Was anyone else disappointed with the A1200?

[Gorf]

Originally Posted by sandruzzo

@Gorf

Amiga need better HW capabilities first..Maybe Dsp would have been great on A4000, but I would have add a PPC, that would have been useful

A DSP would have made up for the non-development of a new Paula.
Good old Paula + a DSP would have solved the sound issue.
A 020 or 030 plus DSP is a powerful yet affordable combination.

[sandruzzo]

@Gorf

Believe me, it is not so easy. What made PC/VGA so succesfull was that is was easy to develop

[hammer]

Originally Posted by babsimov

Tastes and colors
That said, over time, I've come across a lot of people who preferred the ST design in the Amiga community, like me. And despite everything, the design of the case is not everything, since I preferred to take an Amiga for what was inside.

Yes, the Falcon is not free from flaws like the 1200.
But, in many ways the hardware inside is what should have been in the new generation of Amiga (chunky display, more sound channels (16 bit sound) and in addition a DSP). From my point of view, the Atari teams did their job better than those from Commodore on this generation (expect for the 030 on a 16 bit bus).

Falcon's 16-bit 030 was mocked by anti-Atari and 1 MB RAM baseline configuration is previous generation.

Fast 386DX/486SX-25/486SX-33 PC games have 3 to 4 MB requirements that is in line with 3DO ($699 USD)'s 3 MB RAM (2 MB system, 1 MB VRAM).

Falcon 4MB RAM configuration is 599 UKP.
Falcon 4MB RAM configuration is 699 UKP with 64 MB HDD.
Falcon 4MB RAM configuration is 999 UKP.

PC's 16-bit color pack pixels are on a higher clocked 32-bit VLB and SVGA chipsets.

For games machine, no brainer for 3DO's 2 million unit sales murdered the Atari Falcon's 13,000 to 14,000 units.

The games market is larger than the desktop music composition market.

Baseline A1200's 2 MB RAM (399 UKP with a healthy profit margin) is progress from the previous A500/A600 generation, but wasn't enough for games that are ported from 3DO and fast 386DX-33/386DX-40/486SX-25/486SX-33 PC games have 3 to 4 MB requirements.

Officially, Commodore thinks A1200 does not have game performance C2P and their solution is Akiko's C2P that was created within 1 day, hence locking out A1200 from chunky pixels PC game ports.

Commodore was focused on IDE mandate and PCMCIA modifications on A1000+ (turned into A1200) and didn't think about C2P despite AAA project having chunky pixels.

Wing Commander CD32 for A1200 was unofficially patched.

If A1200 had an extra 1MB Fast RAM, hardware assist C2P or optimized blitter assist CP2 as part of the official SDK, the outcome could be different.

Accelerated A1200 had commercial textured map 3D games from smaller game studios.

AGA marketing had 24-bit 16.7 million color palette and 18-bit lossy color compression display capability. In modern GPU marketing, delta color compression (DCC) is a virtue.

https://www.anandtech.com/show/10325...ition-review/8
NVIDIA's marketing delta color compression (DCC) for Pascal generation.

Direct3D GPUs have MPEG (lossy motion compression), lossy and lossless texture compression. Modern GPUs has AV1 and H265 motion compression.

Psygnosis wanted upgraded CD32 for a minimal marginal cost increase.
From Commodore the Inside Story- The Untold Tale of a Computer Giant by David John Pleasance.

About page 128

As a result of this, I was asked by Ian Hetherington (cofounder,
with Jonathan Ellis, of Psygnosis) to arrange a
meeting with Mehdi Ali at their studios in Liverpool.

(skip, read book for extra detail)

Ian explained to Mehdi that with a few seemingly quite modest design changes, the CD32 could have an
incredible boost in performance at very marginal additional cost. 

(skip, read book for extra detail)

Well, it went exactly as expected. Mehdi was rude and ignorant, and clearly had no idea what Ian was talking
about. But instead of just admitting that, he more or less turned on Ian, as though he ‘must be crazy telling us how
to design our computers!’

Mehdi Ahi is not Sony CEO's Sony CEO Norio Ohga and Ken Kutaragi (technical director for Playstation).

[ Show youtube player ]
Steve Ballmer's Developers, developers, developers...

Originally Posted by babsimov

Yes but Commodore never released it.

However, on another forum I had many discussions with the Falcon community about the DSP and also the comparison with the DSP3210.
DSP connoisseurs explained me that the choice of the DSP3210 for the Amiga was quite curious. It is a scientific DSP at its core (floating calculations). To be equivalent in performance (MIPS) to Motorola's DSP 56001, it had to operate at higher frequencies. They also explained to me that Motorola's DSP was already well known at the time and therefore it was quite easy to find already well documented code for this DSP. The DSP3210 seemed less common and therefore less easy to program at first glance.

