Was anyone else disappointed with the A1200?

[Promilus]

SGI's RealityEngine i860 array helped power IRIS GL, the precursor to OpenGL. Intel i860 includes an MMX-like SIMD instruction set which later influenced Pentium's MMX.

That's all irrelevant to what DSP3210 was used in on A3000+ and how it could've accomplish some other things with regards to the actual point of this thread - A1200. And I did point out all the obvious pitfalls and yet you are just spamming more irrelevant stuff.

Before OpenGL, PHIGS 3D API was used. Large workstation vendors typically offered versions of PHIGS for their platforms, including DEC PHIGS, IBM's graPHIGS, and Sun's SunPHIGS.

Amiga was nowhere near anything like that and it's pretty obvious slapping DSP3210 to A1200 wouldn't make it one either.

And as for the mac... for starters it was basically obsolete one year later with introduction of Power Macs. It was also f..ing expensive (~2500$). When it comes to architecture AV, SCSI, Networking, 68040, DSP were sitting on shared bus which was controlled through big PSC DMA controller (obviously video chip had own vram as well). Which means DSP along with 68k were SITTING IN THE MIDDLE OF THE F*** CHIPSET. And guess what - Dave's approach is nowhere near that. So all you did have was 68k bus OUTSIDE of the chipset and pretty darn narrow access to CHIPRAM which is the only way to get through to Paula and Denise (or in this case Lisa).

Which means - you end up with design which requires fast ram to be actually useful (increasing cost well beyond that 20$ of DSP itself), requires slow data transfers back and forth to chipram and with every such transfer it stalls 68k/dsp. And with introduction of 040 and 060 it creates additional problems with caching. And also requires programming a different type of architecture and create some interface for data exchange. And what makes it even funnier - AA3000 recreation team did encounter a number of problems to resolve before it did start to work kind of properly (and the cause were DMAC chips but obviously late 68k CPU models would introduce own set of problems which means retrofitting DSP to A3k and A2k would've been PITA as well).

And again - DSP would not accelerate already existing 68k apps and when it would've become obsolete it would be harder to accelerate DSP-assisted apps by next gen unless acceleration was done purely through the use of high level functions (i.e. libraries which you can just create functional replacements for other architectures) which is NOT the way things were made back then in Amiga world. Commodore did try to enforce that but it was hardly a standard.

[Thorham]

What's the actual point anyone is trying to make?

[hammer]

Originally Posted by Promilus

Yeah...
https://www.phoronix.com/news/LibreO...enCL-Code-2022
Hmmm

LibreOffice 7.6's Calc still supports OpenCL.

Originally Posted by Promilus

Normal user pays for Spotify or Youtube Music or iTunes. This one is for nerds. And even I - being a pretty large nerd - encode audio without GPU. It's damn fast anyway.

For Pinnacle Studio 26 and OBS version 30, I still use NVENC.

Originally Posted by Promilus

I already mentioned all the problems I see with that particular approach. So let's assume A3000+ did kick off. You end up with selling A500Plus (no DSP), A3000 (no DSP) and some A3000+ (DSP) ... doesn't it sound like a great way to make ppl angry and fragment already small market even more?

It's only the "PCJr" mentality to reject $20 DSP3210 for A500 level SKU.

For object manipulator functions, Amiga Blitter does not use its full 32-bit width for its DMA cycle slot.

DSP3210's 8 KB is local on-chip memory which doesn't need to be cache coherent.

https://archive.org/details/dsp_2020...age/6/mode/2up
Commodore pointed out DSP3210's barrel shifter for graphics manipulation functions.

A1200 needs Fast RAM anyway.

Originally Posted by Promilus

So let's assume some ppl would actually code for DSP in A3000+ creating a separate branch of software for Amiga. With a vast majority of Amiga users WITHOUT DSP. Then you release A1200 with DSP and some fast ram (because it would be biggest pile of sh*t without it) - it already increases a price to a level where a 486SX25 systems begin to look tempting.

Prove $20 DSP3210 + 1 MB Fast RAM for A1200 will push into the 486SX-25 price range.

For 1993 retail, 486SX-25's price is around $900 to $1000 in the USA or 799 UKP in the UK.

199 UKP is for the CD32 FMV module with a higher $50 CL460.

CD32's FMV module has the following:

1. 24-bit DAC (STM's STV8438CV) for 16.7 million colors display.

