92
IBM 250 MPC601 @ 66 Mhz, PowerPC 601 SpecINT: 62.6 SpecFP: 72.2 Benchmark date: Sep93
|
You are comparing old CPU to new CPU designs, widely different caches (some on chip and some off), widely different chip processes and widely different priced workstations. The following is from a published source giving SPEC benchmark results.
68040@40MHz SPECint92: 35 SPECfp92: 23 DMIPS: 43.8 DMIPS/MHz: 1.10
68060@50MHz SPECint92: 49 (later "Greater than 50 Integer SPECmarks at 50 MHz" from Moto docs) SPECfp92: ? DMIPS: 90 DMIPS/MHz: 1.80
PPC601@66MHz SPECint92: 75 SPECfp92: 91 DMIPS: N/A
Pentium P54C@100MHz SPECint92: 100 SPECfp92: 81 DMIPS: 138 DMIPS/MHz: 1.38
Motorola Introduces Heir to 68000 Line https://websrv.cecs.uci.edu/~papers/mpr/MPR/ARTICLES/080502.pdf
The problem with the 68040 was that it was late and couldn't be clocked up as much as hoped because it ran hot. The FPU performance was low but not as bad as the 486. The 68060 shows huge integer performance potential but has limited compiler support and a minimalist FPU. The PPC 601 had good performance, especially the FPU, but this was with double the caches of the 68060 and Pentium P54C and the large 32kiB unified cache and shallow 4 stage pipeline limited the ability to be clocked up. The Pentium P54C has a much improved FPU over the 486 but the 5 stage pipeline is more inline with the previous generation of CISC CPUs and limited clock speeds. Sadly, the 68060 with an 8 stage pipeline and cool temps had the most potential to be clocked up but wasn't.
Hammer Quote:
Motorola 88000 is a joke. Check out 88000@ 33Mhz's Spec89 scores i.e. refer to my link. LOL. It's pathetically bad.
|
The Motorola 88000 was released in 1988 like the MIPS R3000. It was claimed to be the fastest 32-bit processor in the world when it was released but you claim the performance is "pathetically bad".
https://en.wikipedia.org/wiki/Motorola_88000#Release Quote:
Aimed at the high-end of the market, it was claimed to be the fastest 32-bit processor in the world when it was released. Running at 20 MHz, it reached 34,000 Dhrystones or 17 VUPS, compared to about 12 MIPS for a 12.5 MHz SPARC of the same vintage in the SPARCstation, or around 3.3 MIPS of the 20 MHz 68030. It was also available as a 25 MHz part at 21 MIPS, 48,387 Dhrystones.
|
The problem is you are comparing apples and oranges. You compare MIPS R3000 cores but not original R3000 cores using a 1200nm process but licensed R3000 cores using newer silicon to the 88000 on original silicon using a 1500nm process. The following is the original 1988 article where, "Motorola claims the family provides the highest performance available in 32-bit microprocessors."
https://techmonitor.ai/technology/motorolas_risc_set_to_go_in_april_is_tailored_for_unix
It wasn't unusual that performance leaders were short lived. As I recall, the PPC 604, HP PA-8000 and DEC Alpha 21164 all shortly claimed to be the highest performance MPU within a few years of your benchmarks above. Moore's law kicked in hard and it was several years before Intel consistently dominated with their economies of scale and CISC performance.
Hammer Quote:
Motorola deserves to go bankrupt or taken over by NXP!
|
As I recall, Motorola's chip division was spun off as Freescale before being bought by NXP. Freescale did almost go bankrupt during a financial crisis (2008-2009?).
Hammer Quote:
IBM PowerPC 601 @ 66Mhz smashed 68040 @ 25Mhz into the ground.
|
1990 68040 4kiB-I+4kiB-D caches, 1.2 million transistors, 800nm process 1993 PPC601 32kiB unified cache, 2.8 million transistors, 600nm process 1994 68060 8kiB-I+8kiB-D caches, 2.5 million transistors, 500nm process
You are really comparing a 1993 PPC601@66MHz to a 1990 68040@25MHz? Not even a 68040@40MHz? Why not compare to a 68060 that is only 1 year apart and has 66MHz benchmarks but strangely no 66MHz rated chips? Why clock up the more expensive to produce 4 stage CPU rather than cheaper to produce 8 stage CPU?
Hammer Quote:
The RISC threat is real, you're in dreamland. Motorola was largely missing in 1992-1993 in the "powerful" workstations. Most of these "powerful" workstations have external 64-bit bus.
Compaq has threatened Intel to improve its offerings or they move to the MIPS CPU family. https://en.wikipedia.org/wiki/Advanced_Computing_Environment Intel quickly released the Pentium P54 with 100 MHz in 1994 and the Pentium Pro with 200 MHz in 1995.
For Amiga's future, my argument position is a full EXIT from Motorola. I'm not referring to Apollo-Core's pretty good AC68080, just needs 68K MMU for Linux68K. Motorola has leadership problems.
|
My argument is that you don't know what you are talking about. Motorola had leadership problems but you are poor at design and benchmark evaluation and comparison. The competition is always a threat. RISC CPUs are simpler and easier to design while CISC CPUs are more powerful. RISC designs had the early advantage that they use less logic to pipeline when there was very limited transistor budgets. This advantage was more than offset by the cache savings of CISC code density a few years later and the CISC advantage continues to grow with cache size increases today.
Hammer Quote:
68060 has a major bottleneck with 4 bytes per cycle fetch from L1. Pentium doesn't have this bottleneck.
SysInfo has a significant amount of 6-byte instructions. To fully use 68060's dual integer pipelines, the programmer has to limit instruction usage to 2 bytes.
68060 needs to be updated to remove this limiter.
|
There is no "major" 68060 fetch bottleneck due to the tiny 4 byte/cycle fetch because of the decoupled fetch and execution pipelines with an instruction buffer. The P5 Pentium and Cyrix 6x86 do not have this so they need a large instruction fetch for high performance. Most high performance superscalar CPU designs today are like the 68060. The instruction buffer is likely made of SRAM and has a significant resource cost but it improves power efficiency.
It is sustained superscalar single cycle instructions of more than 2 bytes that reduce performance and this is uncommon. Non-superscalar and multi-cycle instructions allow the instruction fetch pipeline to catch up and fill the instruction buffer as do any stall cycles. The 68060 has a pair/triplet issue rate of 45-55% for existing 68k code and 50%-65% for 68060 optimized code which is better than most RISC cores but still just over half the time do 2 or more instructions execute together meaning the rest of the time the instruction buffer is likely filling. True dependencies between instructions limit a superscalar CPU so it can not reach 100% multi-issue rate. The poor 68060 results of SysInfo likely has more to do with poor instruction scheduling for a superscalar CPU than large instructions. If large instructions are a problem, it is likely do to the benchmark code being synthetic rather than realistic.