That's propaganda from the Atari side. DSP32's FP32 is 3D game/workstation floating point format.

Atari fans should look at OpenGL's GLfloat which is 32-bit IEEE-754 floating-point value i.e IEEE FP32.

AT&T DSP32 array is targeting SGI's Intel's i860 array market.

http://www.bambi-amiga.co.uk/amigahistory/mikesinz.html
Commodore engineer Mike Sinz's view on DSP32

 It would also so really good speech and sound and math (talk about fast rendering times!)

Atari fans didn't read AT&T's marketing materials for DSP32 i.e. 3D and multimedia DSP accelerator.

IEEE FP32 is standard for PC's vertex OpenGL and Direct3D.

http://bitsavers.trailing-edge.com/c...tion_Guide.pdf
From page 19 of 140, AT&T marketed multiple DSP3210 as a 3D graphics workstation co-processor array. AA3000+ only has a single DSP3210.

From page 20 of 140, "DSP32C, 32-bit CMOS DSP (3-D Graphics Floating-Point Accelerator with Graphics Application Library)"

DSP32 was used as Quickdraw accelerator for Apple's Mac Quadra 840AV (68040 @ 40 Mhz, DSP32 @ 66Mhz) and Quadra 660AV (68040 @ 25Mhz, DSP32 @ 55 Mhz, $1399 USD). Quadra 840AV and 660AV was replaced by PowerMac line in 1994. Quadra AV has video in and video out i.e. Quadra AV and Adobe Premiere NLE combo directly targets Amiga's video market. Fast Amigas with Shapeshifter/Fusion/Emplant can run Adobe Premiere NLE.

Apple >>>>>>>>>>>>>>>> Atari.

[hammer]

Originally Posted by sandruzzo

Why are almost all mad for DSP? Numbers aren't actual performance

SNES has packed pixels Mode 7 and direct color Mode 7. Officially, Nintendo supported DSP and SuperFX add-ons.

"Multimedia DSP" is a cheap math power accelerator.

The reason for GPU is excellent math power per area and per math power per dollar.

http://www.bambi-amiga.co.uk/amigahi...ggebrecht.html
Lew Eggebrecht's official road map had DSP3210 for every Amiga models. Lew Eggebrecht wasn't in the hot seat from 1988 to 1992.

Amiga Hombre's PA-RISC @ 120 Mhz displacing 68EC020 @ 14Mhz mirrors the migration path for PalmOS 5's ARM925T (ARMv4T) @ 126 Mhz to 144 Mhz displacing Dragon Ball MC68VZ328 (68000) @ 33Mhz.

Both PA-RISC and ARM had pretty good code densities. ARM focused on RISC power for the masses.

Phase 5's PowerPC approach wasn't for the masses. It's ARM's turn for the Amiga.

[hammer]

Originally Posted by sandruzzo

@Gorf

Amiga need better HW capabilities first..Maybe Dsp would have been great on A4000, but I would have add a PPC, that would have been useful

Haage & Partner/Phase 5 PPC wasn't low cost for the masses. I re-activated and continued to use my Amigas with ARM-based (PiStorm-Emu68) accelerators instead of storing them after the COVID-19 lockdown.

[Bruce Abbott]

Originally Posted by Gorf

A DSP would have made up for the non-development of a new Paula.
Good old Paula + a DSP would have solved the sound issue.

What sound issue?

[Bruce Abbott]

Originally Posted by hammer

http://www.bambi-amiga.co.uk/amigahi...ggebrecht.html
Lew Eggebrecht's official road map had DSP3210 for every Amiga models. Lew Eggebrecht wasn't in the hot seat from 1988 to 1992.

But if he was in the 'hot seat' earlier, would he have put DSP in the A1200? Probably not.

From your link:-

Most of the games writers want to have 8 channels of sound. Is this going to happen at the low end?

"The current capability is four channels of 8-bit samples at 27 KHz and we forsee that most systems in the future will have CD capability. Most of the sound and music will come from this so it was not as important to put that technology in. Our long term strategy is to put the DSP in every system, obviously. That will be sound in and sound out and you can do pretty much whatever you like."

Does that mean that a AA+ machine will have a DSP?

"We can't make that decision right now - it's something we'll have to look at but in that time frame, even in the low end, every machine is likely to have a DSP. It's a cost thing - although the AT&T chip itself is only $20 to $30 or so. AT&T has a number of lower cost options, as well, that are designed more specifically to go on the motherboard. The problem with the present DSP design is that it has one serial channel and everything you attach to it has to be run through at that channel rate. I think they're looking at having four independent channels running at different clock rates, and with that kind of enhancement, DSP makes a lot of sense."

