@ferrels

Quote:

ferrels wrote: That was my exact experience. I spent a small fortune on an A1200 with an 030/50 Mhz accelerator with FPU and 16MB of FastRAM. Even with the accelerator my A1200 was anemic and CPU benchmarks put it in the same class as a 486-33Mhz and it ran so hot that it nearly melted the case. Lack of chunky graphics was the nail in the coffin for me. My work PCs literally ran rings around my A1200 graphically as well as with CPU intensive tasks so I quickly boxed the A1200 and sent it off to be stored for 12 years before I even got it out again for the sake of nostalgia.

This is what's called 'setting up to fail'.

50MHz 030 accelerator cards were not available for the A1200 on launch. The first 50MHz boards came out in 1994, and they were initially quite expensive. But the A1200 itself wasn't. If you bought an A1200 to put an accelerator card in later then you were trying to do it on the cheap. If you bought both together then you were not buying the latest tech - so why were you expecting it? A 50MHz 030 is about the same processing power as a 40~50MHz 386DX, and general AGA graphics performance about the same as the typical ISA bus VGA card in a 386 PC (though much better for 2D games that were optimized for it).

I love the way Amiga fans complain about their cheap low-end A1200s not having the power of a high-end 486. Did they think Commodore had some magic ingredient that allowed them to make similar hardware perform way better than the price indicated? Perhaps in 1985 they did by being more innovative, but not in 1994. In the 90's PC chip makers were constantly pushing their designs to the limit of what could be achieved, taking full advantage of Moore's law to jam more transistors into faster and larger chips, with huge economy of scale making them cheaper.

Performance skyrocketed as prices dropped, so by 1995 when Windows 95 came out there were systems powerful enough to run it. But not if you had a 2 year old 386DX or SX. I had a 386DX-40 with 4MB of RAM and it ran Windows 95 quite sluggishly, as expected. I certainly did not expect 486DX-33 performance from it. Similarly I didn't expect A4000-40 performance from my A1200 - even after putting a 50MHz Blizzard 1230-IV in it (the combination still costing significantly less than than an A4000 or name-brand 486 system).

This reminds me about a rich farmer's wife who came into my shop looking for a 'multimedia' computer to help with their kid's education. I gave her three brochures with the specs of a low-end (386SX), mid-range (386DX), and high-end (486) system. I gently suggested the 386DX or 486, but she decided to go for the SX because it was cheaper. A few years later she asked me to do some maintenance on it, and complained about it not having the performance she was expecting. I pointed out to her that she bought the low-end model which I did not recommend. Yes, sometimes the sales person is right, and you do get what you pay for!

@bhabbott

Quote:

Yes, I know that - but information on this systems is hard to find. It was apparently released in 1983. It was not intended to be used as a home or personal computer.

Reminder, Amiga 1000 was released in 1985 and Amiga 500 was released in 1987.

Apple Lisa was released in 1983 and it was expensive.

Atari 260ST was launched in Europe on a limited basis. Atari 520STm was released in late 1985.

In 1989, my Dad looked at our first 16-bit era computer and Amiga 500 was recommended by my older cousin who owned a PC 286-16 Mhz with a VGA clone. My Dad looked at Commodore 128, Commodore Colt, 286 PC clones, Atari STe, and Amiga 500 (Rev 6).

Atari STE and Amiga 500 (Rev 6) were the final selection.

Around 1991, my Dad purchased a used IBM PS/2 Model 55SX (386SX-16 with IBM VGA) for $250 when the owner (who was one of my Dad's many contacts in the corporate world, accountant background) upgraded to 486 based PC. I learn to use PC-DOS/MS-DOS from this PS/2 Model 55SX and we have many business software from work e.g. Display Write, Word Perfect, Lotus 123, Havard Graphics and 'etc', and Windows 3.0.

My Dad's corporate friends have A2000s and most of my A500 games are obtained from this source. For my Dad's interest, A500 is used like an Xbox sports gaming machine. Many of my school classmates have A500s and one of my school friends has A1000 and A500.

In 1992, our IBM PS/2 Model 55SX was sold for 386DX-33-based PC when MCA add-on cards weren't priced competitive. Ex-corporate Amiga 3000/030 was also purchased in early 1992 for $850 AUD along with our A500 trade-in. My Dad wasn't aware of Q4 1992 A1200/A4000 release.

Going from A3000/030 (25Mhz CPU, ECS) into A1200 (020 @ 14Mhz, AGA) can be considered as a side grade and A4000/030 wasn't available in Q4 1992 and A4000/040 was expensive.

Can't you see the SKU gap between A1200(020@14Mhz) and A4000/040 in Q4 1992?

From Q4 1992 respective and A4000/040's expensive pricing, my Dad completely shifted towards gaming PC.

