dtynan@sultra.UUCP (Der Tynan) (11/19/88)
In article <1041@raspail.UUCP>, bga@raspail.UUCP (Bruce Albrecht) writes: > In article <2659@sultra.UUCP>, dtynan@sultra.UUCP I wrote: >> 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? > >I'm not quite ready to count myself as a checkbook-in-hand person just yet, but >I might be, if someone can provide a rough estimate of what it will cost to >build one. I got some good figures from Steve Ligett, who says the artwork would cost ~$200, the setup fee for the PC's would be ~$1000, and the per-PC costs would be ~$150. It should be noted, that I haven't checked these figures. Neither has Steve. They are ballpark numbers. With this in mind, approx- imately 20 people would make the BARE PC BOARD cost $200. After that, you're on your own. SIMM's tend to be expensive, but I don't mind paying for a good system, when it is in little (unnoticeable) increments. The PC board would be of the XT (or AT) variety. A case costs $20->40, depending on taste. A PSU can cost up to $70, depending on power, and a keyboard can cost up to $100. These are all "off-the-cuff" figures. As for the minor detail of a 20MHz '532 chip (or chip set?), someone from NSC will have to answer that one. Of course, on top of that, is the OS. - Der -- 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? ---
pcg@aber-cs.UUCP (Piercarlo Grandi) (11/20/88)
In article <2661@sultra.UUCP> dtynan@sultra.UUCP (Der Tynan) writes:
I got some good figures from Steve Ligett, who says the
artwork would cost ~$200, the setup fee for the PC's would be
~$1000, and the per-PC costs would be ~$150. It should be noted,
that I haven't checked these figures. Neither has Steve. They are
ballpark numbers. With this in mind, approx- imately 20 people
would make the BARE PC BOARD cost $200. After that, you're on
your own. SIMM's tend to be expensive, but I don't mind paying for
a good system, when it is in little (unnoticeable) increments.
The PC board would be of the XT (or AT) variety. A case costs
$20->40, depending on taste. A PSU can cost up to $70, depending
on power, and a keyboard can cost up to $100. These are all
"off-the-cuff" figures. As for the minor detail of a 20MHz '532
chip (or chip set?), someone from NSC will have to answer that one.
Of course, on top of that, is the OS.
Stop daydreaming Der! a 532 is too expensive, and probably not needed. I
think a 332 would do. The problem with 532 is that it is not only quite
expensive in small quantities, but it is too damn fast. The real cost of a
system is not in the CPU, is in the support around it. Think over it.
In my never humble opinion, why not have a look at the nice at 286 single board
design from Byte, published several months ago ? If such a design had a 32k CPU
instead of a 286 (and yes, I think this could be done, it is just a bit hard).
In other words, the aim should be to develop a 386SX not a 386 type of system.
And as long as it is as compatible as possible with a 286, so much the better.
Would it be possible to plug a 32k based daughterboard in the socket for a 286
or 386 CPU in an existing AT board ? (no, I am not a jester!)
I am going to buy a 386 box with S5.3.2. Remeber, the only real superiority of
BSD over 5.3.2 is the Fast file system, and Interactive's version of 386 5.3.2
has it... in BSD you do not get mapped files, streams etc... I am going to
use it for developing th next great capability based distributed operating
system. I would for sure use a 32k based cpu instead of the 386 if I could run
UNIX on it. But (BIG NUMBERS ONLY) NatSemi will never come forward with such a
thing, will it ? My only hope is all of YOU.