Please stop spamming dubious data and history revising conclusions.
Last edited by matthey on 29-Jun-2024 at 10:06 PM.
|
| Status: Offline |
| | Hammer
| |
Re: One major reason why Motorola and 68k failed... Posted on 30-Jun-2024 3:22:16
| | [ #283 ] |
| |
|
Elite Member |
Joined: 9-Mar-2003 Posts: 5859
From: Australia | | |
|
| @matthey
Quote:
@matthey Fixed point integers with integer and fraction parts are not comparable to floating point formats with sign, exponent and fraction/mantissa parts. Floating point numbers require normalization after operations and use a wide barrel shifter (68k extended precision FPU requires 67 bit barrel shifter which was available even in the 68881). FPUs have deep pipelines as a result and are resource expensive. Fixed point has a performance advantage but floating point has a number range, quality and ease of use advantage that won out as silicon became cheaper.
|
Read https://en.wikipedia.org/wiki/Single-precision_floating-point_format
The exponent refers to the integer component i.e. the full number component.
PS1's GPU vertex inputs are fixed point integers.
Quote:
@matthey You are comparing old CPU to new CPU designs, widely different caches (some on chip and some off), widely different chip processes and widely different priced workstations. The following is from a published source giving SPEC benchmark results.
68040@40MHz SPECint92: 35 SPECfp92: 23 DMIPS: 43.8 DMIPS/MHz: 1.10
|
http://kpolsson.com/micropro/proc1993.htm Quote:
May 1993, Motorola announces availability of 40 MHz 68040 processor. price is US$393 in 1000 unit quantities
|
https://en.wikichip.org/wiki/intel/80486/486dx2-66 486DX2-66 is released in March 1992.
Against 486DX2-66, Motorola's 68040-40 was missing in action for more than a year.
Apple attached DSP3210 @ 66 Mhz with their 68040 @ 40 for Quadra 840AV which is released in July 1993.
http://www.obsolyte.com/bogomips.html Quote:
From: SPEC list - 1997/06/11
Compaq Deskpro, 80486DX2 @ 66Mhz, 8KB L1 cache and 256 KB L2 SPECint92: 32.2 SPECfp92: 16.0 Info Date: Mar93
|
80486DX2 PC clones are cheaper in H2 1993.
Quote:
@matthey 68060@50MHz SPECint92: 49 (later "Greater than 50 Integer SPECmarks at 50 MHz" from Moto docs) SPECfp92: ? DMIPS: 90 DMIPS/MHz: 1.80
|
Counter, http://www.obsolyte.com/bogomips.html Quote:
From: SPEC list - 1997/06/11 Compaq DeskproXL, Pentium @ 66 Mhz, 8KB L1, 256 KB L2 SPECint92: 65.1 SPECfp92: 63.6 Info Date: Sep93
Intel Xpress, Pentium @ 75 Mhz, 8KB L1, 512 KB L2 SPECint92: 89.1 SPECfp92: 68.5 Info Date: Mar95
|
Quote:
@matthey
PPC601@66MHz SPECint92: 75 SPECfp92: 91 DMIPS: N/A
Pentium P54C@100MHz SPECint92: 100 SPECfp92: 81 DMIPS: 138 DMIPS/MHz: 1.38
|
Counter, http://www.obsolyte.com/bogomips.html Quote:
From: SPEC list - 1997/06/11
Intel Xpress, Pentium @ 90Mhz, 8KB L1 cache and 512 KB L2 SPECint92: 106.5 SPECfp92: 81.4 Info Date: Mar95
Intel Xpress, Pentium @ 100 Mhz, 8KB L1 cache and 512 KB L2 SPECint92: 118.1 SPECfp92: 89.9 Info Date: Mar95
|
Last edited by Hammer on 30-Jun-2024 at 03:59 AM. Last edited by Hammer on 30-Jun-2024 at 03:29 AM. Last edited by Hammer on 30-Jun-2024 at 03:26 AM.
_________________ Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68) Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68) Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB |
| Status: Offline |
| | Hammer
| |
Re: One major reason why Motorola and 68k failed... Posted on 30-Jun-2024 5:25:05
| | [ #284 ] |
| |
|
Elite Member |
Joined: 9-Mar-2003 Posts: 5859
From: Australia | | |
|
| @matthey
Quote:
The Motorola 88000 was released in 1988 like the MIPS R3000. It was claimed to be the fastest 32-bit processor in the world when it was released but you claim the performance is "pathetically bad". |
Who claimed "fastest 32-bit processor in the world"?
Quote:
The problem is you are comparing apples and oranges. You compare MIPS R3000 cores but not original R3000 cores using a 1200nm process but licensed R3000 cores using newer silicon to the 88000 on original silicon using a 1500nm process. The following is the original 1988 article where, "Motorola claims the family provides the highest performance available in 32-bit microprocessors."
|
Stop drinking Motorola's cool aid.
https://groups.google.com/g/comp.benchmarks/c/zvnlKLIgUDk/m/ut4VeY_t0lcJ SPECmark 89
Moto 8608's 88100 @ 20 Mhz with 16/16 cache has 7.8 score. Date: Oct89
DEC 3100's R2000 @16.67 Mhz with 128 cache has 10.1 score. Date: Oct89
HP 834's PA1.0 (PA-RISC) @15 Mhz with 128 cache has 9.5 score. Date: Oct89
Mips M/120-5's R2000 @ 16.67 with 64/64 cache has 11.2 score. Date: Oct89
HP 370 68030 @ 33, 256/256 cache has 3.0 Date: Oct89
Last edited by Hammer on 30-Jun-2024 at 05:32 AM.
_________________ Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68) Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68) Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB |
| Status: Offline |
| | bhabbott
| |
Re: One major reason why Motorola and 68k failed... Posted on 30-Jun-2024 14:59:40
| | [ #285 ] |
| |
|
Regular Member |
Joined: 6-Jun-2018 Posts: 422
From: Aotearoa | | |
|
| @matthey
Quote:
matthey wrote: @agami The broad measure for desktop value may be performance/$ but there are various ways of measuring it including int performance, fp performance, memory performance, drive performance, etc. | Nope.
Definition of value 1. uncountable noun: The value of something such as a quality, attitude, or method is its importance or usefulness. If you place a particular value on something, that is the importance or usefulness you think it has.
Quote:
Then there are features which add value that vary by customer and are difficult to quantify. As you mentioned, the value of a product is relative to similar products on the market. This is where the value problem for CBM came from. The value of Amiga hardware did not improve as fast as the value of PC hardware. |
Value = usefulness. What usefulness did a $399 PC have? None, because they couldn't be built for that price. But what value did a $399 Amiga have compared to a PC? Not much, because it couldn't run PC software.