Not how he is only talking about sound here. No mention of using DSP for graphics.

With CD audio you have Hi-Fi quality music with another 4 channels for sound effects, plenty enough for games and no need for a DSP chip.

[Promilus]

@Gorf - yes, but such DSP has to be properly placed in chipset... just slapping one to 68k bus ain't gonna solve a lot of things. DSP embedded in chipset and obviously with more DMA channels and faster RAM - that's the way to go...

[sandruzzo]

@Promilus

Amen.

[hammer]

Originally Posted by Bruce Abbott

But if he was in the 'hot seat' earlier, would he have put DSP in the A1200? Probably not.

A1200's development state is too far in the product development cycle when Lew was hired for the hot seat.

Bill Sydnes was fired after the A600 debacle.

https://amigaworld.net/modules/newbb...order=0#253077

Dave Haynie: Lew Eggebrecht has nothing to do with the Amiga as you all know it. He was hired to work with Bill Sydnes back in 1991 or so, and took over when they fired Sydnes in 1993.

The only remaining tactic is to execute Commodore UK's accelerated Amiga bundle deals via expansion cards.

A1200's design changes must be completed about 7 to 9 months before the release date. Give an extra month or two for Commodore.

[hammer]

Originally Posted by Promilus

@Gorf - yes, but such DSP has to be properly placed in chipset... just slapping one to 68k bus ain't gonna solve a lot of things. DSP embedded in chipset and obviously with more DMA channels and faster RAM - that's the way to go...

DSP on 020/030 bus is just a math accelerator.

[ Show youtube player ]
For AA3000+, DSP32 works as intended. Mac DSP32 software was ported for the AA3000+.

The audio DSP focus didn't help Atari Falcon's bad 13,000 to 14,000 unit sales results.

[Promilus]

One of the most common misconception is that DSP is math co-processor. No, DSP is DSP with architecture designed to fit that actual job. And while DSP could've supplement 020 with some nice additions related to 3D graphics generation it was hardly anything which particularly suited DSP back then. That's why there was a need to develop SuperFX chip for SNES in the first place, existing DSP-1 designs were hardly suitable.

Actual example how to done things fairly efficient was PS1 - GTE is centered around vector math co-processor which basically translates to SIMD capability present basically in each and every processor out there (and yes, also in some for in AC68080). PS1 had vector math capability. Jaguar did not (but had DSP like MAC, hardware divider and special instructions which allows fairly convenient matrix multiplication). Sega 32X had dual SuperH2 which was nice to develop on but performance was ok-ish. Saturn had the same but with higher clock. Obviously Saturn (and I think 32X also) had some h/w texturing capabilities so at least that one came accelerated somehow. Both Jaguar and Saturn had their processors having direct access to graphics units (and in case of jaguar close relation to blitter and object processor).

So let me get this straight... someone really expects that slapping DSP to 68k bus (instead close to what actually generates graphics) would solve the problems with inadequacies of Amiga hardware in terms of 3D gaming? How naive can you be?!

[TEG]

Originally Posted by Promilus

So let me get this straight... someone really expects that slapping DSP to 68k bus (instead close to what actually generates graphics) would solve the problems with inadequacies of Amiga hardware in terms of 3D gaming? How naive can you be?!

I know nothing about DSP programming but what about the demonstrations here on the Falcon ? Seems not bad at all. The author is @AnimaInCorpore and in another thread, when I asked him, he said the DSP was not such a big challenge to use.

[ Show youtube player ]

[ Show youtube player ]

[Bruce Abbott]

Originally Posted by TEG

Thanks all for the explanations about Jeff Franck and Jeff Porter. However Ali being the top manager, he can't be completely cleared.

Sure. Ali could have canned the 'low cost' A300 idea and gone with just making A500s until nobody wanted them anymore, while pushing for better PCs. That's what all the pundits were recommending, including the bank. Only problem is in 1991/1992 there was a downturn in the PC industry and Commodore got stuck with a bunch of PCs that it couldn't sell. So they would need the Amiga for a while longer.

It's certain that without drivers PCMCIA was useless.

Not true. PCMCIA memory cards worked without a driver. That was the primary reason for putting PCMCIA on the A600.

So it lead once again to the question why they did half the job? Why they didn't sold PCMCIA cards with the good driver at the launch of the A600 and A1200 !??

