The DSP co-processor is brilliant, especially for the time.
A3000+ blows away the 4000. What were they thinking not producing this instead of the 4000?
Dave, was it simply a cost issue?
My friends, if it were ONLY a cost issue. No, it was a stupidity issue.
The A3000+ was mine, mine, all mine! Well, Randell Jesup and Eric Lavitsky (outside consultant) were integrating VCOS/VCAS (the DSP OS) into AmigaOS to go with. The DSP3210 was pretty impressive... it only did 32-bit floating point, but 5x-10x faster than the '040. Apple tapped all our hard work convincing AT&T to sell us the software cheap, a year later, in a series of Macs (Quadra AV or something like that).
The year was 1991, and Mehdi Ali had just appointed Bill Sydnes (the PC Jr. and Franklin Computer guy) head of Engineering. Sydnes first mission was basically just clobbering everything we were working on at the time. So I was basically forced to turn the A3000+ into a development-only platform (the last Rev was dubbed "AA3000" rather than "A3000+", I'll have schematics up one of these days). In fact, we were only allowed to mount these to a block of plywood, not allowed to put them in cases (mine, of course, was in a case... I did the computer graphics for the "Deathbed Vigil" on it. Didn't end well, though... while on loan to a friend, it was stolen).
The other machine we had in the worked was Joe Augenbraun's Amiga 1000+. This was probably the machine that would have really boosted the Amiga's profile. It sat squarely between the A500/A1200 and the A3000/A4000... detached keyboard, two Zorro slots, a new CPU socket, etc. The intent was to ship at around $800 in 1992, with AA chips and a 25MHz CPU (probably an EC020 or EC030, but still).
So, after killing these machines, Syndes went on to mess with the other project in the works. George Robbins was working on a project called the A300... a sub-A500 class machine. He had even worked out a super-cheap Genlock, which was probably going to be built-in.... George knew analog way better than I did at the time, and I was one of the better analog guys all told (George also designed the VIDIOT hybrid, which does the D/A conversion in the A500 and A2000, as well as the A500 and A1200). Syndes has all these changes made, still promising $50 less cost than the A500, but delivering $50 more cost. That was the A600, and as soon as it was done, they cancelled the A500, enough though it was still selling quite well.
They also tried to create a sub-A3000 machine, which was internally called the A2200, but which everyone else called the A1000jr (after the PCjr, of course). This was going to have ECS chips, not AA, and only Zorro II slots, but otherwise based around the A3000 architecture, with Augi's cheap IDE (a couple of PALs... IDE/ATA without DMA is really easy to do) replacing SCSI... that was Joe and Greg Berlin. They knew it wasn't what we wanted, but you sometimes have to listen to the boss.
Here's where the unique nature of Commodore comes in. Commodore did so well internationally because each region had their own marketing and sales company, which ran pretty independently. Sometimes, over the years, you'd see a cool prototype shown off at a CES or Comdex that never made it out. Sometimes it was Jack or some other boss killing it off, but sometimes there just weren't enough orders from the various sales companies to justify production. And that's just what happened to the A1000jr. No orders.
Greg and his team immediately began work on a machine internally dubbed the A3400, which ultimately became the A4000. It was also very A3000-based, but other than the lack of SCSI, not so bad. And it of course did use the AA (Pandora, AGA, etc) chipset, based on the work I did on the A3000+.
During the A2200 project, Greg asked Scott Schaeffer to design "the cheapest 68040 card known to mankind". Scott had done our first '040 card, which you never got to see. This was actually at the A3000 launch in 1990... we had hired Scott specifically for this... he already had '040 experience, even though it hadn't shipped yet. And in fact, we had one of the first OSs actually functional on it... so early, in fact, that was had to get official permission from Motorola to show it off. Which we got... and then the managers decided not to show it. That card was pretty big, with its own L2 cache.
The main reason it wasn't announced at the time was compatibility -- the '040 was designed to run way hotter than any chip we had used in the past; this was right before CPUs started always having heat sinks on them. There was real concern that the '040, and particularly that large module, wouldn't be kept cool enough in the A3000. So no announcement, and that card was cancelled.
So Scott's cheap '040 card was Greg's very good idea to make the A1000jr less embarrassing, at least via upgrade. When the A4000 came around, this was ready, so it became the A4000's default CPU card.
-Dave