When our 386DX-33-based PC was upgraded with ET4000AX and Sound Blaster Pro, this machine acted like an A1200-A4000/030 games machine with Q4 1993 Doom capability.

A4000/EC030 @ 25 Mhz was released in April 1993, it was too late for Xmas 1992 sales and it was priced like 486SX-25 a based PC clone.

This topic is about AGA and the 1992-1993 time period.

AGA being completed in March 1991 and releasing AGA in Q4 1991 is very important. Many of my school friends' parents had purchased 386DX-33-based and 486SX gaming PC from 1992 to 1993.

Before Q4 1993 Doom, 1990 Wing Commander VGA has a major impact on PC gaming's rise.

Quote:

Compaq certainly pulled out all stops to produce the Deskpro 386. But the result was very expensive, and the 32 bit bus proprietary. Very few PC users needed it or could justify the price. If you think the existence of this machine means the Amiga should have used a 386 CPU, you are wrong. First off it was much too 'late to the party', and it was in a completely different class. 6 years later the A3000 was a match to the Deskpro 386 selling at that time (I considered buying one instead of the A3000, but luckily I decided against it - the A3000 was much more upgradable).

1986, Compaq Deskpro 386 (32-bit front side bus) has four 16-bit slots and three 8-bit slots and it was beyond my Dad's $1500 AUD budget in 1989. As long as there is a 16-bit ISA slot, faster SVGA cards like ET4000AX can be used.

1987, IBM PS/2 Model 55SX has a 16-bit proprietary MCA bus. Compaq reached its $1 billion revenue around 1987.

In 1987, Compaq wasn't the only 386-based PC vendor. Direct competition has downward pressure on PC prices.

This topic is about AGA and the 1992-1993 time period.

The American, Canadian, and Australian dollars are stronger than the New Zealand dollar. The New Zealand market is tiny relative to the American, Canadian and Australian markets.

Last edited by Hammer on 17-Oct-2022 at 08:52 AM.
Last edited by Hammer on 17-Oct-2022 at 08:46 AM.
Last edited by Hammer on 17-Oct-2022 at 08:42 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@Hammer

Quote:

Hammer wrote: That's a flawed argument when PC DOS Doom sales smashed under a million A1200 install base. https://doom.fandom.com/wiki/Sales Based on these data, one may place the sales until 1999 in the range 1.9–2.1+ million for Doom (including Ultimate Doom), and 1.5–1.8+ million for Doom II. It is probably safe to deduce that as of 2005, both Doom and Doom II have sold well over 2 million copies each

Er, no - it's completely irrelevant.

But...
Quote:

In a press release by Mindscape Entertainment dated July 19, 1999 [5], another report from PC Data is cited according to which "Doom" sold 2.9 million copies between 1994 and 1999. It is not clear whether this refers to Doom and Doom II, Doom and Ultimate Doom, or some other combination. If it refers to Doom and Doom II, it is inconsistent with the aforementioned figure from PC Data which puts the combined sales for shareware Doom and Doom II at 3.2 million (the difference of one year hardly provides sufficient explanation; it would imply that in the first 20 days after its release, shareware Doom sold 300,000 more copies than Doom and Doom II did combined in 1999). It does however make some sense if taken to mean Ultimate Doom and Doom II combined. In any case, it can probably be ignored in favor of the more specific numbers from the 1993–1998... Due to rampant piracy in Asia, id Software sold the rights to sell the game to the biggest pirates for $1 per copy

All this tells us is that between 1993 and 1999 there were a huge number of PCs that presumably could run Doom, which is not a revelation.

The Asia thing is interesting though. The Amiga also suffered from rampant piracy. I wonder if that might have been a factor in ID not supporting it? Of course we now know that a properly done Doom port would have been a best seller to honest owners of suitably equipped Amigas (A4000, A1200 with accelerator card etc.). But of course that would be a much smaller number than the total for all platforms. ID probably figured the potential sales numbers didn't justify the cost of porting it.

That irrelevancy aside, it is a fact that in 1992 (when the A1200 was released) the majority of PC's were 386SX's or 3866DX's. 486's were high-end premium machines, not at all in the same class as the A1200. And those 386's all had ISA bus graphics cards. In my shop I even sold a 386SX system with equivalent specs to an A1200/40 with monitor for the same price. The A1200 beat the pants off the 386SX in 2D graphics performance (I didn't even try Doom on the PC because I knew its performance would suck). For that price the PC didn't have a sound card or speakers either, so games were meh. Finally, the PC came in a slimline case with limited upgradability. If the owner wanted much better performance down the road their only choice was to buy a whole new machine.

@bhabbott

Quote:

bhabbott wrote: 50MHz 030 accelerator cards were not available for the A1200 on launch. The first 50MHz boards came out in 1994, and they were initially quite expensive.

https://archive.org/details/amiga-world-1993-10/page/n7/mode/2up
From Amigaworld, October 1993, In USD.

