Was anyone else disappointed with the A1200?

[Gorf]

Originally Posted by babsimov

On the contrary, an entry-level performance of a 68060 75 MHz at Christmas 95 seems very good to me, especially in the price range of the 1200 with hard drive. Especially since Ed Hepler also said that there would be a DSP in addition.

Not in addition but instead of a FPU it would have had some DSP-like or better MMX-like features in the CPU...

So no FPU at a time 3D games on PC start making heavy use of it ...

naa - the Hombre (as planed) would have been a complete disaster

Only with a dual CPU setup, where one PA-RISC is responsible for the gfx and an other one for as normal CPU might have has a fighting chance and only if it would have offered some backwards compatibility.
And with all that it would not have been cheap

Playstation would have just rolled over it, like it did with any other console

[sokolovic]

Originally Posted by Gilbert

Was anyone else disappointed with the A1200?

Most Amiga users and magazines seemed to be very happy with the A1200 when it came out. I wasn't at all, and a look at the first games pretty much ended my association with Amiga gaming. I just saw the same games with more colours and a bit smoother. There was no wow factor. After that I stuck with the Amiga 500 (with half meg memory expansion) and my Super Famicom (Jap SNES).

Here's what Commodore got wrong in my opinion

1. Too much focus on creating higher-res screen modes with more colours (and also making the blitter work in these different screen modes) and not enough on enhancing gaming(8 or maybe 16 sprites when the comparitively old Megadrive and SNES could manage 64 and 128 respectively). It's a bit like the original Amiga - yes it can display 4096 colours on screen, but the majority of the games for the system were 16 colours (Albeit some had added some Copper magic) and most didn't even run at 50/60 fps. That was fine back in 1985 but 7 years(!) later you expect a significant upgrade.

2. There was a mild improvement to dual playfield mode. Great!... when the SNES had 5(?) playfields and could scale and rotate whole screens. Commodore seemed to have no sense they were competing here....

2. Sound chip needed 6 channels to get a decent track playing with sound effects. Again SNES and Megadrive have 6 channels each. Using the same sound chip from 1985 was ridiculous!

3. Like the original Amiga, if you wanted to get a good number of objects on screen with a lot of colours and scrolling, you had to spend ages using hardware tricks or specific techniques. Time = money and developers aren't going to want to spend 2 years making an arcade quality game on the A1200 when simpler systems exist....

I do have a CD32 now, but it's not very impressive from a technical point of view, even the mighty Banshee is bettered on both the SNES and Megadrive. The reason I like it is because it offers something a bit different and it's an Amiga It's fairly obvious it had no hope of competing long term. I just find it hard to see what Commodore was thinking with the AGA architecture??

Thanks for asking.

No.

[hammer]

Originally Posted by Mr-Z

Since my A1200 has s PiStorm32+CM4 I'm not disappointed anymore

ARM CPU family started from Commodore's failure to update 65xx CPU family into 16 bit and 32 bit era.

ARMv4T (with MMU, 120Mhz+) displaced Motorola 68000 based DragonBall (33 to 66 MHz) in the smart handheld market.

[hammer]

Originally Posted by Gorf

maybe ... at least for some markets like Eastern Europe

I still doubt the the UK perspective somewhat ... Pleasance was never a big-box-Amiga-guy, he was the games-bundle-guy. He was more or less selling the Amiga as a console. His enthusiasm for the Hombre project underlines that.
But I think he vastly underestimated the computer business growth in the UK ... it was huge like in any other western country - not only for businesses but also for home use. But PCs and Macs got all that business, because of the lack of a mid-range Amiga.

The situation in Germany was clearly different and Commodore Germany was quite right in saying, they couldn't sell anything without a hd anymore.

I highly doubt that race to the bottom was winnable and the aftermarket showed that: Escom managed to sell all produced A4000, but left a huge stock of A1200 ... even Microsystems Draco did sell and kept the company alive during the 90s

For mass production, the Amiga doesn't have Apple's customer base who can spend 1.2 million units in PowerMacs by early Jan 1995. Amiga NG and big box Amiga fans doesn't recognise this fact.

Software sells hardware and Apple has stronger professional software library and Doom in 1994.

The Amiga is closer to Steam Deck's price range, games and tinker. Steam Deck passed 3 million install base which started from 2022. Steam Deck's unit sale pace is similar to A500's golden era.

UK's AGA sales pace for A1200/CD32 in 1993 almost rivals A500's golden era, but this not replicated Germany, second largest Amiga market.

Commodore Germany sold 95,000 A1200 and 25,000 CD32's

Commodore Canada allocated 65,000 CD32 board units for Amitech's big box A2200 clone. For similar money, A2200/030 @ 40 MHz with Akiko C2P is superior when compared to A4000/030 @ 25 MHz.

Commodore Germany's A4000 unit sales are in the 11,000 range.

North American market needed high clock speed 32-bit "big box" Amigas (e.g. 40 MHz) not frozen in 25 MHz. Commodore Canada recognised this "high clock" speed and packed pixels market fact.

[Bruce Abbott]

Originally Posted by hammer

Hint:
1. Gayle didn't exist during 1991's A1000 Jr. Look in https://bigbookofamigahardware.com/b...15&fileId=8468

So? The A2200 (AKA "A1000jr") had IDE in it, implemented much like the A4000 with two PAL (programmable Array Logic) chips and a number of TTL buffers and gates. The A4000 was based on this design.

