home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!noc.near.net!hri.com!spool.mu.edu!yale.edu!jvnc.net!rutgers!cbmvax!daveh
- From: daveh@cbmvax.commodore.com (Dave Haynie)
- Newsgroups: comp.sys.amiga.misc
- Subject: Re: Problems with the A4000
- Message-ID: <35103@cbmvax.commodore.com>
- Date: 14 Sep 92 17:48:50 GMT
- References: <1992Sep12.083819.17967@news.iastate.edu>
- Reply-To: daveh@cbmvax.commodore.com (Dave Haynie)
- Organization: Commodore, West Chester, PA
- Lines: 119
-
- In article <1992Sep12.083819.17967@news.iastate.edu> barrett@iastate.edu (Marc N Barrett) writes:
-
- > The problem comes down to the architecture of the A4000 itself. As has been
- >rumored before, several good systems were canceled not long after DevCon '91,
- >a year ago. What has not been said is that these included all of the systems
- >that Dave Haynie designed that incorporated the AA chipset. Indeed, none of
- >the new Amiga systems -- including the A4000 -- were designed by him. They
- >were designed in a rush by junior engineers at Commodore.
-
- As usual, Mark is, to put it bluntly, talking out of his ass here.
-
- There was ONE system discussed at DevCon, something called I called the A3000+.
- It, like the A4000, was based on the A3000 system architecture and the AA
- chips. I had some other goodies put in there, and I had hoped that this would
- go on as a production machine, but it was just a development platform. We
- did development with it, then moved onto real production. The A4000 is the
- first fruit of that development effort.
-
- The A4000 project was headed by Greg Berlin, the other half of the main A3000
- engineering team. Greg is the only hardware designer who's been here longer
- than I have, by about six months. I was heavily involved in the A4000 design
- in the "AA" and Expansion subsystems. No, it wasn't a full-time job for me,
- basically because most of this had been prefected already, and Greg had two
- junior guys and one other senior guy (Scott Schaeffer, who did the '040 module)
- helping out on day-to-day debugging, etc. Which is just the way I wanted it --
- I started working on "AA" stuff back in the fall of '90, and by the time the
- A4000 project started I was moving onto other things.
-
- > The whole design of the A4000 can be summed up in a fairly short sentence:
- >the A4000 is an A3000 with the CPU moved to a card, the AA chipset kludged in,
- >and with the SCSI subsystem replaced with IDE.
-
- The AA chipset is hardly "kludged" in. The form of the AA chipset is something
- me, Greg, and the AA people (Bob Raible and Bill Thomas) discussed at length
- back before there was an AA. It was designed to drop fairly easily into an
- A3000-class machine, much like ECS did. Actually, the A4000 design is
- considerably cleaned up over my A3000+ prototype, in that Greg's people did a
- new gate array, Bridgette, which manages much of the system's datapaths more
- efficiently than the 9 or so TTL buffers the A3000+ used to do the same job.
-
- Again, Mark, why do you always post stuff you have absolutely no knowledge
- about? Especially things like the genesis of the A4000 and AA, a subject upon
- which only a very few people could produce a real authoritative post.
-
- >Basically, the A3000 was not designed for the AA chipset.
-
- Correct. In point of fact, the AA chipset was designed for the A3000, as I
- explained above.
-
- >The Ramsey and Buster chips in the A3000 were designed for use with the ECS,
- >in particular the 2M Agnus chip.
-
- The RAMSEY and Buster chips have ABSOLUTELY nothing to do with managing the
- chip bus.
-
- Gary and a couple of PALs manage the chip bus in both the A3000 and A4000.
- The signals from Gary are very generic, they work as well with AA as they do
- with ECS. Most of the control for a particular chip set comes from the chip
- bus controller, Alice in AA, Agnus in ECS. Gary just provides a few chip
- selects and a bit of buffer control logic (eg, it's basically controlling
- the CPU gating on and off the Chip bus, which is real simple).
-
- >The new Alice chip in the AA chipset -- the replacement for Agnus -- can
- >potentially address 4M of chip RAM.
-
- No it can't.
-
- > OK, I've established that the A4000 is an A3000 with the AA chipset kludged
- >in. Now it makes sense to me that it would have cost less in development costs
- >to simply leave the A3000's SCSI design there instead of removing it and
- >replacing it with something else. The costs of including SCSI in the A4000,
- >then, would have been practically zero in development costs.
-
- The inclusion of IDE cost a few months of Joe Augenbraun's time. Joe was a
- junior engineer working for Greg (he left C= a little while back to go invent
- HDTV in Princeton or some-such). Say what you will about IDE's inability to
- address lots of cool devices, I'll agree. But it cost practically nothing to
- develop, it's a very, very simple interface. And it costs very little to
- build, it's basically a PAL and a couple TTL buffers, not a full controller
- like SCSI. IDE was designed to be nearly free to add back when the PClone
- industry whipped it up.
-
- > In addition to the costs involved in removing the SCSI subsystem from the
- >A4000's design,
-
- We're now having to pay to remove SCSI from the A4000!?! Buy yourself a
- clue, man!
-
- >The new design was created by Randall Jessup, and includes extra hardware to
- >make the IDE interface and HD look somewhat like a SCSI interface and HD to
- >Amiga software.
-
- Randell did do the software. Far as I know, it's pretty much the same in both
- A600 and A4000, and in both cases, the IDE driver pretends to be a SCSI device,
- since most hard disk utilities know how to deal with SCSI devices. A good idea
- IMHO, no really big deal.
-
- >In the short-term, then, the prices on the A4000 and other new systems will
- >be higher than if they had used a SCSI design.
-
- Not on your life. You obviously know nothing about computer development to
- come up with this conclusion (not that anyone here had any doubts in the first
- place). I'll make up some numbers just to give you your first lesson (eg, I
- don't know what these things cost, but we'll ballpark it). Let's say that
- SCSI cost $25 per system, IDE $2 per system. Let's say that Randell and Augi
- each spent 3 months on IDE. I don't know what Randell and Augi made, but I
- guess around $25,000 would cover the two of them for three months (better than
- cover them, since they're doing other things at the same time).
-
- So how many machines to I need to make to cover the costs of IDE development.
- The differential here is $23, I need to make $25,000. So I'm making a profit
- on this change, all else being equal, on machine number 1087. Again, these
- are made up numbers, I suspect that the IDE stuff is actually paid for much
- sooner, since most of Randell's work was already done for the A600.
-
- --
- Dave Haynie Commodore-Amiga (Amiga 3000) "The Crew That Never Rests"
- {uunet|pyramid|rutgers}!cbmvax!daveh BIX: hazy
- "Work like a horse, drink like a fish" - Psychefunkapus
-