2. MPEG-1 decoder (C-Cube CL450, 352 x 240 pixels @ 30hz, 352 x 288 pixels at 25 Hz, pixel interpolation and frame duplication to produce output formats of 704 x 240 pixels at 60 Hz or 704 x 288 pixels at 50 Hz ),

https://websrv.cecs.uci.edu/~papers/...LES/060803.PDF

CL450 has about 398K transistors with up to 40 MHz. CL450 includes a programmable on-chip "purpose-built" RISC processor with some assist hardware. In quantities of 100K or more per year, the price is less than $50 in 1992.

3. LSI l64111qc (Digital Audio Decoder, 16-bit DAC),

4. 512 KB local RAM, NEC 423260 DRAM 4Mbit (512 KB) with 80 ns. Similar to 512 KB Fast RAM with 80 ns.

5. Lattice ispLSI 1024-60LJ CPLD.

Originally Posted by Promilus

It's 93. Doom kicks in. It is not FPU bound, it's all regular integer and fixed point. And 3210 ain't so great in that. But let's assume Amiga port is released and kind of works. How many of those improved A1200 you'd have to sell to keep Commodore afloat?

DSP3210 @ 50 Mhz with 8K local on-chip memory is still faster than 68030 @ 50 Mhz with 256 byte + 256 byte L1 cache.

DSP3210 has dual pipelines for separate integer and floating point.

You can use FP32 as an integer.

68882's math operation per second rating is slow in the low 1 MFLOPS range.
Even if you use 68882 as an integer compute device, it's still slow. Fixed point ALU is faster games.

Originally Posted by Promilus

Ok, let's assume you sold enough. And CD32 DSP empowered also did kick in. So you know Moto won't produce 68k anymore. You have to dump all that behind.

Not a problem for PowerPC 601 @ 60 Mhz.

DSP3210 has 22 GPR while PowerPC has 32 GPR.

Amiga Hombre 's PA-RISC aimed around 120 Mhz. If Commodore's PA-RISC didn't include an FPU, then DSP3210 @ 66 Mhz (33 MFLOPS FP32) is still useful.

Originally Posted by Promilus

But it's 95. PS1 shows up and smashes your gaming platform to bits. You choose PPC - end up with hand in the potty around 2007. You choose PA-RISC - end up the same even sooner. Maybe then MIPS - like PS1 and PS2? The same result.

Gaming PC adapted into Pentium class gaming experience around 1995 to 1996.

PS1 is not Z-buffer 3D accelerated and purely fixed point device.

[Octopus66]

Originally Posted by Thorham

What's the actual point anyone is trying to make?

lol

[hammer]

Originally Posted by Promilus

That's all irrelevant to what DSP3210 was used in on A3000+ and how it could've accomplish some other things with regards to the actual point of this thread - A1200. And I did point out all the obvious pitfalls and yet you are just spamming more irrelevant stuff.

Reminder, DSP3210's barrel shifter @ 50 Mhz is faster than either 68EC020 @ 7 Mhz barrel shifter effective or Alice Blitter @ 3.5 Mhz.

A CPU ping-pong with Chip RAM is not recommended, hence Fast RAM is needed.

Originally Posted by Promilus

Amiga was nowhere near anything like that and it's pretty obvious slapping DSP3210 to A1200 wouldn't make it one either.

And as for the mac... for starters it was basically obsolete one year later with introduction of Power Macs. It was also f..ing expensive (~2500$).

Quadra 660AV's price is similar to A4000/040, but Quadra 660AV has superior features.

Retail price tells nothing about BOM cost when Quadra 605 (with 68LC040-25) has $1000 USD.

PowerPC 601's FPU is powerful when compared to P5 Pentium.

https://gordonbell.azurewebsites.net...930818%20c.pdf
SpecFP92 score in 1993

486DX2 66Mhz = 16
Pentium 66 Mhz = 56.9
Alpha 200 Mhz = 163
PA-7100 100 Mhz = 150.6
PPC 601 66 Mhz= 80 (PowerMac experience in 1994)
R4400 150 Mhz = 105.2
SuperSPARC 50 Mhz = 85

P54 Pentium ramped up clock speed in 1994.