For the A600 this stuff was simply pushed into Gayle, as was the other stuff in Fat Gary. The only 'new' thing was PCMCIA, and that wasn't hard to do.

2. A1200's Gayle is dependent on A300's Gayle. A1200's AA-Gayle reached "R5".

So? The initial production A600 had Gayle Rev 1.0, with one small bug involving I/O decoding fixed with a small PAL chip. PCMCIA was fine in this first revision, which went into production in December 1991. The initial specification was written up in July 1991 (5 months earlier). That was a normal turnaround time for Commodore - when things went right.

A1200 can't exist without A600's Gary design changes. A1200 is NOT A500p with AGA+Ramsey+68EC020+four TTL bridge chips+Fat Gary+CPU local bus edge connector.

So? If the A1200 had been designed in parallel with or instead of the A300, it could have been released in late 1991 if the AGA chipset was ready in time. But it wasn't. That's also why the A1000+ was turned into the A2200. Do you think they wouldn't have put the AGA chipset in it if they could have? Of course they would have.

Furthermore a new OS had to be written to support the AGA chipset. This was difficult to do without a working chipset. OS programmers were still struggling with AGA graphics bugs as late as May 1992.

But Mehdi Ali insisted that they forge with ahead with the A1200 even though the AGA chipset was buggy. The PCB included space for workaround chips if they couldn't get the bugs out by the time it was ready for production.

Some bugs were worked around in the OS. This is one reason Commodore initially didn't want to publish a detailed AGA hardware spec., as it would prevent them from fixing bugs that hardware-banging code relied on.

BTW note the copyright date on that A2200 pilot production PCB, 1992. No PCMCIA, no Gayle IDE, same old ECS chipset, yet it wasn't out in 1991 like you insist the A1000+ would have been.

[hammer]

Originally Posted by Promilus

@babsimov
A1200 class home computer based on Hombre doesn't have external PA/50
125MHz is a class for PA-7150 which has humongous cache compared to 1K L1D$ and 2K L1I$ in Hombre. Basically all HP PA-RISC processors were known for having abnormal cache size. So once L1 was integrated and relatively small they did include (again) external L2 with a size of few MB! And Hitachi also made such CPU
https://websrv.cecs.uci.edu/~papers/...LES/071104.pdf
PA/50 looks like trimmed down design with no L2 and less execution pipelines.
Reaches supposedly ~100MIPS which is level of 75MHz 060 afaik. Not bad, not great either. Not in XMAS '95 ... I'd say at time it would be a rather big struggle to gain some market share with 1. CPU performance was not so great 2. no backward compatibility, 3. In 96 PC did get a first VooDoo ... also there's quake (but afaik initially in only software rendering so as long as it was actually "tower/desktop" system with external CPU it just might work out. But... trying to sell "multimedia" or "cd console" without one and with fairly crippled PA-7150 inside - sorry Ed, but that was just no go...

Amiga Hombre CD64 targeted CD32's price range, NOT Pentium 100 - 166 + Voodoo 3D price range PC! Amiga Hombre chipset has $40 price target.

Cost constrainted 3DO M2 had IBM PowerPC 602 (1 million transistors) with small cache.

http://cpudb.stanford.edu/processors/307
For 3DO M2's CPU, IBM PowerPC 602 has 66 Mhz, 1 million transistors, 1 KB L1 instruction cache, 4 KB L1 data cache.

Prove if you can deliver $40 68EC060 @ 75 Mhz under 1 million transistor budget!

Prove that the Amiga has the customer base who can spend like Apple's 1.2 million PowerMacs in 1 year pace!

You're in lala land to think the Amiga is like Apple's Macintosh customer base.

Amiga's entry vector is on a low-mid price that is above the game console's price and "bang per buck". Amiga's mass production audience was never about "Phase 5" or PowerAmiga NG overpriced hardware.

Try again.

This is one of my gaming PC's parts list https://pcpartpicker.com/list/FW7DQP with $3315.86 USD spending in 2023 which is $1572.04 USD in 1993 i.e. about A4000/030 spending levels. My other RTX 4090/Ryzen 9 7950/ROG Crosshair X670E hero build is in A4000/040 spending levels.

My Dad spends about $1000 USD on my family's 386DX-33/ET4000 PC clone in 1992! My family has two desktop computers.

The A500's $699 USD in 1987 is about $888.68 USD in 1993. A1200 had $599 USD launch price in 1992.



Reference
https://www.minneapolisfed.org/about...ion-calculator

[hammer]

Originally Posted by Bruce Abbott

So? The A2200 (AKA "A1000jr") had IDE in it, implemented much like the A4000 with two PAL (programmable Array Logic) chips and a number of TTL buffers and gates. The A4000 was based on this design.

My point, Gayle and AA-Gayle was designed in 1991.

My AA-Gayle on my A1200 rev1d4 still has bugs e.g. PCMCIA reset problems which needed 3rd party fix.

https://forum.amiga.org/index.php?topic=62394.0
A1200 with PCMCIA lockups in 2012.

I also encountered A1200 PCMCIA lockups.

The same PCMCIA cards works on DELL Inspiron 5150 (Pentium 4 Northwood 3.4 Ghz mobile). DELL Inspiron 510m (Pentium M 1.6 Ghz) and IBM T20 (Pentium III 700 Mhz) laptops.