As you say, 'value' varies depending on what features the customer values. So what does the customer value? CPU performance is only one factor. The most important metrics are generally 1. functionality - will it do what I need it to? 2. Value for money - How cheaply can it do it?
Quote:
Moore's Law kicked CBMs butt starting in about 1992. This was just as they improved value with a new generation of 68EC020@14MHz, AGA@14MHz, 2MiB chip mem systems but it was too little too late. They barely competed with higher clocked 386+VGA systems that were the low end PC. | Comparing CPU performance of an Amiga to a PC is generally the wrong metric. Apart from a few specific use cases, computing power is way down the list of things of importance.
In the case of the A1200, compatibility with the vast library of Amiga software titles (mostly games) was far more important than how fast the CPU could execute code. In fact having a much faster CPU could actually be counterproductive. Imagine the A1200 with a 68040 CPU - it's now over twice the price and software compatibility is in the toilet. The target market can't afford it, and they don't want it. Performance is better, but value is worse. Moore's law wasn't kicking CBM's butt, IBM compatibility (or the lack of) was. In 1992 the A1200 was kicking the equivalent low-end PC's butt in the area that mattered most to Amiga users at the time - 2D games. But people still bought the more expensive 386SX because it was IBM compatible, which is perfectly understandable considering PCs had 90% of the market and most computer users were not buying them to play games. In many countries you would have a hard time even finding an A1200 to purchase, let alone get any support for it.
How do I know this? Because I was selling A1200s literally alongside 386SX PCs and the first thing every customer asked was "Is it IBM compatible?" - unless they were an Amiga fan and then the PC was of no interest to them. PC purchasers generally had little interest in CPU performance, struggling to understand why a 386DX system cost so much more than a 386SX. Surely the later model number meant it was better?
In one case I recommended a 386DX system to a lady but she bought the SX because it was cheaper. A few years later she accused me of selling her an under-powered system, and I had to point out that it was her choice to go for the better 'value' machine. But I wasn't pushy in my sales pitch because there is a good argument that the 386SX was better value at the time, since it did everything she wanted for less money. That is what was argued in an article in the June 1994 issue of New Zealand PC World magazine. They compared buying a 386SX-16 in 1990, then a 386SX-20 in 1992, and finally a 486SX-25 in 1994, to buying a 386DX-33 in 1990 and expecting it to last 4 years. The 386DX cost $NZ7495. Going the other way cost $2495 + $2395 + $2399 for a total of $7289 - better value overall, and ending up with a machine that better suited the time.
In the PC world it often made better sense to buy a lower performance machine and upgrade when needed than pay for a 'bleeding edge' high spec machine. This is still true today. In the Amiga world it was even more true. In most use cases an A500 or A1200 was a better choice than an A3000 or A4000 even if they were the same price! CPU performance just wasn't that much of an issue for most users, as other attributes were valued more. Quote:
If they had 68EC030@28MHz, AA+@28MHz, 2MiB chip mem and 1MiB fast mem, at least they would be close to matching specs with low end PC hardware. This is why the CBM UK division wanted to add installed accelerators and maybe some games to improve value. |
Games added value because it gave new users something to run on the machine on Christmas day. A faster CPU and more RAM less so. In the UK they didn't even push hard drive equipped A1200s, since most users didn't need one. Retailers could easily have bundled A1200s with RAM boards (only £99 for a Hawk with 1MB), but they didn't because many customers would go for the base unit because it was cheaper and/or they were planning on installing an accelerator card of their choice at some later date anyway.
Quote:
Some Amiga fans would say THEA500 Mini eye candy case too but I consider it a gimmick. At the same time, I expect THEA500 Mini gimmick case will result in more sales than the more functional A600GS case though. Marketing can defy logic as customers can be fickle and impulsive. |
You clearly don't understand the reason people are buying the A500 mini. It's not to get the best performance, but for the retro experience. In this case the most important thing is that the (emulated) CPU have the same performance as an original A500. The case design is important because it makes it look more like an actual A500 and so triggers those nostalgic memories that are an important part of the retro experience. Personally I wish they had given it a functional keyboard too - but that might have priced it out of the market.
Last edited by bhabbott on 30-Jun-2024 at 03:11 PM. Last edited by bhabbott on 30-Jun-2024 at 03:06 PM. Last edited by bhabbott on 30-Jun-2024 at 03:05 PM.
|
| Status: Offline |
| | Kronos
| |
Re: One major reason why Motorola and 68k failed... Posted on 30-Jun-2024 15:53:09
| | [ #286 ] |
| |
|
Elite Member |
Joined: 8-Mar-2003 Posts: 2657
From: Unknown | | |
|
| @matthey
Quote:
matthey wrote: @agami Moore's Law kicked CBMs butt starting in about 1992. This was just as they improved value with a new generation of 68EC020@14MHz, AGA@14MHz, 2MiB chip mem systems but it was too little too late.
|
If C= (or Motorola) had really followed Moore's Law we would have seen a 020+RANGER based A2000 in 1987 or the A1200 (as an A500 replacement) by 1990 by which point would have had a Nyx/AAA based A3000 at the top end._________________ - We don't need good ideas, we haven't run out on bad ones yet - blame Canada |
| Status: Offline |
| | agami
| |
Re: One major reason why Motorola and 68k failed... Posted on 1-Jul-2024 4:36:41
| | [ #287 ] |
| |
|
Super Member |
Joined: 30-Jun-2008 Posts: 1779
From: Melbourne, Australia | | |
|
| @Lou
Quote:
Lou wrote: I think you all keep confusing the purpose of this thread... ... It's about Motorola. |
Yep, and Motorola's actions did have flow-on effects on platforms which used it as the driving engine.
In this thread, many have gone into the minutia of the choices and failings which ultimately doomed Motorola (the micro level). But integrators (OEMs) often plan adoption and integration based on larger movements and trends (the macro level).
Bill Gates' success and that of Microsoft is attributed to a bunch of different decisions, e.g. charging a license for the OS (DOS) in an industry that was focused on HW profits, edging out mainstays such as Wordperfect and Lotus 1-2-3 by making sure Word and Excel were more compatible with new versions of DOS/Windows: These are tactical measures.
Strategically though, the key contributor to Microsoft's success is in their core philosophy as it relates to Moore's Law: Why waste time (money) on optimising software, when in 12 months CPU performance will double, and will also cost less?
Motorola did indeed create an amazing CPU in the 68k. A moonshot of a product which was ahead of the competition. This lead to them resting on their laurels (complacency). And what is the material effect of that complacency? It was them not seeing the shift between the philosophies of the '80s computing industry and that of the philosophy of the '90s.