Page 66 of 104
Amiga 4000/040 @ 25Mhz for $2299 (LOL, )
Amiga 4000/030 @ 25Mhz for $1599

Page 82 of 104
Microbotics M1230X's 68030 @ 50Mhz has $349 USD!
1942 Monitor has $389
A1200 with 85MB HDD has $624
A1200 with 130MB HDD has $724


VS


https://vintageapple.org/pcworld/pdf/PC_World_9308_August_1993.pdf
Gateway Party List, Page 62 of 324

4SX-33 with 486-SX 33Mhz, 4MB RAM, 212MB HDD, Windows Video accelerator 1MB video DRAM, 14-inch monitor for $1495,

4DX-33 with 486-DX 33Mhz, 8MB RAM, 212 MB HDD, Windows Video accelerator 1MB video DRAM, 14-inch monitor for $1795,

Remember Gateway?

Page 292 of 324
From Comtrade
VESA Local Bus WinMax with 32-Bit VL-Bus Video Accelerator 1MB, 486DX2 66 Mhz, 210 MB HDD, 4MB RAM, Price: $1795


https://vintageapple.org/pcworld/pdf/PC_World_9310_October_1993.pdf
October 1993, Page 13 of 354,
ALR Inc, Model 1 has Pentium 60-based PC for $2495.


https://archive.org/details/bub_gb_hqQJaNzN9IcC/page/n603/mode/2up
PC Mag 1992-08, page 604 of 664,
Diamond Speedstar 24 (ET4000AX ISA) has $169.

-------------

Last edited by Hammer on 17-Oct-2022 at 09:07 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@bhabbott

Quote:

Er, no - it's completely irrelevant.

It's relevant for gaming PCs that can run Doom (Q3 1993). If the frame rate is not smooth enough, upgrade the PC.

PC clone is modular with AT motherboard factor standards e.g. my entire A3000 has to be thrown away for the AGA machine.

Fact: Commodore is not willing to sell A2000/A3000 AGA motherboard upgrades. Commodore is treating desktop Amigas like a throwaway game console. LOL

Fact: Any PC 386 with VGA and 4MB RAM enables the user to test Doom and upgrade.

32-bit era PC gaming's rise started in the 1990 Wing Commander moment.

Fact: full 32-bit 020/030 Amigas with OCS/ECS can NOT run Doom until the later EHB version or RTG from 1995 to 1996.

Fact: for Q4 1992, the 32-bit bus enabled 68K AGA install-base started from ground zero.

Fact: A1200 was released in Q4 1992 in limited numbers! The 1993 A1200 couldn't match A500's 1990 sales.


In 1991, Intel is four to five times larger when Commodore reaches $1 billion in revenue. Intel X86 unit sales are mostly unified under a single MS-DOS platform standard.

Last edited by Hammer on 17-Oct-2022 at 04:48 PM.
Last edited by Hammer on 17-Oct-2022 at 09:30 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@bhabbott

https://arstechnica.com/gaming/2017/01/a-history-of-the-amiga-part-10-the-downfall-of-commodore/

REAL FACTS


Jump the lake.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@Hammer

Quote:

Hammer wrote: Can't you see the SKU gap between A1200(020@14Mhz) and A4000/040 in Q4 1992?

Huh? Yeah, I think I can see the difference. The A1200 is the successor to the A500, the A4000 the successor to the A2000/3000. AGA chipset is the same in both machines. CPU and memory are obviously much better in the A4000, but A1200 has a slot big enough to take whatever CPU comes out in the future? (yes). So the A1200 is the poor man's A4000, like the A500 was the poor man's A2000. Except the A1200 can become much more than an A500 could.

Quote:

Going from A3000/030 (25Mhz CPU, ECS) into A1200 (020 @ 14Mhz, AGA) can be considered as a side grade and A4000/030 wasn't available in Q4 1992 and A4000/040 was expensive.

In 1991 I paid NZ$7200 for my A3000. Why would I expect my A1200 to be better than it? Yet a few years later it was. I had a 40MHz 030 in it with SCSI and all kinds of stuff hanging off it. We used it in the shop for demoing software, word processing, sending and receiving faxes, scanning photos and printing them etc. It was a real workhorse as well as fun to relax and play games on.

Quote:

1986, Compaq Deskpro 386 (32-bit front side bus) has four 16-bit slots and three 8-bit slots and it was beyond my Dad's $1500 AUD budget in 1989.

Compaq Deskpro 386 system was ~NZ$8000 here in 1990. I used that price to justify buying an A3000 (and didn't regret it).

Quote:

The American, Canadian, and Australian dollars are stronger than the New Zealand dollar. The New Zealand market is tiny relative to the American, Canadian and Australian markets.