Originally Posted by Bruce Abbott

For the A600 this stuff was simply pushed into Gayle, as was the other stuff in Fat Gary.

Fact: Gayle and AA-Gayle was being designed in 1991.

A600 was released in March 1992.

Commodore's IDE implementation is at 1986 era PIO-0.

Originally Posted by Bruce Abbott

The only 'new' thing was PCMCIA, and that wasn't hard to do.

Oh really? Why PCMCIA lockups on A1200? Why PCMCIA reset issue on A1200?

https://comp.sys.amiga.hardware.nark...e-and-ethernet
20 years ago, A1200 PCMCIA lockups.

You're in lala land. Easy you say.

Originally Posted by Bruce Abbott

So? The initial production A600 had Gayle Rev 1.0, with one small bug involving I/O decoding fixed with a small PAL chip. PCMCIA was fine in this first revision, which went into production in December 1991. The initial specification was written up in July 1991 (5 months earlier). That was a normal turnaround time for Commodore - when things went right.

That's a delay.

Originally Posted by Bruce Abbott

So? If the A1200 had been designed in parallel with or instead of the A300, it could have been released in late 1991 if the AGA chipset was ready in time. But it wasn't. That's also why the A1000+ was turned into the A2200. Do you think they wouldn't have put the AGA chipset in it if they could have? Of course they would have.

There's "more than 6 months delay" due to ECS jobs which is management's fault.

From your own post, "went into production in December 1991". This will miss Q4 Xmas 1991 sales.

A600 was released in March 1992. This will miss Q4 Xmas 1991 sales.

A1200 has added complication with 32-bit Budgie.

Originally Posted by Bruce Abbott

Furthermore a new OS had to be written to support the AGA chipset. This was difficult to do without a working chipset. OS programmers were still struggling with AGA graphics bugs as late as May 1992.

Prove it. I posted my sources. I expected you to do the same.

1. Most AGA games have "kick-the-OS" and all runs in 15 kHz mode.

2. Commodore management wasted "more than 6 months" on ECS jobs.

3. Most of AGA's double scan resolution modes already exist with ECS, the difference is the higher bit-planes i.e. 2 bit-planes vs 8 bit-planes. Double scan resolution modes are useless for 320x200 256 color games.

If I have my 1084S monitor with A1200 combo, 31 kHz AGA modes are useless to me.

Commodore could have released AA500+ with just 15 kHz AGA. 31 kHz capable AGA in 1992's A1200.

With 31 kHz AGA, Commodore attempted to rival entry level SVGA on paper spec e.g. entry level SVGA's 640x480p with 256 colors.

---
Apple M2 has hardware raytracing and it has major issues, hence it was disabled. M3 is Apple's second attempt with hardware raytracing.

Originally Posted by Bruce Abbott

But Mehdi Ali insisted that they forge with ahead with the A1200 even though the AGA chipset was buggy.

Be specific. Which part of AGA has "show stopper" buggy?

Commodore management "wasted more than 6 months" on ECS related Amiga models i.e. A2200, A2400, A3200 and A3400.
https://bigbookofamigahardware.com/b...t.aspx?id=2015

Originally Posted by Bruce Abbott

The PCB included space for workaround chips if they couldn't get the bugs out by the time it was ready for production.

Be specific.

A1200's "AA-Gayle" has an AA part of its name.

Originally Posted by Bruce Abbott

Some bugs were worked around in the OS. This is one reason Commodore initially didn't want to publish a detailed AGA hardware spec., as it would prevent them from fixing bugs that hardware-banging code relied on.

Be specific.

Originally Posted by Bruce Abbott

BTW note the copyright date on that A2200 pilot production PCB,

Hint: AA3000+ revision 0 was in a bootable state in Feb 1991. Revision 1 reached SMD and operational DSP audio subsystem state before it was frozen and switched to "A1000Jr" ECS related Amiga models.

https://www.landley.net/history/mirr...re/haynie.html
There was a mad panic to drop in the frozen AA3000+ AGA chipset on ECS A3400 which turned into A4000.

[hammer]

Originally Posted by babsimov

I had already looked at these links.
As I understand the documents on Hombre, the PA-RISC included in the chipset did not work at 125 MHz in any case.
The frequency of 50 MHz is the one mentioned. The PA/50 probably serving as the basis for the PA-RISC included in Hombre.
They explain that they took things out of PA-RISC to include the features they needed for Hombre. Which may explain why the caches are smaller than the normal PA/50.
As you said, the desktop version (like A4000) would have also included a complete PA/50.
I had looked a little into the comparative performance of RISC processors at the time.
In the document that I found, PA-RISC was the most efficient at equal frequency.
There was the Pentium compared and a PA/50 at 50 MHz was equivalent to a Pentium 60/75 MHz. Of course if I did my calculations correctly.
The link to the document (it is a document in French)


https://hal.univ-smb.fr/INRIA/inria-00073501v1

The comparative table is at the end of the document (page 153). It is not the 7150 generation which is compared, but the following one, it is the Pentium Pro which is indicated. I extrapolated from that for the pentium, maybe I was wrong.

