Path: utzoo!attcan!uunet!husc6!bloom-beacon!mit-eddie!ll-xn!ames!vsi1!daver!mips!sultra!dtynan From: dtynan@sultra.UUCP (Der Tynan) Newsgroups: comp.sys.nsc.32k Subject: Re: AUTHOR SPEAKS: '532 Manifesto Keywords: cheap nsc 532 Message-ID: <2659@sultra.UUCP> Date: 17 Nov 88 03:42:16 GMT References: <433@sdrc.UUCP> Organization: Tynan Computers, Sunnyvale, CA Lines: 122 In article <433@sdrc.UUCP>, scwilk@sdrc.UUCP (Ken_Wilkinson) writes: > > o Ethernet , parallel ports, serial ports would use the bus. We might if > enough room put the serial and parallel ports on the motherboard. Nah! The best way to handle serial ports is with a separate processor. Let's keep the '532 for *real work*, and let some pissant 68K handle the serial lines :-) > o memory will be MAC II style SIMM chips (the fastest we can buy :-}) > for 4 to 8 megs (maybe 4 meg with a daughter board for 8 megs) > NO memory expansion bus. If you need more than 8 megs tuff sh*t. Hmm. I wonder. Right now 4MB is loads, but what about next year? Why not make it so that one can use 1Mbx9 SIMMS or 256Kx9 SIMMS depending on choice (read budget). With 1MB SIMMS, we *should* be able to put 16 meg. on the board. Don't forget also, that the unfortunate (?) thing about a 32-bit bus, is you have to buy SIMMs four-at-a-time. Another concern, here, is parity. What is the overall concensus? Yay or nay? I don't even want to *think* about such budget killers as ECC. > Operating System: > > SYS V - First choice. > Minix - Possible, maybe a parallel effort? Race? > > BSD was the choice of most respondants, but no one has come forward to > do the port. If you use a '532 as a CPU, then Minix is a dead issue here (sorry, AST). Minix was never designed for something like that. Adding all the code to handle a screamer like the '532, would mean major work to Minix. Better to spend the time working on a BSD port. Yes, I said BSD. I writing this under SysV. PLEEZ PLEEZ PLEEZ, don't use SysV!!! It may be just dandy for people who want the stability (?) of AT&T, but they buy IBM anyway. The only reasonable SysV will probably be 4.0, which ain't available. Moreover, 4MB won't hack it (probably). I'll do the port (with help, of course :-). The only problem is, I don't have a source licence. The way I see it, there's the same problem with SysV. I don't think we can collectively put together ~$75K for a source-licence either, so unless we can find a PD version, or someone willing to allow us to use *their* licence, then we're stuck with Minix. Comments anyone? > o Do we use instruction cache? Is there enough room on the cpu motherboard? > Do we really need it, as the 532 has a burst mode. > (no) Assuming something like a 20MHz '532, then the clock cycle is 50ns. Hmm. The memory cycle will probably be something close. I'm not about to buy 4Mb of 50ns SIMMS!. Where's Steve Wilson when ya need him? How about an *optional* 64K cache? Too much? Oh well. > o use same serial uarts as ICM and preserve some of the driver code? > (no) Why not? As I understand it, the ICM uses 16450's. These are *nice* chips. They even have a version with a fifo. What's more, NSC makes them (does the word 'discount' sound familiar?). What did you have in mind instead? The 6850? The nice thing about the '450 (not counting FULL modem control, and interrupts on "change-of-status"), is the internal baud-rate generator. > o how much memory can we fit? > ( 8 megs max easily, but design for 1 meg min.) See above. Don't forget, the GCC binary is >512K. What's the OS going to burn? Don't forget too, that TCP/IP is now available (in SOURCE). What does that mean? Twice as much memory. > o bring out internal bus lines to a connector unbuffered? > (nahhhh) If there are PC-bus slots, to heck with it. HOWEVER, let's be a little more far-reaching than the great Blue Engineer. Make the *address* bus tri-stateable. The world of co-processors would be a *lot* better today, if IBM could think. With all that memory on the motherboard, you don't want to have to wait for a slow old DMA. Do it yourself. > o Clock speed, multi layers 4,6,8? Cost? > (?) Somehow, a CPU speed of 20MHz, sounds nice. It has a nice ring to it (I HOPE NOT :-). I would see the design as "three-tiered". A *fast* CPU, with cache. A not-so-fast main memory (120 -> 150ns), and a dead-slow PC/AT-bus. I would like to see the thing confined to six layers, but I'm not sure that is possible. With some parts running at 40MHz, there *have* to be power & ground planes. > o Memory speed, wait states. > (?) Go with 120 -> 150ns SIMM's. I think this is probably the best price/perfor- mance ratio. > Please comment by e-mail on the concerns and design. I will firm up Using email unfortunately, closes the subject to debate. I don't think the net bandwidth is wasted, when you consider we're plotting the overthrow of IBM, Apple and Sun in one fell swoop :-) > "THE '532 MANIFESTO" and then we can begin the real work! > > Ken Wilkinson (uunet!sdrc!scwilk) The real work has already begun. The actual architecture discussion is one of the most important considerations. I mean, you don't want to be two years down the road thinking "I wish I had more memory". The thought strikes me, that there may be too many chiefs at this stage. I have been collecting names of interested people, as have others. What we need here, is a little "Law and Order". Anyone want to volunteer as the "Gatekeeper". On a related point, in my estimation we need 20 hard-core check-book in-hand people, before any of this is reasonable. Of course, more is good too :-) What is the "head-count" right now? Anyone know? - Der PS: OK, one last point. Does anyone have a nice, catchy name for such a system? I'm sick of calling it the '532 el-cheapo-thing'. How about Merlin? -- dtynan@zorba.Tynan.COM (Dermot Tynan @ Tynan Computers) {apple,mips,pyramid,uunet}!Tynan.COM!dtynan --- If the Law is for the People, then why do we need Lawyers? ---