In the '80s, hardware (CPU/RAM) was so expensive that the best return on investment came from optimising software. In such an ecosystem, there's no point for Motorola to rapidly release newer and faster parts, because the culture of software optimisaiton for any specific platform was actually driving the computing needs.
Moore's law really only focused on the doubling of transistor counts. The economies of scale (reduction in cost of production of silicon) is what actually shifted the economics. In the '90s, anyone who was still thinking low volume with high profit margins, for a software optimisation culture, was doomed.
IBM's open computing platform choice for getting into the PC market cheaply and quickly, ended up working quite well in the new culture were software performance is improved quicker and at lower overall cost by upgrading components of a computer, i.e. hardware optimisation.
It wasn't just about standardisation, because Word didn't work with Wordperfect files, and vice versa. It was about being Agile with hardware. Release HW early, release HW often, Iterate. Goodbye software optimisation.
That's what Motorola and by association Apple, Atari, and Commodore failed to do. While we were bragging about small file and software binary sizes, the IBM-compatible (open computing platform) market was like "who cares, we just doubled the CPU performance, or RAM, or HDD size at half the price of last year's upgrades".
"My Mac or Amiga at 33MHz can do the same job in the same time that requires a 66MHz CPU on a PC". "OK, but my 66MHz PC is cheaper than the 33MHz Mac or the Amiga, and a hell of a lot more upgradeable."
Is it wasteful? Sure. Did it lead to decades of e-waste? Absolutely. But it solidified x86 and Microsoft as the dominant standard in personal/business computing, and it made (and continues to make), a whole bunch of people a lot of money.
The key trend in technology, since the invention of the wheel, has been and continues to be, commoditisation. Motorola (and others like it) worked against comoditisaiton, because it eats into profits, and more importantly it forces companies to move from high value providers to high volume providers. Not everyone is willing to make that shift, and not everyone knows how to make that shift. (I'm looking at you Yellow Pages).
_________________ All the way, with 68k |
| Status: Offline |
| | bhabbott
| |
Re: One major reason why Motorola and 68k failed... Posted on 1-Jul-2024 8:42:19
| | [ #288 ] |
| |
|
Regular Member |
Joined: 6-Jun-2018 Posts: 422
From: Aotearoa | | |
|
| @agami
Quote:
agami wrote:
Is it wasteful? Sure. Did it lead to decades of e-waste? Absolutely. But it solidified x86 and Microsoft as the dominant standard in personal/business computing, and it made (and continues to make), a whole bunch of people a lot of money. |
You say that like it's a good thing. It isn't. Our wasteful non-sustainable lifestyle is catching up with us.
Not only that, the business model is failing. Microsoft doesn't want you to buy a PC and use it forever without paying more, even though there has been no significant progress in a decade (I'm using a decade old PC and it does everything I want). But their attempt to continue monetizing Windows is failing. People are getting sick of it. Desktop PC sales are going down, and people are keeping their old computers for as long as possible because the new ones stink.
Quote:
Motorola (and others like it) worked against comoditisaiton, because it eats into profits, and more importantly it forces companies to move from high value providers to high volume providers. Not everyone is willing to make that shift, and not everyone knows how to make that shift. |
I don't see any evidence of that. What I see is a market dominated by x86 because IBM chose the 8088. I see Motorola doing their best to match what Intel was doing, despite having a much smaller market for their CPUs. I see Motorola having no market for their 68k desktop CPUs after 1994, apart from Apple whose sales dropped to a mere 3% between 1994 and 1997. What were they to do?Last edited by bhabbott on 01-Jul-2024 at 08:43 AM.
|
| Status: Offline |
| | OneTimer1
| |
Re: One major reason why Motorola and 68k failed... Posted on 1-Jul-2024 18:05:10
| | [ #289 ] |
| |
|
Super Member |
Joined: 3-Aug-2015 Posts: 1053
From: Unknown | | |
|
| @Kronos
Quote:
if C= (or Motorola) had really followed Moore's Law ...
|
I was curious and made some kind of extrapolation on Moore's Law. It's said: The number of transistors on a chip will double every 2 years. So I can predict what CPU a 68k computer should have and how many RAM. We also have some estimations for Clock frequency depending on what happened in the PC industry.
1987: A500 CPU 68000 RAM: 512kB 8 MHz
1989: Ax00 CPU 68020 (there is a big gap between 68000 and 68020) RAM: 1024kB 16 MHz could be a bit less
1991: Ax00 CPU 68030 (could have been an LC/EC type) RAM: 2048kB 28 MHz
1993: Ax00 CPU 68LC040 RAM: 4096kB 60 MHz (at least something between 40-60MHz)
I made this estimation for the low cost model, other features would have need an update too, but there was no fast and cheap 68040 or 75MHz 68060 from Motorola and after the main focus of Motorola switched to PPC it was discontinued as a desktop CPU.
|
| Status: Offline |
| | matthey
| |
Re: One major reason why Motorola and 68k failed... Posted on 1-Jul-2024 22:09:22
| | [ #290 ] |
| |
|
Elite Member |
Joined: 14-Mar-2007 Posts: 2270
From: Kansas | | |
|
| bhabbott Quote:
Nope.
Definition of value 1. uncountable noun: The value of something such as a quality, attitude, or method is its importance or usefulness. If you place a particular value on something, that is the importance or usefulness you think it has.
|
Performance is a measure of usefulness for a computer. The more performance is available, the more useful it can be. The needed features and software have to be available to make use of the performance though.
bhabbott Quote:
You clearly don't understand the reason people are buying the A500 mini. It's not to get the best performance, but for the retro experience. In this case the most important thing is that the (emulated) CPU have the same performance as an original A500. The case design is important because it makes it look more like an actual A500 and so triggers those nostalgic memories that are an important part of the retro experience. Personally I wish they had given it a functional keyboard too - but that might have priced it out of the market.
|
Compatibility is as important as performance for retro systems. However, more performance is more useful if compatibility can be maintained. Customers would love to be able to play 3D Amiga games at 100 fps on THEA500 Mini but it isn't happening even though better value hardware is possible that could without increasing the hardware price. The RPi Foundation understands the importance of value for the low end market and is seeing great success because of it while A-Eon hardware has horrible value and Amiga1 is a failure because of it. The path to Amiga success is aggressively improving performance/$ like RPi hardware while retaining compatibility and providing more modern features. Amiga has the games that the RPi does not, at least without emulation which it is successfully using due to the lack of competition (e.g. Retro Pi).
@OneTimer1 We would have liked to have seen 68k Amiga progression more like you posted. I tried making my own chart of 68k CPUs starting with the 68000 and improving at the pace of the Moore's Law prediction of doubling the number of transistors every 2 years but maybe I did something wrong?