Commodore was aiming for about $40 Amiga Hombre chipset i.e. effectively high clock speed single thread PA-RISC driven 3D accelerator via instruction set extensions in a similar compute power and 1 million transistors budget as Sony's PS1 (33 MIPS CPU + 66 MIPS geometry, 1 million transistors).

Remember, A1000 was about delivering mid 1985 era workstation graphics for a lower price. IBM PGC already has 256 colors 640x480 with 4096 color palette in 1984. NEC PC-98 (uPD7220) has 640×400p with 16 colors display and 4096 color palette from 1982.

[Thorham]

Originally Posted by hammer

I also encountered A1200 PCMCIA lockups.

I don't with my network card.

[babsimov]

Originally Posted by Gorf

Not in addition but instead of a FPU it would have had some DSP-like or better MMX-like features in the CPU...

Yes, I meant that the DSP was complementary to the floating point calculation. Hombre had no FPU. On the other hand it had MMX type instructions. On this point he was also ahead of Intel who would only offer MMX 2 or 3 years later if I remember correctly.
Besides, I have the impression that HP will recover Commodore's work on this point and that it will partly become the MAX instructions for the 7100LC.
https://en.wikipedia.org/wiki/PA-7100LC.

So no FPU at a time 3D games on PC start making heavy use of it ...

naa - the Hombre (as planed) would have been a complete disaster

If I remember correctly the PS1 didn't have an FPU and yet that didn't stop it from being a real commercial success. Hombre targeted this market for the CD64.
With hindsight, on the contrary I think that Hombre was a step in the right direction.

Only with a dual CPU setup, where one PA-RISC is responsible for the gfx and an other one for as normal CPU might have has a fighting chance and only if it would have offered some backwards compatibility.
And with all that it would not have been cheap

For Amiga compatibility, there would have been the possibility of the SOC AGA+EC020 which Ed Hepler had spoken about.
I did the math a bit and an A1200 type machine with this SOC and Hombre, hard drive and 4/8MB of RAM was around 1000 dollars.
It seems pretty good to me for hardware 3D integrated into a home computer. I would have bought without hesitation, especially with an AmigaOS port. An A4000 type version with an additional PA/50 would also have been interesting.
On the other hand, ideally this Hombre range should have arrived at the maximum by September 1994. Exactly as the AGA should have been on the market in the Amiga 3000 of 1990 or at the maximum in 1991.

Playstation would have just rolled over it, like it did with any other console

On the console market it is likely that the PS1 has indeed do that.

But Hombre was also for the home computer market and at that time the Amiga had a good image there.
If Commodore had released the AGA earlier he would have kept it and Hombre would have put the Amiga back on track with 24 bit 3D hardware.

[babsimov]

Originally Posted by hammer

Commodore was aiming for about $40 Amiga Hombre chipset i.e. effectively high clock speed single thread PA-RISC driven 3D accelerator via instruction set extensions in a similar compute power and 1 million transistors budget as Sony's PS1 (33 MIPS CPU + 66 MIPS geometry, 1 million transistors).

Remember, A1000 was about delivering mid 1985 era workstation graphics for a lower price. IBM PGC already has 256 colors 640x480 with 4096 color palette in 1984. NEC PC-98 (uPD7220) has 640×400p with 16 colors display and 4096 color palette from 1982.

Yes, I had read that the target price for Hombre was $40.
And, i agree, for the CPU and chipset together it's not expensive at all.

Likewise I agree that the Amiga 1000 attempted to bring workstation graphics to the masses. And that's why, in hindsight, I find that Hombre was a step in the right direction. Lew Eggebretch said that Commodore wanted to bring the 3D capabilities of workstations to the masses. During the interview we didn't know about the Hombre project, but that's what he was talking about, no doubt.

[Gorf]

Originally Posted by hammer

Be specific. Which part of AGA has "show stopper" buggy?

I also would really very much like to know, what exactly did not work and if this was a real show stopper.

So far I only read sources claiming it was ready or having only minor bugs.

If such bugs could have been squashed by some external circuitry, this should have been implemented in an AA3000+ in 1991, while putting the necessary effort in R&D to have a fixed version in early 92.

As for the 31KHz modes:
Amber was already there and 2 of them work fine with AGA 24bit
This could have been used in high-end and mid-range models as a pure line-doubler (making the field-RAM optional).
A line-doubler makes more sense anyways for dblPAL and dblNTSC, as it puts less stress on the ChipRAM.

The OS not being ready ... this did not stop the A3000 to ship, when 2.0 was not yet complete: early models came with OS 1.4

[Promilus]

Originally Posted by babsimov

Yes, I meant that the DSP was complementary to the floating point calculation. Hombre had no FPU. On the other hand it had MMX type instructions. On this point he was also ahead of Intel who would only offer MMX 2 or 3 years later if I remember correctly.
Besides, I have the impression that HP will recover Commodore's work on this point and that it will partly become the MAX instructions for the 7100LC.
https://en.wikipedia.org/wiki/PA-7100LC.

The only instructions I came across were all about packed pixels and not generic calculations.

If I remember correctly the PS1 didn't have an FPU and yet that didn't stop it from being a real commercial success. Hombre targeted this market for the CD64.
With hindsight, on the contrary I think that Hombre was a step in the right direction.