Apple's customer base can sustain 1.2 million unit shipments during 1994 which is different from Amiga's customer base relative to mass production targets. Most of Amiga's customer base is above the game console's price range.

https://citeseerx.ist.psu.edu/docume...760071fcdabfe9
At 80 Mhz, the PPC 601 processor used by Parsytec is rated at 80 Mflops peak and at 93 SPECfp92. PPC 601 reached 1 IPC for its FPU. Parsytec's PPC was introduced in September 1994.

https://netlib.org/performance/html/..._94.notes.html
1994's Pentium 100 Mhz reached SPECfp92 = 81.8.

Pentium 100 Mhz and PPC 601 80Mhz were not in the $20 DSP3210 price range in 1994.

https://archive.computerhistory.org/...-05-01-acc.pdf
From DataQuest 1995, Page 227 of 417 for 1996
68040-25 = $93.75 , crazy to stay on 68K.

68LC040-25= $46.00
80486DX4-75 = $81.00

Pentium-66 = $92.50
Pentium-75 = $92.50
Pentium-100 = $154.57

PowerPC-601-66 = $97.45
PowerPC-601-80 = $128.08
Power PC 603-80 = $90.82

Originally Posted by Promilus

When it comes to architecture AV, SCSI, Networking, 68040, DSP were sitting on shared bus which was controlled through big PSC DMA controller (obviously video chip had own vram as well). Which means DSP along with 68k were SITTING IN THE MIDDLE OF THE F*** CHIPSET. And guess what - Dave's approach is nowhere near that. So all you did have was 68k bus OUTSIDE of the chipset and pretty darn narrow access to CHIPRAM which is the only way to get through to Paula and Denise (or in this case Lisa).

Did you F*** forgotten CPUFastBilt patches on the Amiga with CPU accelerators with 32-bit Fast RAM?

[ Show youtube player ]
Star Wars Dark Forces 68K on the A1200's AGA display with compute power provided by PiStorm32-Emu68.

320x200p 256 colors passed 50 fps into the near 60 fps range.
---------------

[ Show youtube player ]
Beats Of Rage (OpenBOR) on the A1200's AGA display with compute power provided by PiStorm32-Emu68.

At typical action gaming resolution for 1992 to 1994, AGA (Lisa) is fast enough as a frame buffer. AGA chipset is for backward compatibility with OCS.

AGA is too slow for a Quadra 66AV's desktop GUI performance.

Examples with Quake demo1 benchmark with PiSTorm32-Emu68-RPi 4B on AGA (Lisa) display,
640x200p 256 colors = 29 fps. Twice the pixels, half the frame rate from Star Wars Dark Forces's near 60 fps example.
640x400i 256 colors (NTSC Lace) = 14.5 fps.
640x400i HAM6 (NTSC Lace, 6 bitplanes) = 13.5 fps.

640x400p 256 colors (Double NTSC) = 10 fps. Diablo wouldn't be fast.
640x400p HAM6 (Double NTSC, 6 bitplanes) = 12.38 fps.

The graphics raster needs to be on Fast RAM.

Again, a fast CPU (or compute device) ping-pong with Chip RAM is not recommended, hence Fast RAM is needed. Comparing the difference between frames and sending the difference can reduce data transfers between Fast RAM and Chip RAM domains.

On desktop GUI performance with games like StarCraft, PiStorm32-Emu68 CPU on A1200's AGA display is slower than my Pentium 166 with S3 Trio 64 UV+ PCI. StarCraft and Diablo for Windows 95 still runs on Windows 11. Diablo on the Amiga needs native PCI RTG.

Phase 5's Blizzard PPC/Blizzard Vision (3DLabs Premedia 2 PCI) only needs a CD32 card, hence faster Amiga clones could be built. Phase 5 couldn't expand the Amiga install base by themselves.

Originally Posted by Promilus

Which means - you end up with design which requires fast ram to be actually useful (increasing cost well beyond that 20$ of DSP itself), requires slow data transfers back and forth to chipram and with every such transfer it stalls 68k/dsp. And with introduction of 040 and 060 it creates additional problems with caching.

Reminder, 3.5 Mhz x 16 bit = 7 MB/s.

DSP3210's on-chip 8 KB is local memory i.e. cache coherence not required.

68040 uses MMU to mark the Chip RAM address range as data uncachable. The Amiga chipset is cache coherence incompetent.

Originally Posted by Promilus

And also requires programming a different type of architecture and create some interface for data exchange.

No different from Amiga blitter. Both Blitter and DSP3210 have setup costs.

