18 Dec
2001
18 Dec
'01
11:49 a.m.
From: Richard Erlacher <edick(a)idcomm.com>
feature,
then
No it wasn't, the 765 design was introduced in late '79, by then you were
locked to WD.
> Munged
> wacky formats like using deleted address mark for address mark {you can
post
format with
deleted data} and it was not designed to pump out all the raw
bits/splices/marks from the media.
What? I do seem to remember that the data fields could be written during
formatting with the WD parts but I don't think anybody ever used that though it would have been a good/smart feature for
software duplication.
There
was some confusion about whether it worked properly
because the part
responded
to some bytes by generating an address mark, though I
doubt it did that
while
writing the data field.
Two differing things. The WD part you could format a disk with Deleted data
marks
where data marks are normally found. Infact you could put all sorts of odd
stuff
in strange placesn using WD. The 765 was ucoded to do IBM standard formats
so a lot of the who's where is already known and mapped.
Things it did
do that the WD never had: Multiple seeks or recals, timing
for the stepper, head load delay, head settle delay.
Did it do "implied seeks" wherein the controller looked at where it was
and automatically computed the difference before moving the
heads?
No, I know of no chip that did. It made for a peice of that.
The biggest
difference: register based programming vs command packet to a
"port".
Back in the '77-'82 period I was probably responsible for the use of 100K
Western chips and it might as well have been decided on a coin toss. I was vs packet programming though. Perhaps you could cite
an example? The
Western
part is certainly register based. Isn't the NEC
part also just a register
set?
Internally the 765 has "registers". However, you feed it via one port
addres with a
command packet and after the data IO is done you read a status packet of 1-9
bytes
based on the command issued. It's obsious when you look at the part, the
765
has A0 for the port addressing (status and command/data) where A0 only has
meaning{it's only active during /CS not /DACK) during non-dma ops.
That was an advantage for those who were invested in a
software base, but
nobody
knew that back in '78-79. What's the 36C766?
Google comes up empty. I've
seen
some 37C665/666 types, but 36Cnnn? Who made them?
Several vendors including SMC and UMC. They were variations of the 765 with
rate generators and interface to disk plus IOports (parallel, serial and
even IDE).
Aimed at PCs they replaced the two serial ports, IDE, FDC and parallel
boards.
What's interesting, BTW, is that even Western, with
its institutional
prejudice
toward analog PLL's went with the 765 core once it
went to the fully
integrated
all-digital FDC, having dealt with the lower data rates
in the 1770/72/73
chips,
which were not shown to be capable of 500 Kbps for some
reason. Perhaps
there
was some advantage in the 765 core that made it more
amenable to
integration
with a digital clock extraction circuit at the higher
data rate. I doubt
that
Intel would have gone for the 765 type if there
weren't some manufacturing
advantage inherent in the silicon. That may be what's made the difference.
Intel certainly would have chosen the chip that was more economically
manufacturable, though maybe their primary economy came from the
already-established relationship (which they'd sabotage later) with NEC.
The 765 core did the step rate and may of the external things that the 1793
needed external hardware for. The preference for digital data seperation
was
pushed by NEC as it could be done with a small 32x4 prom and a latch with
good reliability compared to the often difficult analog designs. Also
digital
fits on silicon of the time better. I have a design and samples we did in
late
'81 to put the floppy side "glue" on one 2500 gate array that allowed for
data
sep, write pre comp, drive and motor selects and all the other things that
would end up on the super chips. The end result was a complete FDC two
chip combo that was half the price of discrete 765 or WD 1793 designs with
no performance compromizes. It was never marketed for obtuse reasons
and less than three years later several vendors were putting 765+glue on
one chip.
And the d7265 wa the ISO 3.5" tuned version that had a shorter VCO sync
time{post index gap time} and a shortend index gap. I believe most of the
765 cores are of the 7265 flavor.
The aside to that is that the SMC 9229 was a digital data sep/clock/precomp
that worked with both the 765 and 1793 with equal perfomance for all rates.
SMC also had a 765 core with analog PLL (9265 or 66) for those that prefered
analog. The 1770/2/3 problem was not data seperator in itself but process
speed
of the die, they{WD} flat out could not do the required 16mhz stuff then for
the
data sep and the other rate generators. The 1793 with external 9229 works
great at 500khz but, the 1793 only has to see something like 2 or 4mhz max
and therein lies the difference.
Allison