Year | Predicted Transistors | CPU | Actual Transistors 1979 68,000 68000 68,000 1980 102,000 1981 136,000 1982 204,000 68010 84,000 1983 272,000 1984 408,000 68020 190,000 1985 544,000 1986 816,000 1987 1,088,000 68030 273,000 1988 1,632,000 1989 2,176,000 1990 3,264,000 68040 1,170,000 1991 4,352,000 1992 6,528,000 1993 8,704,000 1994 13,056,000 68060 2,530,000 1995 17,408,000 ... 2024 427,819,008,000 Apple M4 28,000,000,000
The results are suspect. Maybe the 68k didn't keep up with Moore's Law and maybe Moore's Law has slowed down enough to make Apple's M4 SoC off by that much?
Last edited by matthey on 01-Jul-2024 at 10:53 PM. Last edited by matthey on 01-Jul-2024 at 10:35 PM. Last edited by matthey on 01-Jul-2024 at 10:10 PM.
|
| Status: Offline |
| | OneTimer1
| |
Re: One major reason why Motorola and 68k failed... Posted on 1-Jul-2024 23:32:28
| | [ #291 ] |
| |
|
Super Member |
Joined: 3-Aug-2015 Posts: 1053
From: Unknown | | |
|
| @matthey
Quote:
I tried making my own chart of 68k CPUs starting with the 68000 and improving at the pace of the Moore's Law prediction of doubling the number of transistors every 2 years but maybe I did something wrong? |
You made nothing wrong, you started when the 68k was released 1979. I started when the Amiga with 68k was released, my idea was not to show the development of a super CPU, I focused on the usage in a low cost computer like the A500.
My Idea was when the A500 started with a CPU built of 68000 transistors its successor two years later should had a CPU with 136000 transistors, this would have been less than a 68020 but more than a 68010 and the next step would have been a 68030 EC/LC and so on.
|
| Status: Offline |
| | agami
| |
Re: One major reason why Motorola and 68k failed... Posted on 2-Jul-2024 4:27:01
| | [ #292 ] |
| |
|
Super Member |
Joined: 30-Jun-2008 Posts: 1779
From: Melbourne, Australia | | |
|
| @bhabbott
Quote:
bhabbott wrote: @agami
You say that like it's a good thing. It isn't. Our wasteful non-sustainable lifestyle is catching up with us. |
I'm saying that as a factual thing. Personally, I hate how wasteful things have been, but my personal views on the matter cannot change what had transpired.
Those who rode high on the commoditisaiton wave, I'm sure they saw it as a good thing. And were someone to point out the e-waste problem to them, they'd state that it's someone else's problem, downstream.
Quote:
I don't see any evidence of that. What I see is a market dominated by x86 because IBM chose the 8088. I see Motorola doing their best to match what Intel was doing, despite having a much smaller market for their CPUs. I see Motorola having no market for their 68k desktop CPUs after 1994, apart from Apple whose sales dropped to a mere 3% between 1994 and 1997. What were they to do? |
Nobody has a market for anything in perpetuity without innovating. The fact remains that Motorola were milking the 68k IP for all it was worth during the '80s, that they didn't spend enough on innovation. When the '90s rolled around, they were found sleeping at the proverbial wheel. If they innovated into a commodity business, frequently changing and updating the microarchitecture and core designs, and waking up to the fact the less and less people will be developing in Assembly, then they wouldn't have needed to be a founding member of the AIM "Alliance". x86 would still have been a dominant architecture, but the innovation and evolution of the 68k at that time could've given the market an alternative like what ARM is today, and what PowerPC never was.
If they thought more like a commodity business, they would've licensed the design to a second source manufacturer. This in turn could've helped them if they wanted to focus more on embedded and portable applications while the second source might be more optimised for desktop and workstation/server applications, or vice versa.
Last edited by agami on 02-Jul-2024 at 04:43 AM.
_________________ All the way, with 68k |
| Status: Offline |
| | Hammer
| |
Re: One major reason why Motorola and 68k failed... Posted on 2-Jul-2024 5:31:49
| | [ #293 ] |
| |
|
Elite Member |
Joined: 9-Mar-2003 Posts: 5859
From: Australia | | |
|
| @bhabbott
Quote:
Value = usefulness. What usefulness did a $399 PC have? None, because they couldn't be built for that price. But what value did a $399 Amiga have compared to a PC? Not much, because it couldn't run PC software.
As you say, 'value' varies depending on what features the customer values. So what does the customer value? CPU performance is only one factor. The most important metrics are generally 1. functionality - will it do what I need it to? 2. Value for money - How cheaply can it do it?
|
1. Before business software's release on a platform, prove the mass-produced Amiga model (e.g. A500) has a stable high resolution for business.
For the Mac 512K, Apple established the necessary hardware infrastructure for GUI MS Excel (1985), MS Word (1985), Aldus PageMaker (1985), Adobe Illustrator (1986) and QuarkXPress (1987) from the get-go. Apple licensed PostScript from Adobe. Normal people have day jobs. Apple attracted business customers in sufficient numbers for the Mac platform.
Meanwhile, Commodore leadership and engineers were arguing about Hi-Res Denise and post-A500 evolution. Hi-Res Denise evolved from monochrome to four colors and was mostly completed by late 1987 and Color Hi-Res Denise's release was delayed to 1990.
Amiga's WordPerfect 4.x port from the PC's DOS version is just a weak "zero-sum" i.e. it's not "next-gen" GUI MS Word on Mac.
ProWrite 1.11 (1987)'s on Amiga OCS's interlace resolution is not good and has no spell checker.
MS Word version 3.0 (1987) for Macintosh adds a built-in spell checker, outliner, style sheets, page preview, hyphenation, and sorting.
ProWrite 2.0 (1988) has a spell checker.
Both Atari ST (e.g. SM125 monochrome) and Amiga (i.e. "Hedley Hi-Res" grayscale hack) need special high-resolution monitors while it's unified on PC VGA.
Microsoft/Aldus recycled their Mac GUI experience for Mac ports of Excel (1987-1988), Word (1989), and PageMaker (1988) on Windows 2.x. Standalone Windows 1.x/2.x runtime version is free for developers. Mac GUI experience is important for Microsoft since its ideology beats text-based Lotus 123/WordStar/WordPerfect establishment.
2. The Amiga platform is not "second-sourced".
Quote:
Comparing CPU performance of an Amiga to a PC is generally the wrong metric. Apart from a few specific use cases, computing power is way down the list of things of importance.
In the case of the A1200, compatibility with the vast library of Amiga software titles (mostly games) was far more important than how fast the CPU could execute code. In fact having a much faster CPU could actually be counterproductive. Imagine the A1200 with a 68040 CPU - it's now over twice the price and software compatibility is in the toilet. The target market can't afford it, and they don't want it. Performance is better, but value is worse.
|
The fault for 68040's incompatibility with 68000 is on Motorola.