Er, so? Currency conversion is easy. What's harder is factoring in freight costs, import duties, markups etc. NZ had access to cheap PCs because we had a trade agrement with Taiwan. I delt with two wholesalers here who were Asians with ties to manufacturers in Taiwan. They had excellent prices. We used to buy built-up PCs as well as parts for upgrades and building our own custom PCs. Good times! If you are interested I recently discovered some documents sitting on my A1200's hardrive from 1996/7, including wholesale price lists. Brings back some memories...

@Hammer

You still don't seem to get it. Commodore's unit sales and revenue have nothing to do with how good (or not) the AGA chipset was.

BTW that graph is a bit deceptive. Sales and revenue at the end of 1994 were 'zero' because Commodore didn't exist. On the day the bankruptcy was announced we were still selling plenty of Amigas. My only regret was in not stocking up beforehand.

But I've said it before and I will say it again - Commodore needed to die and this was the right time. Had they carried on the Amiga would have lost its identity, either due to having too many architectures or becoming too much like a PC (probably both). The 'disappointment' of AGA being 'not that much better than OCS' is actually its strength. Some great new Amiga games now being developed for OCS abolutely fly on AGA - the benefit of coding for limited hardware. Yet they still maintain the character of games of that era, when every pixel counted and the graphics were real works of (digital) art. Today's 3D games strive to be more realistic and yet end up losing meaning.

Clearly Amiga was the last thing that kept CBM afloat.
Too bad that Medhi Ali & co did not like it.

AGA was OK in 1993.
Nice for video stuff, suberb for multimedia, nice for still GFX editing.
Nice for Audio.
Poor for high-res cad etc. (flixer at 1440x580).
Poor speed for some high color use cases.

CBM policy was a rotten one. R&D money was used in jet fuel etc.

CD32 and A1200 should have had some fast RAM.
Every A1200 should have been shipped with HDD.
A500+, A600 were futile.

In 1993/1994 CBM should have had 2D RTG + RTA ready for third party chip makers.

AAA would have also been ok, if released as originally planned, but anyway RTG more important.

IIRC, Hombre might have taken "Amiga" to windows and sure death vs GFX+Audio chip makers.

Last edited by KimmoK on 17-Oct-2022 at 01:39 PM.
Last edited by KimmoK on 17-Oct-2022 at 01:38 PM.
Last edited by KimmoK on 17-Oct-2022 at 01:36 PM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@bhabbott

Quote:

In 1991 I paid NZ$7200 for my A3000. Why would I expect my A1200 to be better than it? Yet a few years later it was. I had a 40MHz 030 in it with SCSI and all kinds of stuff hanging off it. We used it in the shop for demoing software, word processing, sending and receiving faxes, scanning photos and printing them etc. It was a real workhorse as well as fun to relax and play games on.

https://www.poundsterlinglive.com/bank-of-england-spot/historical-spot-exchange-rates/usd/USD-to-NZD-1991
U.S. Dollar / New Zealand Dollar Historical Reference Rates from Bank of England for 1991


https://archive.org/details/amiga-world-1991-06/page/n87/mode/2up
Amiga World Magazine (June 1991) page 88 of 104 in USD
From Commodore,
Amiga 3000-25/100 has $1500 (25 Mhz 68030, 100MB HDD)
Amiga 3000-25/50 has $1250 (25 Mhz 68030, 50MB HDD)
Amiga 3000-16/50 has $1150 (16Mhz 68030, 50MB HDD)

Amiga 3000-25/100 should cost about $2550 NZD to $3000 NZD when directly translated from USD to NZD.

My Dad purchased an ex-corporate Amiga 3000-25/120 (68030 @ 25Mhz, 120 MB HDD, 1 MB Chip RAM, 1 MB Fast RAM) for about $850 AUD with A500 1MB trade-in. Amiga Vision project data files are still on the hard disk.

$7200 NZD is abnormally high unless it has other extras e.g. 1992 OpalVision Card for PAL or VideoToaster with PAL converter (e.g. GVP TBC+), extra RAM, and hard disks. NZ's weaker economies of scale seem to be a factor.


Quote:

Compaq Deskpro 386 system was ~NZ$8000 here in 1990. I used that price to justify buying an A3000 (and didn't regret it).

https://archive.org/details/pc_world_australia-1991_05/page/5/mode/2up

PC World Australia May 1991, Page 67
OCT 386C-33 PC has $2,445 AUD.
OCT 386-25 PC has $1,845 AUD.
OCT 386SX-16 PC has $1,295 AUD

Last edited by Hammer on 17-Oct-2022 at 04:53 PM.
Last edited by Hammer on 17-Oct-2022 at 04:24 PM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@bhabbott

Quote:

bhabbott wrote: @Hammer You still don't seem to get it. Commodore's unit sales and revenue have nothing to do with how good (or not) the AGA chipset was. BTW that graph is a bit deceptive. Sales and revenue at the end of 1994 were 'zero' because Commodore didn't exist. On the day the bankruptcy was announced we were still selling plenty of Amigas. My only regret was in not stocking up beforehand.