Sure it didn't and that was the reason of infamous texture warping and not only - there were other artifacts in 3D on PS1. It was not such big issue when playing relatively low res on TV because it did look good anyway. With computer, 640x480 resolution and crystal clear pixels things like that look absolutely horrible. Now without external PA/50 the most Hombre might've accomplish was to reach the same level as PS1 which was released earlier and with great support of Sony to developers. I'm sorry but it just would not work.

For Amiga compatibility, there would have been the possibility of the SOC AGA+EC020 which Ed Hepler had spoken about.

Sure, but he didn't get even that far to make POC (even Hombre was just mock-up at that point) and with "classic Amiga on PCI" you risk using it for different computers to run emulated Amiga environment on mixed PC/Amiga hardware the same way as PC software was run on Amiga with PC XT/AT hardware emulators.

I did the math a bit and an A1200 type machine with this SOC and Hombre, hard drive and 4/8MB of RAM was around 1000 dollars.

[/quote]
Without FPU it was a quick loss to all those 486DX2/4 machines

It seems pretty good to me for hardware 3D integrated into a home computer

well that "hardware 3D" is rather "software assisted 3d" ... as in case of "CD64" and "A1200 class" there's only ONE CPU and it is responsible for both "hardware 3d" and actual execution of application binary. So basically same thing as running quake in software mode on pentium. It was ok-ish but Quake did run a lot better (despite looking, well, different, I could even go as far as bland) with actual offloading the first portion of 3d pipeline to graphic card (every few years more and more elements were done by GPU like T&L, then actual shading, then modifications of object geometry etc.)

On the other hand, ideally this Hombre range should have arrived at the maximum by September 1994. Exactly as the AGA should have been on the market in the Amiga 3000 of 1990 or at the maximum in 1991.

And it was not possible, from the very same reason... the first silicon die is usually buggy as hell, something relatively slow simulation cannot catch. That's why both performance number of Ed Hepler's simulations and target date of the working silicon are overly optimistic.

But Hombre was also for the home computer market and at that time the Amiga had a good image there.

With no software whatsoever and additional money to pay for compatibility with older system and run it on native older systems speed ... I'd say that's not something really interesting.

If Commodore had released the AGA earlier he would have kept it and Hombre would have put the Amiga back on track with 24 bit 3D hardware.

If commodore did invest in own plant they might've done many things a lot earlier. If commodore did adopt industry standards sooner and cooperated with 3rd pary amiga hw manufacturers it might've created RTG a lot earlier and jump on the modular Acutiator instead of integrated Hombre.
If commodore actually did cooperate with developers to make some launch titles for CDTV and CD32 those machines might see a much better reception.

That's a lot of "if" in regards to the company we know was incapable of doing things this way.

[hammer]

Originally Posted by babsimov

Yes, I meant that the DSP was complementary to the floating point calculation. Hombre had no FPU. On the other hand it had MMX type instructions. On this point he was also ahead of Intel who would only offer MMX 2 or 3 years later if I remember correctly.

Intel has MMX-like extensions for the i860. i860 is good at floating point and 64-bit SIMD integers.

i860's 64-bit SIMD influenced Pentium's MMX.

Originally Posted by babsimov

Besides, I have the impression that HP will recover Commodore's work on this point and that it will partly become the MAX instructions for the 7100LC.
https://en.wikipedia.org/wiki/PA-7100LC.

MAX SIMD is a good influence for Commodore's instruction set customization.

Originally Posted by babsimov

If I remember correctly the PS1 didn't have an FPU and yet that didn't stop it from being a real commercial success. Hombre targeted this market for the CD64.
With hindsight, on the contrary I think that Hombre was a step in the right direction.

PS1 is a fixed point integer.

Originally Posted by babsimov

For Amiga compatibility, there would have been the possibility of the SOC AGA+EC020 which Ed Hepler had spoken about.

That's pretty much what we have with A1200 and "cheap RISC" RPi 3A+/3B/4B/CM4.

There are Amiga AGA clones via FPGA in development that would support PiStorm32.

Originally Posted by babsimov

I did the math a bit and an A1200 type machine with this SOC and Hombre, hard drive and 4/8MB of RAM was around 1000 dollars.

3MB to 4MB is the minimum.

"CD32 SoC" is needed for 68K/AGA/Akiko backward compatibility.

$299 retail "Amiga Hombre CD64's dual chips" add DSP3210 and "CD32 SoC".

Amiga Hombre's A1200 replacement would have the lessons from CD32's cost reduction.

Commodore Canada/Amitech's A2200 clone used the CD32 board as the modular "Amiga card" for the larger "Agent 88" board, hence leveraging mass production for big box Amigas.

A1200's cost reduction direction was advanced by CD32.

Originally Posted by babsimov

It seems pretty good to me for hardware 3D integrated into a home computer. I would have bought without hesitation, especially with an AmigaOS port. An A4000 type version with an additional PA/50 would also have been interesting.
On the other hand, ideally this Hombre range should have arrived at the maximum by September 1994. Exactly as the AGA should have been on the market in the Amiga 3000 of 1990 or at the maximum in 1991.

On the console market it is likely that the PS1 has indeed do that.

ARM and RPi covered the tinker "cheap RISC" market. Commodore's $40 PA-RISC also covers the tinker "cheap RISC" market.

Sony's PlayStation doesn't cover tinker "cheap RISC" i.e. it's a close-walled garden.