The fault for the disappearing C0000 memory address range is on Commodore.
The fault for hitting Kickstart's library by non-OS method is 3rd party developers. Needs Relokick.
By default, the PC (before UEFI) starts in legacy 8086's real mode. The lesson is the same, any newer Amiga should start "A500 legacy" mode until there's a specific request for advanced mode.
Quote:
Moore's law wasn't kicking CBM's butt, IBM compatibility (or the lack of) was. In 1992 the A1200 was kicking the equivalent low-end PC's butt in the area that mattered most to Amiga users at the time - 2D games.
|
For Xmas Q4 1992, there were only about 40,000 A1200 in the UK. 1993 was A1200's full sales year.
In 1992, the SNES was kicking Amiga's butt in the area that mattered most to budget gamers at the time - 2D games.
SNES = 60 million worldwide install base. Amiga = 5 to 6 million worldwide install base.
The larger Gaming PC market avoids direct competition against SNES's strong 2D gaming experience by delivering a "full 32bit" texture-mapped 3D gaming experience.
https://www.youtube.com/watch?v=uQwAE2EWdl0 1991's Formula 1 Grand Prix (PC/DOS) "World Circuit" 1991 with ground texture-mapped.
https://www.youtube.com/watch?v=KxMt-bIm5bk 1990's Wing Commander on 386DX-25 MHz.
https://youtu.be/uW7wvTn-In0 1990's Wing Commander on 386SX-33 MHz with ET4000AX.
https://youtu.be/5o9yOBBWPgM?t=3567 At 56:29 in the YouTube video, PC's 286 @16 Mhz with a fast VGA clone playing Prehistorik 2 with a parallax background and 256 colors.
Quote:
But people still bought the more expensive 386SX because it was IBM compatible, which is perfectly understandable considering PCs had 90% of the market and most computer users were not buying them to play games.
|
In absolute numbers, the gaming PC is larger than Amiga's install base.
386SX-16 with fast VGA like ET4000AX is 2D gaming capable.
Quote:
How do I know this? Because I was selling A1200s literally alongside 386SX PCs and the first thing every customer asked was "Is it IBM compatible?" - unless they were an Amiga fan and then the PC was of no interest to them. PC purchasers generally had little interest in CPU performance, struggling to understand why a 386DX system cost so much more than a 386SX. Surely the later model number meant it was better?
|
Your country is tiny New Zealand with a weaker NZ dollar and weak economies of scale.
In Australia, when the A500 was at its peak from 1989 to 1991, many of my classmates had A500s, but the signs for 1990's Wing Commander were already in play.
A500 was offered for sale across many Norman Ross (precursor to Harvey Norman) stores and my A500 starter pack was purchased from Norman Ross in 1989.
Quote:
In one case I recommended a 386DX system to a lady but she bought the SX because it was cheaper. A few years later she accused me of selling her an under-powered system, and I had to point out that it was her choice to go for the better 'value' machine. But I wasn't pushy in my sales pitch because there is a good argument that the 386SX was better value at the time, since it did everything she wanted for less money. That is what was argued in an article in the June 1994 issue of New Zealand PC World magazine. They compared buying a 386SX-16 in 1990, then a 386SX-20 in 1992, and finally a 486SX-25 in 1994, to buying a 386DX-33 in 1990 and expecting it to last 4 years. The 386DX cost $NZ7495. Going the other way cost $2495 + $2395 + $2399 for a total of $7289 - better value overall, and ending up with a machine that better suited the time.
|
From Australia's Your Computer December 1990 in AUD page 109 of 132 386DX-25 based PC = $1950 AUD 386DX-25 based PC with cache = $2350 AUD 386DX-33 based PC with cache = $3400 AUD
Configured with full tower case, 1 MB RAM, 1.2 MB FDD, serial/parallel/game port, MCGA video card, 220 watts PSU, keyboard, MS-DOS.
GCS vendor has 10 branches across Australia.
-------------------- From Australia's Your Computer December 1992 in AUD page 94 of 132 Am386DX-40 with 64K cache-based PC = $1855 Includes 4MB RAM, 120 MB HD, 1MB video RAM Paradise Windows accelerator card, SVGA monitor, 1.44MB FDD, 2 serial, 1 parallel, 1 game port, and keyboard.
1MB video RAM Paradise Windows accelerator card is $275 Trident 16/512 KB video RAM = $75 Trident 16/1024 KB video RAM = $110
--- page 91 of 132 386DX-40 (with 128K cache) based PC = $1679 (Am386DX-40) 486SX-33 based PC = $1829 These systems include 2MB RAM, 85 MB HD, 1MB SVGA card, SVGA monitor, 1.44MB FDD, 2 serial, 1 parallel, 1 game port, keyboard, mouse and MS-DOS 5.0. This vendor has four branches in Sydney and the central coast.
386DX-40 (with 128K cache) based PC = $1450 (Am386DX-40) 486SX-25 based PC = $1518.00 These systems include 2MB RAM, 85 MB HD, 512KB SVGA card, SVGA monitor, 1.44MB FDD, 2 serial, 1 parallel, 1 game port, keyboard, mouse, MS-DOS 5.0 and case.
page 90 of 132 1 MB Teseng = $105 Teseng MegaEVA/2 = $230
386DX40 with 64KB cache based PC = $1710 (Am386DX-40) 486SX25 with 64KB cache based PC = $1800 These systems include 4MB RAM, 80 MB HD, 512KB SVGA Trident card, SVGA monitor, 1.44MB FDD, 2 serial, 1 parallel, 1 game port, keyboard, mouse, MS-DOS 5.0, 200watts PSU and case.
page 44 of 132 Advert for Wolfenstein 3D
page 39 of 132 386DX-33 with 12KB cache based PC = $1549 386DX-40 with 64KB cache based PC = $1539 (Am386DX-40) 486SX-25 based PC = $1609 These systems include 2MB RAM, 105 MB HD, 1MB SVGA card, SVGA monitor, 1.2 MB FDD, 2 serial, 1 parallel, 1 game port, keyboard, mouse, MS-DOS 5.0 and case. This vendor has seven branches in NSW, VIC, and SA states.
---------------------
From Australia's Your Computer December 1991 in AUD page 97 of 132 386DX-33 based PC = $2450 include 4MB RAM, 85 MB HD, 512KB SVGA card, SVGA monitor, 1.44 MB FDD, 2 serial, 1 parallel, 1 game port, keyboard, mouse and case.
page 67 of 132 386DX-33 based PC = $1850 486SX-25 based PC = $2250 These systems include 1MB RAM, 42 MB HD, 512KB SVGA card, SVGA monitor, 1.44 MB FDD, 2 serial, 1 parallel, 1 game port and case.
page 10 of 132 386DX-33 with 128 KB cache based PC = $1895 (without tax is $1650) These systems include 1MB RAM, 85 MB HD, 512KB SVGA card, SVGA monitor, 1.2 MB FDD, 2 serial, 1 parallel, 1 game port, 200w PSU and case.