You didn't read my 1993 A1200 sales i.e. there's a full year for AGA sales in 1993!

1993 was the 'make or break', according to David Pleasance.

AGA has a major install base issue while PC 386DX/486SX/486DX's sales weren't gimped.

By 1994, only the operations in Canada, Germany, and the United Kingdom were still profitable.
Commodore International announced voluntary bankruptcy liquidation on May 6, 1994.

There's no way in hell that AGA can narrow the massive install base gap with gaming PC 386DX-33, 486SX-25, and 486DX-33.

Last edited by Hammer on 17-Oct-2022 at 04:43 PM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@KimmoK

With a fast CPU and Fast RAM, AGA was okay for Doom and Wing Commander AGA.

Without a fast CPU, the AAA chipset alone wouldn't solve the Doom problem.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@bhabbott

Quote:

But I've said it before and I will say it again - Commodore needed to die and this was the right time. Had they carried on the Amiga would have lost its identity, either due to having too many architectures or becoming too much like a PC (probably both). The 'disappointment' of AGA being 'not that much better than OCS' is actually its strength. Some great new Amiga games now being developed for OCS abolutely fly on AGA - the benefit of coding for limited hardware. Yet they still maintain the character of games of that era, when every pixel counted and the graphics were real works of (digital) art. Today's 3D games strive to be more realistic and yet end up losing meaning.

https://screenrant.com/genshin-impact-fortnite-gta5-first-year-revenue/


Genshin crossed the $1 billion mark six months after its release.

Genshin Impact's impressive first-year earning puts them ahead of successful first year launches like Fortnite, Grand Theft Auto V, and Call of Duty: Modern Warfare.

Anime toons artwork style for win... billions $$$.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

Quote:

With a fast CPU and Fast RAM, AGA was okay for Doom and Wing Commander AGA.

Any dumb framebuffer (that's what AGA is with Doom) is okay with a fast CPU + RAM.

Last edited by Leo on 17-Oct-2022 at 05:44 PM.

_________________
http://www.warpdesign.fr/

@Hammer

Quote:

Hammer wrote: @KimmoK With a fast CPU and Fast RAM, AGA was okay for Doom and Wing Commander AGA.

With a fast CPU and Fas RAM AGA was (is) little more then an overcomplicated framebuffer.

Thats how it was used in those games and and pretty much all the 060/PPC demos.

Thats something even a basic 16Bit VGA could do better in 93 and hence AGA was to little to late.

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

cdimauro Quote:

Possible. But a guy in a post said that Lightwave increased the speed by 300% when it was properly compiled for the 68040. So, I assume that OxyPatcher wasn't required at least for this processor and which also means that it was good for the 68060 as well, since it had an extra instruction (which was removed on the 68040).

SAS/C may generate more optimal code for the 68040 and maybe even the 68060 but it is still using plenty of trapped 6888x FPU instructions. It takes a significant amount of work to update the backend and support code of a compiler and remove all trapped instructions. I'm not even sure GCC has done it for 68k FPU code generation although the number of 6888x instructions has been greatly reduced to where they are uncommon and don't make much difference to performance. Vbcc shouldn't use any trapped instructions when compiling for the 68040 or 68060 which is more important for embedded use where trap support like the 68060SP may not exist. It was a little easier to change because the backend was simple, not generating trapped instructions to begin with. It was primarily the fp math libraries that needed to be changed (and created in the case of the m060.lib) which is what I worked on. Despite the simplicity of the vbcc 68k FPU support, it appears to work well and outperforms GCC and SAS/C in floating point performance on the 68060.

If you have the last versions of Lightwave, try diassembling it. The version of ADis I updated does a good job of disassembling it. I could have started optimizing it but it is big. Maybe back in the desktop video days if I had my knowledge and the tools I have today, I could have optimized/patched it for the 68040 and 68060 and there would have been money in it. I'm surprised it wasn't simply compiled with GCC for the Amiga which should have been an improvement. GCC was one of the most logical candidates for compiling the x86 version of Lightwave but if it was used then why didn't Newtek compile the Amiga version with GCC also?

cdimauro Quote:

What was left I think it was the instructions ordering for the 68060's superpipeline.