You don't understand. Apart from memory cards complying with the PCMCIA standard, each card needs its own driver. Obviously cards designed for PC laptops would only come with drivers for DOS and/or Windows. Even other PC OS's such as Linux were in the same boat as the Amiga in this respect (which was good for me because I was able to crib off the Linux driver source code when writing the CNET driver).

Commodore didn't have to put PCMCIA in the A300. They could have just made a cut down A500 that wasn't at all expandable, for people who just wanted the base model to play games. They didn't need to put PCMCIA in the A1200 either. People would find other ways to get Ethernet into it (eg. through the clock port) and we wouldn't know what we were missing. But I'm glad they did, because apart from the increased functionality I learned a lot about how PCMCIA and Ethernet worked.

Finally do we know why the A600 was more expensive to manufacture than the A500? From what is said here, there was some TTL added to manage the IDE, the IDE connector and the PCMCIA connector.
I don't understand how it can overtake the gain of the CMD process and the remove of the numeric pad.

The A600 was more expensive primary because it used smd parts and a 4 layer PCB, which at that time were significantly more expensive - particularly for Commodore because they had little experience with smd. But they would need that experience if they wanted to move forward. The A1200 benefited greatly from it.

The TTL chips mentioned were not a big deal. There was no need for a special custom chip in the A600 because it only had a 16 bit bus. This is a lot less complex than the A1200 and A4000 which have a 32 bit bus that needs to be steered to 16 bit for some devices. The A600 only needed two 74LS245 chips to interface between the CPU and custom chips. The A4000 didn't have PCMCIA so of course it wouldn't need any buffers for that, but Bridgette replaced six 74F646 and four 74F245 chips used in the A3000.

[TCD]

Originally Posted by Bruce Abbott

Sure. Ali could have canned the 'low cost' A300 idea and gone with just making A500s until nobody wanted them anymore

But that's exactly what they did with the C64. At the very least the decisions weren't consistent.

[Bruce Abbott]

Originally Posted by hammer

A1200's development state is too far in the product development cycle when Lew was hired for the hot seat.

This is silly. Lew Eggebrecht's vision for the Amiga was to concentrate on the low-end and produce something else (not Amiga compatible) for the high end. If brought in earlier he wouldn't be for putting DSP in low-end machines (ie. Amigas) any more than he was in 1993.

Dave Haynie: Lew Eggebrecht has nothing to do with the Amiga as you all know it.

Dave Haynie also said:-

the Amiga, as we did it back in the day, was cutting edge stuff. The idea was to keep advancing the technology, to do things better.

Yes, and what good did it do? Miner and co. kept trying to make it more 'cutting edge' with expensive VRAM and high resolutions. Dave Haynie himself worked on the A3000, which was a high-end disaster. They were all trying to 'advance the technology' while ignoring the 'outdated' A500's popularity. They wasted years trying to produce an AAA chipset that would blow PC's out of the water - and ended up with nothing.

But fans here would do the same - piss away expensive R&D on their 'dream machine' which would have so much stuff in it that it would be impossible to build, let alone be cheap enough to become popular. Then they would blame management for their own failure to produce a machine that the market wanted.

[Bruce Abbott]

Originally Posted by TCD

But that's exactly what they did with the C64. At the very least the decisions weren't consistent.

Keep doing the same thing and you are accused of not changing with the times. Do something different and you are 'inconsistent'.

I'm not saying it wouldn't have been better for Commodore to just keep making the A500. In fact I agree with it. Hell, it might even have been better for Commodore if they had just stuck to making office furniture.

But I don't care about what was best for Commodore - I care about what computers they produced. A Commodore that only made the A500 and then switched to whatever was the hottest product at the time (eg. this office chair) is not the Commodore I want. If producing the A600 meant they would go bankrupt, so be it.

But Commodore didn't just keep making the C64 until it died. They also produced the C16 and Plus 4, the C128, the C64GS and the C65 (pity they didn't make more of them). So we are actually seeing some consistency here. Commodore had a long history of bringing out 'lower cost' models on the idea that it would boost sales. It worked for the VIC-20, didn't work for the C16, didn't work for the C64GS... and we can't put the blame for the C16 on Gould or Ali or Sydnes because that was Jack's idea.

[TCD]

Originally Posted by Bruce Abbott

But Commodore didn't just keep making the C64 until it died. They also produced the C16 and Plus 4, the C128, the C64GS and the C65 (pity they didn't make more of them).

But they never replaced the C64 with an 'updated' model. Releasing a replacement for the A500 didn't make a lot of sense, but releasing two of them in such a short amount of time makes it very clear that Commodore didn't have a clear strategy for the Amiga. It almost seems like a lot of decisions were second guessed the moment they made it to the market.