If a PC is purchased based on "work-related", income tax can be lowered.
Last edited by Hammer on 02-Jul-2024 at 08:52 AM. Last edited by Hammer on 02-Jul-2024 at 08:50 AM. Last edited by Hammer on 02-Jul-2024 at 08:48 AM. Last edited by Hammer on 02-Jul-2024 at 08:28 AM. Last edited by Hammer on 02-Jul-2024 at 06:16 AM. Last edited by Hammer on 02-Jul-2024 at 06:06 AM. Last edited by Hammer on 02-Jul-2024 at 05:34 AM.
_________________ Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68) Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68) Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB |
| Status: Offline |
| | Hammer
| |
Re: One major reason why Motorola and 68k failed... Posted on 2-Jul-2024 9:01:20
| | [ #294 ] |
| |
|
Elite Member |
Joined: 9-Mar-2003 Posts: 5859
From: Australia | | |
|
| @bhabbott
Quote:
Not only that, the business model is failing. Microsoft doesn't want you to buy a PC and use it forever without paying more, even though there has been no significant progress in a decade (I'm using a decade old PC and it does everything I want). But their attempt to continue monetizing Windows is failing. People are getting sick of it. Desktop PC sales are going down, and people are keeping their old computers for as long as possible because the new ones stink. |
Recently, Microsoft has been attempting to follow Google's Android example, but there's an argument for hard X86-64 v2 level support.
I plan to migrate towards ArchLinux for my non-WIndows 11 compliant PCs.
https://www.pcgamesn.com/gaming-hardware/pc-sales-rise-2024 Quote:
After the better part of two years of declining sales after the gaming PC boom during the COVID pandemic, PC sales are finally on the rise again. The first three months (Q1) of 2024 saw a 1.5% rise in PC sales compared to the same period last year
|
Quote:
I don't see any evidence of that. What I see is a market dominated by x86 because IBM chose the 8088. I see Motorola doing their best to match what Intel was doing, despite having a much smaller market for their CPUs. I see Motorola having no market for their 68k desktop CPUs after 1994, apart from Apple whose sales dropped to a mere 3% between 1994 and 1997. What were they to do?
|
In 1992, Motorola didn't factor in AMD's 386DX-40 wildcard.
Last edited by Hammer on 02-Jul-2024 at 09:06 AM. Last edited by Hammer on 02-Jul-2024 at 09:03 AM.
_________________ Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68) Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68) Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB |
| Status: Offline |
| | matthey
| |
Re: One major reason why Motorola and 68k failed... Posted on 3-Jul-2024 2:50:29
| | [ #295 ] |
| |
|
Elite Member |
Joined: 14-Mar-2007 Posts: 2270
From: Kansas | | |
|
| OneTimer1 Quote:
You made nothing wrong, you started when the 68k was released 1979. I started when the Amiga with 68k was released, my idea was not to show the development of a super CPU, I focused on the usage in a low cost computer like the A500.
|
Ok, different base line then. There are different ways to look at it. The 68k Amiga was more about the chipset than a super CPU. The chipset offloads the CPU for low end affordability so CPU performance is not as important as a good ISA and ease of use which the 68k has. CBM abused the offloading the CPU advantage to buy cheap CPUs thus keeping Amiga performance anemic. The 68000 was 6 years old in 1985 when the Amiga launched and the 68EC020 low end standard was 10 years out of date in 1994 when they ended. The 68000 originally used a 3500nm HMOS (NMOS) process while the 68020 originally used a 3000nm HCMOS (CMOS upgrade good but more transistors used) process. A 3000nm or 2500nm CMOS process was possible in 1979 so the 68000 was about 2 die shrinks behind and a 1500nm process by 1984 so the 68020 was about 3 die shrinks behind. The A1222 QorIQ P1022 SoC came out in circa 2008 using a 45nm process so it is 16 years old and state of the art today is 3nm which is about 11 die shrinks difference. Low end ARM emulation of the Amiga is likely to be worse performance than the A1222 but this is Amiga Neverland.
The Amiga chipset has too much missing and unverified info to draw conclusions.
Year | Amiga chipset | Chip count | Transistors | Process 1985 OCS 3 ? NMOS-5000nm 198? Ranger ? ? 1990 ECS 60,000 ? 1992 AGA 3 450,000 CMOS-1500nm/Lisa (CIA transistors included?, Alice+Lisa=300,000 transistors) ???? AA+ 2 200,000? ? (Wiki claims 200,000 transistors but unlikely even with integration savings) ???? AAA 4 750,000 ? (32 bit AAA) ???? AAA 6 1,000,000+ ? (64 bit AAA) ???? Hombre 2 ? ?
There was a huge leap in transistor budget between 1990 ECS and 1992 AGA perhaps with a CBM suicidal "no new chips" mandate in between. If 5000nm was the original chipset process, it would have been 4-5 die shrinks and about 10 years away from best silicon at that time. Paula was still used in 1994 which would have been about 19 years behind the curve. It may have been challenging to upgrade the analog parts of the chip for a CMOS SoC so this is partially understandable. A 1500nm AGA Lisa in 1992 would have been about 10 years and 7-8 die shrinks from cutting edge.
Questionable sources for chipset: https://en.wikipedia.org/wiki/Amiga_Original_Chip_Set https://en.wikipedia.org/wiki/AA+_Chipset CBM post bankruptcy docs (I don't recall link)
OneTimer1 Quote:
My Idea was when the A500 started with a CPU built of 68000 transistors its successor two years later should had a CPU with 136000 transistors, this would have been less than a 68020 but more than a 68010 and the next step would have been a 68030 EC/LC and so on.
|
More like Mac improvements and less like Amiga to C64 cost reductions?
Last edited by matthey on 03-Jul-2024 at 03:09 AM. Last edited by matthey on 03-Jul-2024 at 03:05 AM.
|
| Status: Offline |
| | Hammer
| |
Re: One major reason why Motorola and 68k failed... Posted on 3-Jul-2024 4:37:20
| | [ #296 ] |
| |
|
Elite Member |
Joined: 9-Mar-2003 Posts: 5859
From: Australia | | |
|
| @matthey
Quote:
Ok, different base line then. There are different ways to look at it. The 68k Amiga was more about the chipset than a super CPU. The chipset offloads the CPU for low end affordability so CPU performance is not as important as a good ISA and ease of use which the 68k has. CBM abused the offloading the CPU advantage to buy cheap CPUs thus keeping Amiga performance anemic. The 68000 was 6 years old in 1985 when the Amiga launched and the 68EC020 low end standard was 10 years out of date in 1994 when they ended.
|
68EC020 was available about mid-1992.
