From: dhaynie@jersey.net (Dave Haynie) Newsgroups: comp.sys.amiga.misc Subject: Re: Mac & Amiga Smeared on The Site/MSNBC (T Date: Wed, 20 Aug 1997 03:10:28 GMT Organization: PIOS Computer Message-ID: <33fa59d8.1482047882@news.jersey.net> References: NNTP-Posting-Host: swip129.jersey.net X-Newsreader: Forte Free Agent 1.1/32.230 Lines: 112 Path: 195.95.96.10!news.unisource.be!planetinternet.be!newsgate.unisource.nl!bullseye.news.demon.net!demon!dispatch.news.demon.net!demon!feed1.news.erols.com!europa.clark.net!205.252.116.205!howland.erols.net!infeed2.internetmci.com!newsfeed.internetmci.com!192.48.96.125!in3.uu.net!209.66.0.12!news.jersey.net!not-for-mail On Sat, 16 Aug 97 13:10:20 EST, damocles@nostromo.gate.net (Randy Vice) wrote: >On Wed 13-Aug-1997 11:43p, wwt6591@lithium.cs.rit.edu wrote: > >w> I agree with this. Zorro3 is aging, and not nearly as fast as PCI. A new >w> Zorro4 could relieve that problem, but we'd still be forced to pay >Dave was working on Zorro IV, till the day he read the official stats on PCI, >and he just dumped the Zorro IV papers into the waste can without a second >thought. Kinda-sorta, but not quite. Back in 1991, I was working on a new overall system architecture, code-named "Acutiator". The goal was to develop a highly modular architecture, in which the system chips were vertically rather than horizontally integrated, so that each piece could be indepently replaced in a system. The motherboard was to be CPU-independent, running instead on a new inter-module bus I called the AMI bus (Amiga Modular Interconnect). The goals of the AMI bus were simple. Since it was crystal clear, even then, that the 68K's days were numbered, the system had to be CPU independent. The main idea was building efficient chip to chip interconnects, so the bus had to be physically small -- a minimal connection ran about 45 pins, while a 68030/68040 connection can easily take twice that number. This would free up more pins for actual use in each module, or make the modules cheaper just by cutting packaging costs. The bus had to be much faster than we had run before, I was targeting 100MB/s to start with, with a fairly simple and pre-engineered solution to move to 64-bits. Each module would be based on a custom chip which plugs directly onto the AMI bus, fully autoconfiguring. This eliminates user configurations, including options for things that Zorro II and even Zorro III still had to do with jumpers (AMI bus actually used an I2C bus for card configuration ROMs and programmable jumpers, so this stuff didn't have to be in the custom chip unless you wanted it that way). In mid '92, Intel released the PCI bus specs, and spun control of the PCI bus off to an independent group. I studied PCI, and even in its first incarnation, it basically hit every one of my requirements dead on: small footprint (about 45 pins minimally per chip), fast speed (132MB/s in the 32-bit, 33MHz profile), upgradabilty (64-bit, and later 66MHz profiles), autoconfig, as cheap on a card as on the motherboard, etc. And one thing I could never do on my own: it had, at the time, half the industry backing it. Today, it's basically everyone: PClone, PowerMac, HP, DEC, IBM, Sun, Intergraph, etc. The reasoning for this is simple: no individual company can hope to compete across the board with something this prevalent. No, it's not perfect, but they thought of things (like nested buses; a system can have up to 255 PCI buses if you really want) that I didn't think of. Ok, they don't have TPUs yet... >That is apparently what Dave thought as well. Remember, his last days were >looking into SVGA chipsets for the next gfx chips since they were faster, and >cheaper then the AAA or Hombre. Not exactly, but close. Basically, by the fall of '93, it was clear C= was out of money, nothing could be spent on AAA anymore, even if we wanted to. And there was a great deal of belief that it was just too little, too late. Hombre was going ahead, but realize, this design was still a ways off. At best, it might have been in game-machine format for Christmas of '94, but that would have presumed enough resources from Commodore and probably help from HP -- Ed Hepler might tell the tale more accurately and in better detail. Anyway, near the end we systems guys were called on to re-invent systems solutions to extend the life of the A4000. So Greg Berlin rounded up a few guys to turn the fast '040/'060 board design back on (this had been in progress in '92, but management cancelled it, even though it would have paid for itself in two months of A4000 production and delivered much faster '040 and '060 Amigas). And I started looking at doing a graphics board with a scan conversion function, using standard PCI graphics chips. And yes, by then, you could get AAA class specifications for about $25 or so. That's not to say a AAA clone, of course; most of these parts had faster but less sophisticated blitters (the AAA blitter had some neat features, new arithmetic functions like add with or without saturation, sort and tally ops, a copper interrupt, etc), and they could deliver AAA resolutions using DRAM rather than VRAM. And they existed for real, in working silicon. >Why should >w> it be a bad thing if we could take advantage of PC goodies? And do realize that "PCI" extends well beyond the old PC Clone these days. Part of that is simply that "PCs" have been moving into areas previously occupied only by workstations. Some of that's the idea of UNIX or even NT on the PC, but it's also the fact that PCI is faster than the predominant workstation buses (VME, for example) of earlier times. This is also why most of the workstation companies have moved to PCI themselves. >w> Custom stuff >w> is no longer really woth the money involved, unless it has performance >w> far beyond the current popular standards. If AI came up with something >w> like this, fine. A killer bus, or chipset, whatever. If it could be >w> adapted for use in a PC, even better, higher volume means lower costs. It's like everything else -- you have a certain amount of money to spend on developing a computer. If you don't chose the projects carefully, you'll wind up with an inferior product, or possibly even no product if you overrun the costs. Would a custom chip set do the whole Amiga community more good than, say, 20-30 more programmers working full time for the same development period? That's exactly the way you have to balance the equations. >Kinda sounds like a PIOS-ONE, huh? =) The PIOS One is basically what happens when you try to build an Acutiator type system entirely from off-the-shelf parts. Dave Haynie V.P. Hardware Engineering PIOS Computer haynie@pios.de "...no RISC, no fun"