[hammer]

Originally Posted by Promilus

One of the most common misconception is that DSP is math co-processor. No, DSP is DSP with architecture designed to fit that actual job. And while DSP could've supplement 020 with some nice additions related to 3D graphics generation it was hardly anything which particularly suited DSP back then. That's why there was a need to develop SuperFX chip for SNES in the first place, existing DSP-1 designs were hardly suitable.

Actual example how to done things fairly efficient was PS1 - GTE is centered around vector math co-processor which basically translates to SIMD capability present basically in each and every processor out there (and yes, also in some for in AC68080). PS1 had vector math capability. Jaguar did not (but had DSP like MAC, hardware divider and special instructions which allows fairly convenient matrix multiplication). Sega 32X had dual SuperH2 which was nice to develop on but performance was ok-ish. Saturn had the same but with higher clock. Obviously Saturn (and I think 32X also) had some h/w texturing capabilities so at least that one came accelerated somehow. Both Jaguar and Saturn had their processors having direct access to graphics units (and in case of jaguar close relation to blitter and object processor).

So let me get this straight... someone really expects that slapping DSP to 68k bus (instead close to what actually generates graphics) would solve the problems with inadequacies of Amiga hardware in terms of 3D gaming? How naive can you be?!

My comment about DSP on 020/030 bus is only for the DSP3210 context.

I'm aware of less flexible DSPs.

AT&T marketed DSP3210 as "3D and multimedia" DSP.


Again, http://bitsavers.trailing-edge.com/c...tion_Guide.pdf
From page 19 of 140, AT&T marketed multiple DSP3210 as a 3D graphics workstation co-processor array. AA3000+ only has a single DSP3210.

From page 20 of 140,

 "DSP32C, 32-bit CMOS DSP (3-D Graphics Floating-Point Accelerator with Graphics Application Library)"

AT&T has other DSPs designed for audio.


https://user.xmission.com/~legalize/vintage/esv.html
Evans & Sutherland ESV Workstation used 44 DSP32C for its 3D acceleration!

The graphics subsystem consists of a command FIFO filled by the CPU with a stream of commands.

The command stream is dispersed to any one of up to 44 AT&T DSP32C signal processing chips. 

The model number roughly correlates to the number of DSP chips in the system, with four in the base ESV/3 
and 44 in the ESV/50. 

In my collection I have an ESV/10 and an ESV/50. 


The DSP chips perform vertex processing operations and produces a command-stream for the custom 
ASIC pixel processor chip that is connected to the 88-bit deep frame buffer. 

The pixel processor command stream handled 2D Blit operations and antialiased rasterization of point, line and span primitives. 

A 'span' primitive is an aliased line used for scan conversion of polygon primitives. The polygons are rasterized into a series 
of spans by the DSP chips.

Intel i860 wasn't the only competitor in the early 3D acceleration.

Amiga Hombre with a single PA-RISC with custom 3D extensions for 3D... this is NOT a many-core GPU characteristic of early "proper" workstation class 3D and modern-day GPUs. At best, Amiga Hombre is just a low-end games console.

SGI employed many core i860 as its geometry engine.

There's a reason why modern GpGPUs companies are founded by workstation graphics engineers.

https://en.wikipedia.org/wiki/Race_Drivin%27

Race Drivin' improved upon its predecessor in several ways:

It had improved handling courtesy of a faster microprocessor and more efficient software. The TMS34010 used for car modeling was replaced with an AT&T DSP32C which is faster and has floating point. Instead of modeling a car with only two wheels as Hard Drivin' did, Race Drivin' could model a car with all 4 wheels.

This Atari Games Inc. is separate from Jack Tramiel's Atari Corporation (Atari's consumer products division).

For 3D, DSP32C is faster than TMS34010.

Ignorance is bliss.

Reference
https://en.wikipedia.org/wiki/Atari_Games

MAME that runs Race Drivin's machine emulation would have a DSP32C emulator.

https://ia903005.us.archive.org/Book...ale=4&rotate=0
During 1989, DSP32C is around 24 Mhz. This is used as an FPU accelerator add-on card for the PC.

AA3000+'s DSP3210 is the 50Mhz variant. The DSP3210 is a variant of the DSP32C with a more powerful interface to external memory.

Commodore is just a follower, not a leader at this point.

[Gorf]

Originally Posted by sandruzzo

@Gorf

Believe me, it is not so easy. What made PC/VGA so succesfull was that is was easy to develop

Mainly because of software abstraction layers and not because the hardware would have been that much easier to program directly.

Commodore could have dome more to support developers, but that's besides the point here ...

I just tried to answer the question why DSP was somewhat hyped back then and why it might have been a reasonable and economical decision to go that route.