When Motorola released 68040 @ 40Mhz in May 1993, Apple released Quadra 840AV with 68040 @ 40Mhz in July 1993. Notice Apple's quick response.
Quadra 840AV shares a similar design with Quadra 660AV's 68040 @ 25Mhz.
Commodore couldn't respond in two months with 68EC020's mid-1992 release when AGA wasn't completed due to frozen status and focusing on ECS.
Commodore wasted R&D time on releasable monochrome hi-res Denise. Commodore wasted R&D time on near releasable C65 256 color display capable chipset. Commodore wasted R&D time on moonshot AAA with 64-bits VRAM via four 32-bit custom chips.
Read Commodore The Final Years by Brian Bagnall for extra 1986 to 1989 details.
Amiga Ranger R&D was canceled in mid-1986. Commodore wasted time on monochrome hi-res Denise vs color hi-res Denise debates.
By mid-1987, Commodore was suprised by IBM and Apple was catching up with 256 color displays. Commodore didn't factor in IBM's 1986 MCGA's mode 13h and D15 port that was reused for VGA.
PC's 256 color display and 640x480p monitor signal use case existed with IBM's PGC in 1984. A small adjustment to PGC monitor's H. FREQ pot will display VGA. https://www.vogons.org/viewtopic.php?p=1090799#p1090799
IBM's 1984 PGC has 256 colors at 640x480p which established the use case for 256 colors and double scan 640x480p resolution.
Quote:
Low end ARM emulation of the Amiga is likely to be worse performance than the A1222 but this is Amiga Neverland.
|
Emu68 with RPi 4B @ 2.0 Ghz is competitive against SAM460 @ 1Ghz and A1222.
Quake 1 demo1 software rendering at 640 x 480:
On CaffeineOS R928's Clickboom Quake 68K port on A1200-PiStorm-Emu68-stock RPi 4B's ARM Cortex A72 @ 2.0 Ghz with default voltages,
Using 68040/68060 copy functions, 640x480 8 bit demo1 = 59.7 fps. 640x480 8 bit demo2 = 68.2 fps.
I would need to upgrade the PSU or remove Indivision AGA MK3 since the PSU is running on the very warm side.
-----------
https://www.amiganews.it/forum/viewtopic.php?f=37&t=17550&start=60#p236718 Quake 1 software rendering
640 x 480 AmigaOneXe @ 933 Mhz = 47.8 fps 640 x 480 Sam460 @ 1Ghz = 61 fp
Quake1 on A1222 has issues, read http://www.eliyahu.org/tabor/games.html
Last edited by Hammer on 04-Jul-2024 at 06:24 AM. Last edited by Hammer on 03-Jul-2024 at 02:30 PM. Last edited by Hammer on 03-Jul-2024 at 02:29 PM. Last edited by Hammer on 03-Jul-2024 at 05:05 AM. Last edited by Hammer on 03-Jul-2024 at 05:05 AM. Last edited by Hammer on 03-Jul-2024 at 04:59 AM. Last edited by Hammer on 03-Jul-2024 at 04:55 AM. Last edited by Hammer on 03-Jul-2024 at 04:52 AM. Last edited by Hammer on 03-Jul-2024 at 04:49 AM. Last edited by Hammer on 03-Jul-2024 at 04:47 AM.
_________________ Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68) Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68) Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB |
| Status: Offline |
| | ppcamiga1
| |
Re: One major reason why Motorola and 68k failed... Posted on 3-Jul-2024 10:37:51
| | [ #297 ] |
| |
|
Cult Member |
Joined: 23-Aug-2015 Posts: 858
From: Unknown | | |
|
| @Hammer
emu68 is emulator it always be many times slower than native code use quake on linux it run native full screen with opengl support
by using pistorm you change your amiga into nothing more than joystick mouse keyboard interface for rpi. it is extremelly stupid way to waste money use ordinary joystick mouse keyboard it is cheaper and better
|
| Status: Offline |
| | pixie
| |
Re: One major reason why Motorola and 68k failed... Posted on 3-Jul-2024 12:48:55
| | [ #298 ] |
| |
|
Elite Member |
Joined: 10-Mar-2003 Posts: 3287
From: Figueira da Foz - Portugal | | |
|
| @ppcamiga1
Quote:
by using pistorm you change your amiga into nothing more than joystick mouse keyboard interface for rpi. |
I guess you couldn't even say that about 'your' ppc machine...
_________________ Indigo 3D Lounge, my second home. The Illusion of Choice | Am*ga |
| Status: Offline |
| | Hammer
| |
Re: One major reason why Motorola and 68k failed... Posted on 3-Jul-2024 14:19:19
| | [ #299 ] |
| |
|
Elite Member |
Joined: 9-Mar-2003 Posts: 5859
From: Australia | | |
|
| @ppcamiga1
Quote:
ppcamiga1 wrote: @Hammer
emu68 is emulator it always be many times slower than native code use quake on linux it run native full screen with opengl support
|
Emu68 is a bare metal emulator just for the CPU, everything else is Amiga's Auto-Config "plug and play" and AmigaOS device drivers e.g. Broadcom Video Core, WiFi.
PRi's device drivers are written for AmigaOS and the intended hardware device.
There's no Linux host with Emu68.
Warp3D acceleration for GLQuake, Quake 2 GL and Quake 3 is functional for Broadcom IV IGP.
Quote:
by using pistorm you change your amiga into nothing more than joystick mouse keyboard interface for rpi.
|
In terms of concept, PiStorm-RPi is similar to Phase 5's BlizzardPPC with BVision RTG.
RPI's RAM is Amiga's bare metal Fast RAM.
Look in the mirror.Last edited by Hammer on 04-Jul-2024 at 06:32 AM. Last edited by Hammer on 03-Jul-2024 at 02:25 PM. Last edited by Hammer on 03-Jul-2024 at 02:23 PM.
_________________ Amiga 1200 (rev 1D1, KS 3.2, PiStorm32/RPi CM4/Emu68) Amiga 500 (rev 6A, ECS, KS 3.2, PiStorm/RPi 4B/Emu68) Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB |
| Status: Offline |
| | Mr-Z
| |
Re: One major reason why Motorola and 68k failed... Posted on 3-Jul-2024 17:45:42
| | [ #300 ] |
| |
|
Regular Member |
Joined: 24-May-2005 Posts: 191
From: De Keistad, Netherlands | | |
|
| @Hammer
ppcamiga1 will never understand how Emu68 works spare yourself the effort
_________________ Amiga is additive coz it is fun to use |
| Status: Offline |
| |
|
|
|
[ home ][ about us ][ privacy ]
[ forums ][ classifieds ]
[ links ][ news archive ]
[ link to us ][ user account ]
|