The Diab Data 3.4A C compiler is the only compiler I know of with an instruction scheduler for the 68060. Motorola had developed their own compiler which they were using for the 88k just a couple of years earlier but stopped development to save money and worked with other compiler developers. Mitch Alsup complains about this being the start of the downward spiral at Motorola and he has experience working on compilers. The Diab Data compiler/toolchain is owned by Wind River now and switched to LLVM (Wind River is the VxWorks OS embedded guys). LLVM doesn't sound like a good embedded compiler with its bloat but I guess all compiling is done remotely anymore. Maybe a smaller footprint compiler with few dependencies like vbcc isn't needed for embedded systems.

An instruction scheduler for vbcc would be especially good for the 68k. It could do optimizations that Volker expects the assembler to do but can't (vasm can only do relatively simple peephole optimizations). The instruction scheduler has information where the branches are so can do more advanced optimizations besides instruction scheduling. A good example is the following.

movem.l d2,-(sp) ; 4 bytes pOEP only

to

move.l d2,-(sp) ; 2 bytes pOEP|sOEP

Vasm can't make this optimization because MOVEM.L does not affect the flags while MOVE.L does. With branch information, I believe the instruction scheduler could optimize this in most cases while keeping the backend simpler and optimizations modular like Volker wants. An instruction scheduler could provide optimizations for earlier 68k CPUs which don't need instruction scheduling although scheduling all code for the 68060 is worthwhile as it is no worse on earlier CPUs (except for 68040 optimized FPU code which is better with parallel FMOVE scheduling).

cdimauro Quote:

As you know, I don't like those synthetic benchmarks: I always prefer real world applications. A rich set, if possible.

The ByteMark benchmarks come from real algorithms like the SPEC benchmarks. Lightwave rendering is a good benchmark but I expect it shows the weakness of SAS/C code generation on the Amiga likely partially saved by OxyPatcher when used. There is likely no SAS/C 68060 instruction scheduling either but the strength of the in-order 68060 with unscheduled code still shines through.

cdimauro Quote:

They look strange to me. On the usual code (not fully optimized for the 68060) 16-bit instructions didn't even reached 50% of the total (on average). The 68060 was able to execute a pair of instructions only if they were both 16-bit. Let's simplify the calculations and say that 16-bit instructions were exactly 50% of the total. It means that the probability to have (and execute) a pair of them is 50% * 50% = 25%. So, very very distant from the above numbers.

Don't be fooled by the 4 byte/cycle instruction fetch of the 68060 even though it is half of the less efficient Pentium and PPC 603e competition. The key is the decoupled fetch (IFP) and execution (OEP) pipelines with a FIFO between them. The IFP consistently fetches, predecodes instructions and places them in the FIFO when the OEPs are stalled or executing multi-cycle instructions. This is often enough that the FIFO rarely empties with just a 4 byte/cycle fetch and the FIFO often contains enough instructions for multi-issue into the OEPs. The IFP may spend multiple cycles fetching a long instruction but it doesn't matter unless the FIFO is nearly empty. Too many large instructions together can starve the FIFO as can a large instruction after a pipeline refill but these are rare as the average instruction length is less than 3 bytes. An 8 byte/cycle fetch may have increased performance slightly but it is not worthwhile until other improvements are made, especially cache size increases.

Some 68k developers believe the instructions coming out of the FIFO are very restricted in size and thus the instructions multi-issued are weak. I believe this is due to various errors and lack of clarity in 68060 documentation.

FIFO instructions (i+1),(i+2) or (i+2),(i+3) issued to the OEPs

Some developers believe i=2B because a basic 68k instruction or OP=2B. The 68060UM says that the FIFO is 96B and other documentation states that there are 16 FIFO locations (96B/16=6B). Even with this belief, i=6B already allows much stronger instructions to be executed. I suspect there is an error in the 68060UM and the 96B FIFO should actually be a 16x96b FIFO.

Superscalar Architecture of the MC68060 Quote:

The four-stage instruction fetch pipeline, which the FIFO instruction buffer decouples from the dual execution pipelines, performs the chip's instruction prefetch. The third stage is a table lookup that uses the 16-bit opcode to produce a 32-bit longword of decode information. ... Another field provided by the early decode is the instruction length. Having this information available allows packaging of the stream of fetched variable-length instruction into machine instructions before they move into the instruction buffer. The buffer provides 16 storage locations, each location containing a 16-bit operation word, 32-bit extension words, and the early decode information. Typically, instructions require a single instruction buffer entry, although the more complex instructions require multiple locations.

16b OP (operation word) + 32b (extension words) + 32b (early decode) = 80b

Joe Circello is describing a 16x80b FIFO which is significantly larger than the 16x48b (16x6B=96B) FIFO of the 68060UM. Now what I believe Joe left out.