Originally Posted by Promilus

And what makes it even funnier - AA3000 recreation team did encounter a number of problems to resolve before it did start to work kind of properly (and the cause were DMAC chips but obviously late 68k CPU models would introduce own set of problems which means retrofitting DSP to A3k and A2k would've been PITA as well).

My A3000's SDMAC DMA/SCSI/Ramsey works. Super Buster has problems.

[Thorham]

Originally Posted by Octopus66

lol

[pandy71]

Originally Posted by hammer

YOU. Again, who are you again? You're from Poland!

This is not the first time - do you have something against Poland?

[hammer]

Originally Posted by pandy71

This is not the first time - do you have something against Poland?

No. The cost perspective could be different from the US.

[pandy71]

Originally Posted by Promilus

It's great to see you are so adamant about providing examples of such great systems basically everyone in the world had... So the first link ... yeah, so popular system, and shortcommings of 32016 ... could've been easily cured with 32532 released in '87 but who cares...

Depends on many factor, price availability, software existence etc...From some reason someone decided to use TMS320C10 as accelerator for number crunching.I just tried to show that you intentionally narrowing hypothetical areas where DSP can be used to simply win dispute.

Originally Posted by Promilus

And the rest publications... well there were some publications about using GPU pipelines way before they became able to "general computing" to do some stuff by encoding tasks as graphic data and then decoding it from processed pixels on screen. So if there was a publication about that it means GPUs back then were widely used that way, right? Because that's exactly what you're trying to pull here with all those links. No, there were no 8x DSP picture processors in consumer electronics, nor there were numeric coprocessors made from DSP for home computers. There were only things like Macintosh Quadra with crippled by design 68040 and DSP offloading graphic calculations... So a predefined role for DSP, not any generic accelerator to audio, video, media...

True and we all know that x86 shine in dBase III and Lotus 123. DSP can only perform ROM based code and they can't be programmed to perform other tasks even if their HW and ISA can do other things.

Originally Posted by Promilus

Every popular platform - DSP has specific role (audio or network or graphics, or media decoding, as a peripheral unit) and not generic role (as co-processor sitting beside main CPU). The role Commodore guys assigned to 3210 was to handle audio and modem stuff. And believe that would not change a bit for A1200 and the outcome of Commodore business.

If you say so then it must be true... DSP are worthless piece of silicone - simple FUD - for normal user 640KiB is more than enough. And Commander Keen on CGA is superb piece of entertainment.

Originally Posted by Promilus

yes, but not the way you propose it for A1200...

Of course not - A1200 with MC68060 easily outperform 20$ worth piece of sh..silicone such as DSP3210.

Originally Posted by Promilus

That's way faster interface than most of the DSP uses in consumer, entertainment equipment (ISA, ClockPort, Zorro, PCI vs PCIE). And it really wasn't an issue even with PCIe gen 2 when the whole HPC GPGPU thing started (milkyway system 1), I think we're at gen 5 now. And yet... basically no GPGPU in everyday's use. Maybe slightly more in smartphones but it's hard to say when there's NPU as well and a lot of apps uses that instead for image processing.

All true - high bandwidth with latency of hundred++ cycles is most desired thing for real time application.

Originally Posted by Promilus

And that's yet another over decade old approach with actually no real world application. And those combo processors do have zero copy memory ...

Who need GPU if we have multicore CPU's?

Originally Posted by Promilus

No, I'm not mixing anything, Handbrake uses experimental OCL scaling engine. It has nothing to do with actual encoding/decoding but is a use of GPGPU.

You can use https://upscale.wiki/wiki/Main_Page if you wish - strangely there is no efficient GPGPU video encoding despite DSP like GPU capabilitiesSo perhaps other things are important too - maybe SIMD/MIMD is not optimal architecture to perform video encoding or perhaps people are too stupid - who knows...

Originally Posted by Promilus

Yeah...https://www.phoronix.com/news/LibreO...enCL-Code-2022Hmmm

Yes, support is limited - perhaps because most people do not need GPGPU in their everyday PC usage cases or perhaps it is still easier to achieve satisfactory results with general CPU's but your point need to be shown from proper perspective - how many years took general applications to be multi threaded to efficiently use mullticore/multithread CPU's? 15 years perhaps?Everything need time.

Originally Posted by Promilus

Normal user pays for Spotify or Youtube Music or iTunes. This one is for nerds. And even I - being a pretty large nerd - encode audio without GPU. It's damn fast anyway.

Depends on settings and HW but OK - you don't use GPU for this - fine for me.