NVIDIA's ShieldTV covers "cheap RISC" with games and tinker markets.

Sony's PlayStation hasn't eliminated RPi and ShieldTV.

[hammer]

Originally Posted by Promilus

The only instructions I came across were all about packed pixels and not generic calculations.

Why generic? Games should be the focus.

Originally Posted by Promilus

Sure it didn't and that was the reason of infamous texture warping and not only - there were other artifacts in 3D on PS1. It was not such big issue when playing relatively low res on TV because it did look good anyway. With computer, 640x480 resolution and crystal clear pixels things like that look absolutely horrible. Now without external PA/50 the most Hombre might've accomplish was to reach the same level as PS1 which was released earlier and with great support of Sony to developers. I'm sorry but it just would not work.

PS1's wobble 3D was on PC's Tomb Raider port. PC's Tomb Raider 3D wasn't stable like on PC's Quake.

Fixed point 3D has render issues.

Lew argued for DSP3210.

Originally Posted by Promilus

Sure, but he didn't get even that far to make POC (even Hombre was just mock-up at that point) and with "classic Amiga on PCI" you risk using it for different computers to run emulated Amiga environment on mixed PC/Amiga hardware the same way as PC software was run on Amiga with PC XT/AT hardware emulators.

Both PA-RISC and 68K CPUs are big-endian CPUs e.g. PRi with Amiga.

Commodore's direction with $40 Hombre PA-RISC is closer to the "cheap RISC" ARM mentality.

Originally Posted by Promilus

Without FPU it was a quick loss to all those 486DX2/4 machines

486's FPUs weren't fast enough for Quake. Is your argument being framed as a "big box" Amiga?

Originally Posted by Promilus

well that "hardware 3D" is rather "software assisted 3d" ... as in case of "CD64" and "A1200 class" there's only ONE CPU and it is responsible for both "hardware 3d" and actual execution of application binary.

PS1 and Saturn don't have Z-buffer 3D acceleration. Both consoles weren't as feature-rich when compared to N64.

Originally Posted by Promilus

So basically same thing as running quake in software mode on pentium. It was ok-ish but Quake did run a lot better (despite looking, well, different, I could even go as far as bland) with actual offloading the first portion of 3d pipeline to graphic card (every few years more and more elements were done by GPU like T&L, then actual shading, then modifications of object geometry etc.)

For Pentium, Quake's geometry was on FPU and pixel processing on integers.

Quake's workload type was applied to DX8's floating point vertex shaders (e.g. FP32) and integer pixel shaders (e.g. INT16).

Moving the 3D geometry processing load on asymmetric dual integer pipelines while processing pixels wouldn't be optimal for Pentium. Heavy integers are 68060's and Cyrix 6x86's advantage.

Originally Posted by Promilus

And it was not possible, from the very same reason... the first silicon die is usually buggy as hell, something relatively slow simulation cannot catch. That's why both performance number of Ed Hepler's simulations and target date of the working silicon are overly optimistic.

Ed Hepler's statements include PA-RISC licensing and adding custom 3D instruction extensions. It depends on the implementation.

Originally Posted by Promilus

With no software whatsoever and additional money to pay for compatibility with older system and run it on native older systems speed ... I'd say that's not something really interesting.

If commodore did invest in own plant they might've done many things a lot earlier. If commodore did adopt industry standards sooner and cooperated with 3rd pary amiga hw manufacturers it might've created RTG a lot earlier and jump on the modular Acutiator instead of integrated Hombre.

Acutiator has its own PA-RISC vision. There are reasons for maintaining "big-endian".

[babsimov]

Originally Posted by Promilus

The only instructions I came across were all about packed pixels and not generic calculations.

Maybe. I'm not technical enough to have analyzed that. But, I still feel like Hombre looked pretty good. Even Dave Haynie says it in relation to AAA.

Sure it didn't and that was the reason of infamous texture warping and not only - there were other artifacts in 3D on PS1. It was not such big issue when playing relatively low res on TV because it did look good anyway. With computer, 640x480 resolution and crystal clear pixels things like that look absolutely horrible. Now without external PA/50 the most Hombre might've accomplish was to reach the same level as PS1 which was released earlier and with great support of Sony to developers. I'm sorry but it just would not work.

So already, at the time, the 3D graphics of the PS1 were a slap in the face to everyone who saw them (including me).
Besides, I just wanted to have a computer with equivalent 3D capabilities at home.
If I understood correctly Hombre was at least equivalent. This is what Ed Helpler said in an interview.
Rereading the documents, it also indicates that some floating point instructions were included (which is not the case for the PS1).
For the price of a 1200 configuration with hard drive, all that would have been very good. Much less disappointing than the AGA was in any case.

Sure, but he didn't get even that far to make POC (even Hombre was just mock-up at that point) and with "classic Amiga on PCI" you risk using it for different computers to run emulated Amiga environment on mixed PC/Amiga hardware the same way as PC software was run on Amiga with PC XT/AT hardware emulators.

Yes, but here we are talking about a Commodore which would not have wasted its time with the AAA. Would have marketed the ECS in 87/88 and the AGA (or a little better) in 1990. In short, a Commodore which would have been well managed and properly funded research. In this case, Hombre could have been ready for 1994 from my point of view.

