grr@cbmvax.UUCP (04/10/87)
In article <18261@ucbvax.BERKELEY.EDU> schoet@ernie.Berkeley.EDU.UUCP (Steve Schoettler) writes: > There must be something I missed. > If worst case design of peripherals for the A1000 is so bad that >manufacturers could only make 1 or 2 slot card cages, how is it that >now so many vendors can offer 2000-and-1 boxes? > I can understand how a redesign of a circuit board and adding custom >chips can give you more timing flexibility, but how can they do it to the >A1000 without adding wait states? The key issue is buffering the expansion connector signals with minium DC and capacitive loading. The A1000 expansion design basically gives you a bunch of "wires" dangling from processor pins. Putting any more loading on these lines than necessary may make a system stop working or become flakey. Normal production variation results in some machines that will never have problems, some worst case systems that are very sensitive and a statistical sample of worse than worst case cripples that have problems with any expanion. If an expansion box has buffers, then you should be able to have as many slots as the characteristics of your bus drivers allow. If an expansion box is not buffered, it may not work on some systems due to loading violations and will also be limited as to the number of slots. If an expansion device "passes the bus" without buffering and/or with excessive loading the same applies. If an expansion device passes through a buffered bus, then the delays from the multiple levels of buffering will eat up the timing margins and you are limited to about two daisy-chained devices. Why have many of the vendors designed devices contrary to these notions? 1) optimism - not understanding how many worst case systems you get out of a production run of ~150K units. Also expecting Commodore testing to catch all the marginal units. 2) price - doing it right costs money - people want cheap/affordable devices. 3) market expectations - customers want the "pass the bus" because it allows adding other expansion devices. 4) lack of a good example - C=A never made expansion boxes for developers to serve as a role model. The best example of a conforming A1000 "Zorro" expansion box is the PAL made by Byte-by-Byte which employs the C=A designs for backplane, memory and disk controller boards. It is quite expensive compared to the plug in expansion devices. CSA's expansion racks and ASDG's mini rack-D are similar design. The Allegra memory card doesn't try to pass the bus. Sorry if I've missed any of the other strictly conforming devices... Many of the most popular devices fall into a grey area in that they do not strictly follow the recommentations in the expansion documentation, but do work well in the vast majority of cases. I have no wish to argue with success, but I offer the following warnings: 1) Make sure that when you purchase the equipment that you reserve the right to return it (and get your money back) if it does not work in *YOUR* system. A note to this effect on your receipt can save you untold grief and pain. 2) If the vendor represents that modifications to your Amiga may be necessary or that additional devices are needed, assume that you will need these and factor them in on your cost comparisons. 3) Check the reviews in magazines with Amiga coverage and more important, the transactions on the various networks. This may not change your plans, but at least you can avoid unpleasant surprises. Well, I think that covers it. Life would have been much simpler if the Amiga had *no* expansion, or provided a high-dollar standardized bus interface like a Mac-II or a mini-computer/workstation. The A2000 represents an improvement, a compromise somewhere in between the two, but you pay a little more... -- George Robbins - now working for, uucp: {ihnp4|seismo|rutgers}!cbmvax!grr but no way officially representing arpa: cbmvax!grr@seismo.css.GOV Commodore, Engineering Department fone: 215-431-9255 (only by moonlite)
perry@sfsup.UUCP (04/10/87)
In article <18261@ucbvax.BERKELEY.EDU>, schoet@ernie.Berkeley.EDU (Steve Schoettler) writes: > If worst case design of peripherals for the A1000 is so bad that > manufacturers could only make 1 or 2 slot card cages, how is it that > now so many vendors can offer 2000-and-1 boxes? > I can understand how a redesign of a circuit board and adding custom > chips can give you more timing flexibility, but how can they do it to the > A1000 without adding wait states? > I have ultimate faith in ASDG; if it can be done, they will do it. > But why were we (at least I was) led to believe by so many hardware > vendors in the past that it couldn't be done? Proper (full spec) buffering allows a reasonably large number of expansion slots to be placed in ONE expansion box. This is what the framers of the Zorro spec envisioned from the start. Current slap-on-the-side-with-uh-oh-pass-thru products DO NOT PROPERLY buffer. You're gonna need proper buffering to allow proper operation. A slap-on-the-side-with-pass-thru product (if one existed which properly buffered the bus) would squander the bus buffering on exactly one board. This means downstream products which would presumably also buffer the bus would begin to incur wait states (slots in a single card rack though do not incur wait states if the containing card rack is the first card rack in the chain - and - to my knowledge no one seriously expects two fully buffered card racks to be plugged into each other and work reliably.). To sum up: Buffering is essential. Buffering more than once leads to wait states downstream. Sots products do not buffer because they are inefficient card racks (they contain only one board) and to buffer properly would mean wait states for products downstream. Card racks are more efficient card racks (like that one?) in that their bus buffering ammortizes over several boards. Also, thanks for the faith. You're right. If it can be done - we can do it. Perry S. Kivolowitz ASDG