Jules Richardson wrote:
Dave McGuire wrote:
...
In 1988 or
so, I used a similar OCR setup at work. I think it was
called Calera, but I'm not sure. It used an ISA board with a 16MHz
68020 and a bunch of memory on it. It ran under DOS. Its code was
all RAM-based, and was downloaded into the board at boot time by
programs run from autoexec.bat.
That's interesting - so I wonder if they made a design decision to use
the 68k, or did they already have a product based around a 68k (possibly
on a Mac?) and it made the porting effort to the PC platform a lot
easier just to sell the product with a copro board? Back then most OCR
products commanded a huge price tag, so doubtless adding the cost of an
ISA card didn't make much difference (but I assume that they bought in
the card from somewhere else rather than making in-house?)
The first generation Palantir (later renamed Calera) OCR machines contained a pipeline of
four 8 MHz 68000s, each doing a bit of the pipeline. It was a $30K machine.
The second generation machine upgraded to 68020s running at 16 MHz, as I recall, but it is
kinda blurry since I also spent a fair amount of time prototyping systems with 16 MHz
68000s from Thompson. We tried a number of configurations to see which gave the best
performace/$. The second generation machine came out at $30K, and they continued to sell
the first generation machine at a reduced price, like $24K. The funny thing is the first
generation machine was more expensive to build, so in time they switched to shipping 2nd
generation machines with delay loops to the customers who could afford only the $24K price
tag.
The 68020-based ISA accelerator was $2K, and ran as fast as the first generation machine.
Although a 16 MHz 68020 isn't four times as fast as four 8 MHz 68000s, the pipeline
nature of the 1st gen machine meant that frequently one or more processors were idled
waiting for input or output. The 68020 on the ISA card ran full out all the time that
work was available.
So why the 68020 on the ISA card? 68K code compatibility was part of the decision,
although the vast majority of code was done in C. The bigger reason was that at the time,
RISC was a new trend. At the time I don't know if Sun had announced SPARC yet. MIPS
was
around, but the R2K was expensive and hard to fit on an ISA card. R3K was manageable, and
that came later.
BTW, Caere also came out with 68020 accelerator board for doing OCR around the same time
Calera did, although they didn't use any ASICs for acceleration. It was all
software.
Eventually Caere acquired Calera, or they merged. Caere's Omnipage name lives on as a
Scansoft product, although they recently got acquired by a different company. I've
read
that the latest couple of versions (I think they are at Omnipage 16 now) are a step
backwards.