Without FPU it was a quick loss to all those 486DX2/4 machines

It seems somes are included.

well that "hardware 3D" is rather "software assisted 3d" ... as in case of "CD64" and "A1200 class" there's only ONE CPU and it is responsible for both "hardware 3d" and actual execution of application binary. So basically same thing as running quake in software mode on pentium. It was ok-ish but Quake did run a lot better (despite looking, well, different, I could even go as far as bland) with actual offloading the first portion of 3d pipeline to graphic card (every few years more and more elements were done by GPU like T&L, then actual shading, then modifications of object geometry etc.)

I looked on YouTube and it still looks like the PS1 can run Quake (it's even Quake II here). So I don't see why Hombre couldn't have done the same thing or a little better. In any case, it is much better than what the AGA allowed by default.


[ Show youtube player ]

And it was not possible, from the very same reason... the first silicon die is usually buggy as hell, something relatively slow simulation cannot catch. That's why both performance number of Ed Hepler's simulations and target date of the working silicon are overly optimistic.

With no software whatsoever and additional money to pay for compatibility with older system and run it on native older systems speed ... I'd say that's not something really interesting.

If commodore did invest in own plant they might've done many things a lot earlier. If commodore did adopt industry standards sooner and cooperated with 3rd pary amiga hw manufacturers it might've created RTG a lot earlier and jump on the modular Acutiator instead of integrated Hombre.
If commodore actually did cooperate with developers to make some launch titles for CDTV and CD32 those machines might see a much better reception.

That's a lot of "if" in regards to the company we know was incapable of doing things this way.

Of course a lot of "if". But it is precisely the point of asking what could have happened if things had been better managed.
That's not to say it wouldn't have been difficult for Commodore to survive the 90s, but I think it would be interesting if one day someone tried to recreate Hombre in FPGA, that way we would really have a good idea of ??what it could have been compared to what was being done at the time.
Of course Ed Hepler himself said that 3D PC cards would do better. But, these cards would have cost the price of a CD64, not to mention the price of the PC needed to use it.
In terms of performance/price, a 1200 Hombre would have done it for me. But, with competent Commodore management, of course.

[babsimov]

Originally Posted by hammer

Intel has MMX-like extensions for the i860. i860 is good at floating point and 64-bit SIMD integers.

i860's 64-bit SIMD influenced Pentium's MMX.

Thanx for the information.

MAX SIMD is a good influence for Commodore's instruction set customization.

I'm not sure MAX SIMD exist when Hombre was in design phase.

PS1 is a fixed point integer.

That's what I thought

That's pretty much what we have with A1200 and "cheap RISC" RPi 3A+/3B/4B/CM4.

There are Amiga AGA clones via FPGA in development that would support PiStorm32.


3MB to 4MB is the minimum.

"CD32 SoC" is needed for 68K/AGA/Akiko backward compatibility.

$299 retail "Amiga Hombre CD64's dual chips" add DSP3210 and "CD32 SoC".

Amiga Hombre's A1200 replacement would have the lessons from CD32's cost reduction.

Commodore Canada/Amitech's A2200 clone used the CD32 board as the modular "Amiga card" for the larger "Agent 88" board, hence leveraging mass production for big box Amigas.

A1200's cost reduction direction was advanced by CD32.

ARM and RPi covered the tinker "cheap RISC" market. Commodore's $40 PA-RISC also covers the tinker "cheap RISC" market.

Sony's PlayStation doesn't cover tinker "cheap RISC" i.e. it's a close-walled garden.

NVIDIA's ShieldTV covers "cheap RISC" with games and tinker markets.

Sony's PlayStation hasn't eliminated RPi and ShieldTV.

I agree, to me Hombre was the right decision.

I say that, although at the time I did not understand the interest in Hombre because it was not compatible with the Amiga. I preferred AAA. But since then I have read a lot of things and the SOC AGA+EC020 would have solved this compatibility problem. That's why it's been a while since I changed my mind about Hombre and why it seemed the right direction.

[hammer]

Originally Posted by babsimov

I looked on YouTube and it still looks like the PS1 can run Quake (it's even Quake II here). So I don't see why Hombre couldn't have done the same thing or a little better. In any case, it is much better than what the AGA allowed by default.

[ Show youtube player ]

Pocket Quake project has rewritten substantial portions of the source code as fixed-point integers.

[hammer]

Originally Posted by babsimov

I'm not sure MAX SIMD exist when Hombre was in design phase.

Rendition Verite V1000 has an MIPS-like CPU with 3D extensions.

https://vintage3d.org/verite1.php#sthash.nVkUWaAD.dpbs
"Canopus Total3D fall short of average framerates of S3 Virge /GX2 by 7%."

"V1000L-P to be 46% of Voodoo performance overall, were OpenGL Quake games are significantly lower at 33% however VQuakes are running at 75%-95% of Voodoo".

Rendition Verite V1000 wasn't optimized for OpenGL, hence the large difference between Verite specfic VQuake vs other OpenGL games.

This is for V1000L-P which replaced the initial V1000-E.
----------------
https://fabiensanglard.net/vquake/
For the initial V1000-E chip,

From Walt Donovan (Algorithm Architect)

The v1000 was basically just a slow CPU (25MHz) that had a 1-clock 32*32 multiply (which took up a significant fraction of the chip!), a 1 clock approximate reciprocal instruction (so a 2 clock approximate integer divide), and the usual set of RISC instructions. Oh, and one more thing - a "bilinear load" instruction which effectively read a 2x2 block of linear memory and performed a bilinear filtering based on the u and v fraction passed in the instruction. There was a tiny cache, I believe it was just 4 pixels. So if you had overlapping 2x2's you'd get a reduction in memory bandwidth, but not if the 2x2's didn't overlap.