16b (OP) + 16b (brief/full extension word) + 32b (immediate/displacement) + 32b (early decode) = 96b

The 16b OP and 16b brief/full extension word together provide basically a 32 bit instruction. The brief/full extension word has to be read to calculate the instruction length when using (d,An,Xn*Scale) addressing modes. There can be two brief/extension words for MOVE mem,mem but these instructions are broken into 2 RISCier FIFO instructions and the load half executed separate from the store half (pOEP until last & 2 cycles). The 32b immediate/displacement is what Joe calls extension words of the instruction. I believe this matches the capabilities of the 68060 with instructions that require multiple brief/full extension words or more than 32 bits of immediate/displacement data requiring more than one FIFO instruction slot, requiring more than one cycle to execute and do not multi-issue. An internal 96b fixed length instruction format allows for powerful 68k instructions to fit in one FIFO instruction slot to be multi-issued (pOEP|sOEP). The P6 Pentium and successors used an internal 118b fixed length RISC format for awhile although this was eventually widened for more powerful instructions. A 64 bit 68060 would likely use an internal 128b fixed length instruction format for 64 bit immediates/displacements. This should allow single cycle double precision FPU immediates which could then fit in one FIFO instruction slot as well.

cdimauro Quote:

Thanks for the above data. Which confirms the stupid decision to drop the 68k family in favor of RISCs. Bah...

The funny thing is that the 32 bit 88110 integer registers were 32 bit so pairs of 32 bit registers were needed to do 64 bit integer math. This was actually not uncommon on the 88k which even did SIMD on register pairs. This was considered acceptable for a RISC CPU.

The 64 bit integer math instructions on the "CISC" 68060 had to be removed to be more "RISC" like though. It's not like Motorola philosophy changed in 2 years. It's more like start axing features to save transistors, area and chip cost for embedded use. RISC philosophy even then seems to be keep it simple and today we have micro-coded OoO monstrosities with CISC like features except for reg-mem accesses, at least not yet. Here is another good quote by Mitch Alsup.

Mitch Alsup https://groups.google.com/g/comp.arch/c/SrAHwT4lfEo Quote:

A 16KB L1 cache (small) is already bigger than the register file, forwarding logic, and all integer execution stuff, and other interfaces this section talks to. Adding RMW operations to the cache does not add "that much" logic or "that much" to verification.

The last, most important and most taboo separation of RISC and CISC isn't really that much. The caches use more logic than reg-mem which saves caches. Requiring more caches must be more RISC like though. Moore's law is over and the battle to get to a newer chip fab process first has changed into a battle of CPU design and ISA/ABI efficiency.

cdimauro Quote:

Yes, it had many advantages. But Pentium had also its own: 64-bit data bus, larger instructions fetch buffer (so, being able to pair longer instructions), all instructions supported (no trapping and/or software patches or special library versions needed).

The 64 bit data bus increased cost of not just the CPU due to more pins but also memory required for it while providing a small performance benefit. A good compromise would have been to allow 32 or 64 bit memory like the PPC 603(e) allowing for cheaper or higher performance memory but it still requires more pins/multiplexing and logic. It's more compelling to use a 64 bit data bus today as 64 bit memory is more common and cheaper. Most high performance CPUs are integrated into a SoC where as much memory bandwidth as possible is more advantageous.

The 68060 decoupled IFP and OEPs handle variable length instructions including larger instructions with elegance and efficiency instead of Pentium brute force as explained above. Any Pentium performance advantage here is "marginally faster". The only time there is a performance advantage is when the FIFO is nearly empty (i

Last edited by matthey on 18-Oct-2022 at 08:58 AM.
Last edited by matthey on 17-Oct-2022 at 11:26 PM.
Last edited by matthey on 17-Oct-2022 at 11:20 PM.

@bhabbott

Quote:

bhabbott wrote: @ferrels Quote: ferrels wrote: That was my exact experience. I spent a small fortune on an A1200 with an 030/50 Mhz accelerator with FPU and 16MB of FastRAM. Even with the accelerator my A1200 was anemic and CPU benchmarks put it in the same class as a 486-33Mhz and it ran so hot that it nearly melted the case. Lack of chunky graphics was the nail in the coffin for me. My work PCs literally ran rings around my A1200 graphically as well as with CPU intensive tasks so I quickly boxed the A1200 and sent it off to be stored for 12 years before I even got it out again for the sake of nostalgia. This is what's called 'setting up to fail'. 50MHz 030 accelerator cards were not available for the A1200 on launch. The first 50MHz boards came out in 1994, and they were initially quite expensive. But the A1200 itself wasn't. If you bought an A1200 to put an accelerator card in later then you were trying to do it on the cheap. If you bought both together then you were not buying the latest tech - so why were you expecting it? A 50MHz 030 is about the same processing power as a 40~50MHz 386DX, and general AGA graphics performance about the same as the typical ISA bus VGA card in a 386 PC (though much better for 2D games that were optimized for it). I love the way Amiga fans complain about their cheap low-end A1200s not having the power of a high-end 486. Did they think Commodore had some magic ingredient that allowed them to make similar hardware perform way better than the price indicated? Perhaps in 1985 they did by being more innovative, but not in 1994. In the 90's PC chip makers were constantly pushing their designs to the limit of what could be achieved, taking full advantage of Moore's law to jam more transistors into faster and larger chips, with huge economy of scale making them cheaper. Performance skyrocketed as prices dropped, so by 1995 when Windows 95 came out there were systems powerful enough to run it. But not if you had a 2 year old 386DX or SX. I had a 386DX-40 with 4MB of RAM and it ran Windows 95 quite sluggishly, as expected. I certainly did not expect 486DX-33 performance from it. Similarly I didn't expect A4000-40 performance from my A1200 - even after putting a 50MHz Blizzard 1230-IV in it (the combination still costing significantly less than than an A4000 or name-brand 486 system). This reminds me about a rich farmer's wife who came into my shop looking for a 'multimedia' computer to help with their kid's education. I gave her three brochures with the specs of a low-end (386SX), mid-range (386DX), and high-end (486) system. I gently suggested the 386DX or 486, but she decided to go for the SX because it was cheaper. A few years later she asked me to do some maintenance on it, and complained about it not having the performance she was expecting. I pointed out to her that she bought the low-end model which I did not recommend. Yes, sometimes the sales person is right, and you do get what you pay for!

This ^

All day, every day.

Last edited by agami on 18-Oct-2022 at 01:03 AM.

_________________
All the way, with 68k

@Hammer

Quote:

Can't you see the SKU gap between A1200(020@14Mhz) and A4000/040 in Q4 1992?

I remember this baffling me when I was in Maxwells Computer Centre almost 30 years ago to the day, as I was gladly handing over hard-saved money from a minimum wage post-uni job for the machine that would go on to change everything for me. While waiting for Lee to get everything together, I can vividly recall how much I coveted the A4000 040, and how it would've been nice to have a machine in between the A1200HD and the A4000, the latter being more than 3x the price of the former.
In my mind I started thinking about future upgrades that would bridge the gap, but there was still a missed opportunity in not having that mid-tier SKU, or add-ons that could immediately elevate the A1200 to the mid-tier position.

Interestingly, Apple did the same during the Steve Jobs comeback tour in the late '90s as they had the new and simplified SKU approach, with the all-in-one, colourful and popular iMac at one end, and then a big jump to the short-lived PowerMac G3 quickly replaced by the PowerMac G4.
They did have a misguided attempt at creating a mid-tier machine in 2000, with the PowerMac G4 Cube.


Quote:

It's relevant for gaming PCs that can run Doom (Q3 1993).

Doom didn't come out until December 1993, so for all intents and purposes in terms of influencing PC purchase decisions, Doom is a 1994 game.

I agree that in late 1993, the A1200 (with or without accelerator) and the newly released CD32, were not looking so good to an objective, first-time personal computer buyer.
If I hadn't got the A1200 a year earlier, and didn't already have a decent library of Amiga game titles and productivity software, in all honesty I can't say that I would not seriously consider a 386 machine at that time. Certainly, if I stuck with the A500 until mid-to-late 1994, which some friends did, then anything AGA would not even be a factor unless it was at stupendously reduced "clearance" prices.

But one thing I can tell you for certain. If I had stuck it out with the A500 and then joined the Doom gaming generation on the 486 like a couple of my friends did, I would never have developed any sentimentality or passion for the PC platform as I did for the Amiga.
The Amiga was more than a collection of silicon, copper, and plastic. It was a dream machine. Commodore's poor business management can be blamed for the delays in AGA, the SKU strategy and pricing, but the software and hardware engineers almost as an act of rebellion, made sure that even at the low A1200 end it was still the best little multi-media/multi-tasking personal computer they were proud to produce.

Last edited by agami on 18-Oct-2022 at 02:04 AM.

_________________
All the way, with 68k

@Leo

Quote:

Leo wrote: Quote: With a fast CPU and Fast RAM, AGA was okay for Doom and Wing Commander AGA. Any dumb framebuffer (that's what AGA is with Doom) is okay with a fast CPU + RAM.

1987 IBM VGA is a slow dumb framebuffer and throwing Athlon XP 2200+ at it wouldn't fix IBM VGA's roll-up slide show.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)

@Kronos

Quote:

Kronos wrote: With a fast CPU and Fas RAM AGA was (is) little more then an overcomplicated framebuffer. Thats how it was used in those games and and pretty much all the 060/PPC demos. Thats something even a basic 16Bit VGA could do better in 93 and hence AGA was to little to late.

Not with IBM VGA. It must be a fast 16-bit VGA clone.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)