Originally Posted by Promilus

I already mentioned all the problems I see with that particular approach. So let's assume A3000+ did kick off. You end up with selling A500Plus (no DSP), A3000 (no DSP) and some A3000+ (DSP) ... doesn't it sound like a great way to make ppl angry and fragment already small market even more?

Let's assume that all AGA machines are equipped with HDD and 20$ 32 bit DSP - DSP is required to boost performance required to fully utilize AGA capabilities where main CPU was simply to slow.

Originally Posted by Promilus

So let's assume some ppl would actually code for DSP in A3000+ creating a separate branch of software for Amiga. With a vast majority of Amiga users WITHOUT DSP. Then you release A1200 with DSP and some fast ram (because it would be biggest pile of sh*t without it) - it already increases a price to a level where a 486SX25 systems begin to look tempting. Where are those killer apps?

This is moment when you don't know and i don't know as this never happened.

Originally Posted by Promilus

It's 93. Doom kicks in. It is not FPU bound, it's all regular integer and fixed point. And 3210 ain't so great in that. But let's assume Amiga port is released and kind of works. How many of those improved A1200 you'd have to sell to keep Commodore afloat? Ok, let's assume you sold enough. And CD32 DSP empowered also did kick in. So you know Moto won't produce 68k anymore. You have to dump all that behind.

Same as above - you don't know how market response and i don't know - we know that CBM probably collapse anyway as it was too late to manage CBM being in deep shit.

Originally Posted by Promilus

But it's 95. PS1 shows up and smashes your gaming platform to bits. You choose PPC - end up with hand in the potty around 2007. You choose PA-RISC - end up the same even sooner. Maybe then MIPS - like PS1 and PS2? The same result. And again - even more fragmented market and community. And along the way you abandon chipset and adopt PC solutions like graphic cards, sound cards, network cards etc. - first on PCI, then PCIe. So in te end if you're lucky you end up being different from PC world by OS and CPU. If you are lucky and developers want to write for your OS.

From my perspective seem this EOT especially as racial slur already happened twice in the thread - side to this have no pleasure to perform resuscitation over Gould rotten corpse.

[pandy71]

Originally Posted by hammer

No. The cost perspective could be different from the US.

There is many things in Europe different than in the US - same apply for France, Germany, Netherlands, Portugal, Spain, Italy or UK.

[dreadnought]

Originally Posted by hammer

No. The cost perspective could be different from the US.

You're from Australia. The cost perspective could be different from the US.

Or, we can allow that people from different countries are capable of processing and considering simple information regarding other countries.

[Bruce Abbott]

Originally Posted by dreadnought

The amazing thing is that this thread seems to be accelerating, not slowing down.

Someone must have inserted a DSP chip into it!

[Bruce Abbott]

Originally Posted by hammer

PPC 601 66 Mhz= 80 (PowerMac experience in 1994)
...
P54 Pentium ramped up clock speed in 1994.

[b]Apple's customer base can sustain 1.2 million unit shipments during 1994

Why are you talking about 1994? The A1200 was introduced in 1992.

https://archive.computerhistory.org/...-05-01-acc.pdf
From DataQuest 1995, Page 227 of 417 for 1996
68040-25 = $93.75 , crazy to stay on 68K.

Which 68k Amigas did Commodore introduce in 1995?

68LC040-25= $46.00
80486DX4-75 = $81.00

Pentium-66 = $92.50
Pentium-75 = $92.50
Pentium-100 = $154.57

PowerPC-601-66 = $97.45
PowerPC-601-80 = $128.08
Power PC 603-80 = $90.82

Not sure what you are trying to say here. Does the price matter or not? If it matters then that's 'PCjr mentality', right? If it doesn't then why are you quoting prices?

Did you F*** forgotten CPUFastBilt patches on the Amiga with CPU accelerators with 32-bit Fast RAM?

If you mean FastBlit, it's a disgusting hack that made no noticeable difference in my A1200.

Star Wars Dark Forces 68K on the A1200's AGA display with compute power provided by PiStorm32-Emu68.

320x200p 256 colors passed 50 fps into the near 60 fps range.

If you are trying to show how great the A1200 design was that it can use a chip released a decade or more after it was, I agree.

AGA is too slow for a Quadra 66AV's desktop GUI performance.

The computer that cost US$2,300 in 1993, only lasted one year, and couldn't run Amiga software? What has it got to do with this thread?