There was no hardware support for Z buffers. So the software that ran on the v1000 had to read Z, do the compare, then decide to write or not.
---

Rendition took the approach of being fully programmable, but didn't have any kind of clever pipelining or fast clocks. So, if it took 25 instructions to write a pixel, then you would get only 1 megapixel/sec. With fixed function hardware, you can pipeline the equivalent of these 25 instructions, and get 25 megapixel/sec. 3dfx's people came from SGI, and they took the approach that the right solution was to build a fixed function triangle engine in hardware, with a bunch of knobs that was a subset of OpenGL functionality. The designers of the V1000 came from a very different background, had no knowledge of OpenGL, and decided that the right approach was to build a CPU.

Amiga Hombre's entry point is based on price.

[TuRRIcaNEd]

Originally Posted by hammer

Apple's Macintosh Display Card 8·24 GC for its Macintosh IIfx(1990), featuring AMD's 30 MHz Am29000 RISC processor, 64 KB static RAM cache, and 2 MB of VRAM. Am29000 accelerates Quickdraw for Macintosh IIfx. For Macintosh Quadra 840av and 660av, AT&T DSP3210 (RISC-DSP for 3D and multimedia) accelerates Quickdraw. The RISC threat is real.

And how much did that setup cost?

I skipped "Geforce 256" for GeForce 2 MX along with Athlon Tbird 1.133 Ghz.

Apropos of nothing, the second PC I built was set up around an OG (Slot A) Athlon 700 and the GeForce 256 - the VIA chipset was annoyingly problematic at times...

My comments are for the A1200's context i.e. leadup to AGA and 1992 and 1994 date range.

As I said before, it's reasonable to state that when AGA arrived it was too little, too late - however it's also reasonable to point out that it was a nifty, elegant and cost-effective bit of engineering given the constraints the team had to deal with.

During the late 1990s, X86 CPUs engaged in the Ghz race, and the PowerPC camp didn't keep up.

Not exactly - though I'm sure you know that... Intel tried to leverage the Pentium 4, only for AMD to eat their lunch by hiring a bunch of former DEC engineers and designing a superscalar processor that was effectively an x86 emulation layer hooked up to a turbocharged RISC back-end derived from the DEC Alpha. It's also worth pointing out that CBM were looking into the DEC Alpha to power the next-gen Amiga platform if I recall correctly.

I'm aware of Apple's late 1998 difficulties.

For the 1994 context, Apple came out swinging with PowerPC and the floating point was strong. The RISC threat is real.

But the PowerPC was an IBM technology that Apple just happened to use.

Intel released the Pentium Pro in 1995.

And in commodity PC terms, it was a sales flop in its original form. The P6 architecture was re-tooled later, but that's another story.

This issue is covered by https://eab.abime.net/showthread.php?t=91337

For the Amiga, 68060 @ 50 Mhz and Shapeshifter was a 1995 experience.

Doesn't matter - it's still but one more piece of evidence demonstrating that Apple hardware has always been overpriced, underpowered garbage. Tramiel Sr. spotted that back in 1977, that's why he told Jobs and Woz to take a hike in the first place!

They weren't cost-effective and not enough to dislodge incumbents with existing PCs text-based Lotus 123 and Word Prefect.

They were more capable in general, and with Gould defenestrating Rattigan, the resulting power vacuum meant that the concept was never properly marketed.

CBM's PET has had early success in the business markets.

With all due respect, barely.

Windows 95 allows backward compatibility with DOS games.

Far from universally...

(I was there; I remember...)

Nope,

Correct me if I'm wrong, but that graph only shows an uptick in PC compatible sales from 1984 onwards, with the significant uptick starting in 1987.

Furthermore, the author of that blog post doesn't seem to grok how to use Excel's chart functionality correctly. They're using a stacked area chart, when a stepped area or line chart would be more appropriate. The stacked chart makes it look like the Macintosh was outselling the PC compatibles, which is just laughable.

British Empire didn't exist in 1984.

What's that got to do with anything?

A1000s' release has near zero business applications. Normies has day jobs.

CBM was a mess in 1985 - no arguing with that.

1985 Macintosh has business apps i.e. MS Excel, MS Word, Aldus PageMaker, and 'etc'. Mac's QuarkXPress in 1987.

But the platform lacked sufficient RAM out-of-the-box and expandability to make that software work in a practical manner.

Read https://en.wikipedia.org/wiki/Clean_room_design

Don't need to - I'm well-aware of how Compaq went about reverse-engineering the BIOS. What I was getting at was that the legal state-of-the-art at the time was lagging behind the realities of the technical state-of-the-art. The case could have gone either way.

IBM had a Professional Graphics Controller in 1984 https://en.wikipedia.org/wiki/Profes...ics_Controller

Which cost just shy of $3000 in 1984 on top of the same amount (or more) for the base machine.

The Amiga didn't partition its graphics architecture.

VideoToaster says "What am I, chopped liver"?