[Thorham]

Originally Posted by Bruce Abbott

If you mean FastBlit, it's a disgusting hack that made no noticeable difference in my A1200.

Works well enough on my A1200 with 50mhz 68030. Be sure to also use FText (big difference when using 16 color text editor syntax coloring).

[Bruce Abbott]

Originally Posted by hammer

It's only the "PCJr" mentality to reject $20 DSP3210 for A500 level SKU.

Cost of all the custom chips (and other chips except for DRAM) in the A1200 was $11. The A1200 was supposed to be a replacement for the A500 that only cost a little more (and would eventually be cheaper), not hundreds of dollars more. Putting a DSP chip in it would be an extravagance.

If by 'PCJr mentality' you mean building it to a price that fans could afford, what's wrong with that?

That's not why the PCjr bombed. The problem with the PCjr was that they cheapened it in areas that mattered, making it worse than the original PC (which was hardly great itself). Number one mistake was giving it a horrible chiclet keyboard (the PC's keyboard was its most praised feature). Next was making the effective CPU speed slower because it shared memory with the video system, and not giving it enough memory to run typical PC programs. Memory expansion was limited and the CPU couldn't be upgraded.

The PCjr did have improved graphics (16 colors in 320x200 was significantly better than 4) and sound (SN76489 was a lot better than PC speaker) but these improvements were not enough to make up for its performance and ergonomics being worse than the PC. And even after cutting it to the bone it still cost more than an Amiga 1000 did a year later.

Comparing the A1200 to that is silly. The A1200 had the same lovely keyboard as the A500. It had a much faster CPU, more RAM, a bigger ROM with the latest OS, built in hard drive interface and greatly improved graphics, while everything else was fully compatible. Furthermore it could be upgraded with no limit via the 32-bit expansion slot. All that for the same price the A500 was a year earlier.

The A1200 was the exact opposite of the PCjr, being at least as good as the previous model in every way and better in many ways. Sure it didn't have exotic new features such as a DSP chip, but at that price it wasn't expected. In fact in 1992 nobody was expecting a DSP chip in any Amiga. It certainly was not something anyone was disappointed about at the time.

[Bruce Abbott]

Originally Posted by Thorham

Works well enough on my A1200 with 50mhz 68030. Be sure to also use FText (big difference when using 16 color text editor syntax coloring).

I generally use Topaz 8, which is already rendered using the CPU. And I only run 4 or 8 colors, not 16. Perhaps that's why I didn't notice any difference.

[TCD]

Originally Posted by Thorham

What's the actual point anyone is trying to make?

That nobody cares about this thread anymore. There was once a time when threads would get closed once they were fully derailed.

[pixie]

Originally Posted by hammer

On desktop GUI performance with games like StarCraft, PiStorm32-Emu68 CPU on A1200's AGA display is slower than my Pentium 166 with S3 Trio 64 UV+ PCI. StarCraft and Diablo for Windows 95 still runs on Windows 11. Diablo on the Amiga needs native PCI RTG.

You're not using pistorm own gfx card driver?

[Thorham]

Originally Posted by Bruce Abbott

And I only run 4 or 8 colors, not 16. Perhaps that's why I didn't notice any difference.

Text scrolling is hideous in 16 colors with DblPal without FBlit FText combo.

Originally Posted by TCD

There was once a time when threads would get closed once they were fully derailed.

Very strange it hasn't happened yet

[Bruce Abbott]

Originally Posted by Gorf

Commodore needed a better product in 91/92.
It did not have such a product and did not survive.

But why didn't they have a better product? Commodore could easily have put a DSP chip in the 'A1000jr', but would fans be satisfied with that? Of course not. The same old ECS chipset with DSP thrown in to compensate would not go down well.

Dave Haynie put DSP in the A3000+, but couldn't finish the design because the AGA chipset wasn't working. Until they got it going there wouldn't be any Amiga with DSP. Furthermore the A3000+ was going to be expensive, and they needed a much cheaper machine to attract more customers. Haynie got upset when the A3000+ was cancelled, but it was the right decision.

Perhaps if the engineers had concentrated on producing the A1200 in 1991 rather than playing around with exotic DSP chips, it might have been released early enough to avoid Commodore's demise in 1994. Even when late it was a good seller, with demand constantly exceeding supply.

The really disappointing thing about the A1200 wasn't that it didn't have DSP or some other exotic feature, but that it didn't come out in 1991 after Gould 'promised' it would.