--
Piercarlo "Peter" Grandi INET: pcg@cs.aber.ac.uk
Sw.Eng. Group, Dept. of Computer Science UUCP: ...!mcvax!ukc!aber-cs!pcg
UCW, Penglais, Aberystwyth, WALES SY23 3BX (UK)
ken@gatech.edu (Ken Seefried iii) (11/21/88)
In article <256@aber-cs.UUCP> pcg@cs.aber.ac.uk (Piercarlo Grandi) writes: >Stop daydreaming Der! a 532 is too expensive, and probably not needed. I >think a 332 would do. The problem with 532 is that it is not only quite >expensive in small quantities, but it is too damn fast. The real cost of a >system is not in the CPU, is in the support around it. Think over it. > I agree with this whole hartedly, and will be detailing my thoughts on the subject in the next few days... > <stuff deleted> > >I am going to buy a 386 box with S5.3.2. Remeber, the only real superiority of >BSD over 5.3.2 is the Fast file system, What about things like long file names and symbolic links to name a few? Sorry, but Sys V.3 is still not that great. Maybe System V.4......... > and Interactive's version of 386 5.3.2 >has it... For the record, no it doesn't. It just calls what it has Fast File System. What ISC has done is make the in kernal stuff more efficient. It is, however, still the System V file system and bears no simblance to the BSD FFS. > in BSD you do not get mapped files, streams etc... I'll buy that...but you also forgot other good things like HDB uucp. I'd still much rather have BSD, but thats a religious issue. > I am going to >use it for developing th next great capability based distributed operating >system. I would for sure use a 32k based cpu instead of the 386 if I could run >UNIX on it. This is silly. Writing an OS is a big enough pain in the ass without having to deal with a horrid bastard architecture like the '386 (although, to be fair, it does have some interesting features). For the price of a MINIMUM, which would be FAR from what an OS developer would need, '386 based system (16MHz, 2MB, 40-60MB disk, EGA, ISC 386ix v.2.0 == $5500+) you should be able to dig up a usable 32k based Unix machine. Hell, there are tons of 680x0 based systems that fall into that price range that would be better than a '386. As you are at a University, you could probably get a Sun 3/50 with big disk and tape for under $7K (what a decent '386 system REALISTICLY costs). Whoops, I forgot, theres that nasty BSD again...;') > But (BIG NUMBERS ONLY) NatSemi will never come forward with such a >thing, will it ? My only hope is all of YOU. The way things look, this is probably true. If the '386 is the hottest selling Unix chip around (recently), its only because NS fumbled the ball... >-- >Piercarlo "Peter" Grandi INET: pcg@cs.aber.ac.uk >Sw.Eng. Group, Dept. of Computer Science UUCP: ...!mcvax!ukc!aber-cs!pcg >UCW, Penglais, Aberystwyth, WALES SY23 3BX (UK) ken seefried iii ...!{akgua, allegra, amd, harpo, hplabs, ken@gatech.edu masscomp, rlgvax, sb1, uf-cgrl, unmvax, ccastks@gitvm1.bitnet ut-ngp, ut-sally}!gatech!ken "Children is more trouble than human beings!" -- Popeye the Sailor
vixie@decwrl.dec.com (Paul Vixie) (11/21/88)
## [...] the only real superiority of BSD over 5.3.2 is the Fast file system, This is debatable, and as Ken says below, it's a religious issue anyway. Not that I'd mind wasting SysV in public for the Nth time, but that's not what this group is about so let's table it. ## and Interactive's version of 386 5.3.2 has it... # # For the record, no it doesn't. Quite right, it doesn't. Similar in name only. # [...] For the price of a MINIMUM, which would be FAR from what an OS # developer would need, '386 based system (16MHz, 2MB, 40-60MB disk, EGA, ISC # 386ix v.2.0 == $5500+) you should be able to dig up a usable 32k based Unix # machine. Heck, Symmetric 375's cost about that much. Not that I particularly recommend the '375, but I agree that anyone who wants to program an OS for the 32k should probably be programming WITH a 32k if it's at all possible. -- Paul Vixie Work: vixie@decwrl.dec.com decwrl!vixie +1 415 853 6600 Play: paul@vixie.sf.ca.us vixie!paul +1 415 864 7013
dtynan@sultra.UUCP (Der Tynan) (11/22/88)
In article <256@aber-cs.UUCP>, pcg@aber-cs.UUCP (Piercarlo Grandi) writes: > >Stop daydreaming Der! a 532 is too expensive, and probably not needed. I >think a 332 would do. The problem with 532 is that it is not only quite >expensive in small quantities, but it is too damn fast. The real cost of a >system is not in the CPU, is in the support around it. Think over it. What support? Do you mean the DMA, memory, or discrete circuitry? As I said in the previous posting, I believe the system should run real fast to memory, but slower to the PC-Bus. I don't agree that a '532 is not needed. The current migration of software (both applications, and OS), is toward Big Time. Sure, the '332 may be just great today, but what about next year? Will we have to start over again? One of the systems I use at home, is a vanilla IBM-PC. It is now totally and completely obsolete. It can barely handle the stuff I put on it. Anyone want to buy an IBM-PC (4.77MHz) for $2,500?? I thought not. My point is, that I want a system that will grow with the technology. Four years ago, I thought 10Mb was more than enough. I seriously thought about removing the 'DELETE' command. Want to guess how much space X-Windows eats? I want a system which, if it won't perform at next-years rates, at least allows me to upgrade. Why should I sell myself short? >In my never humble opinion, why not have a look at the nice at 286 single board >design from Byte, published several months ago ? If such a design had a 32k CPU >instead of a 286. Interesting point. What issue(s) of Byte was that in (I cancelled my subscription long ago)?? >In other words, the aim should be to develop a 386SX not a 386 type of system. Huh? I think I missed something. What is a 386SX? >And as long as it is as compatible as possible with a 286, so much the better. >Would it be possible to plug a 32k based daughterboard in the socket for a 286 >or 386 CPU in an existing AT board ? (no, I am not a jester!) Sure. This could be done, but this is nothing more than a "stop-gap" measure. You'd basically have to give the 32K a "frontal lobotomy". In which case, you end up with a system not that far removed from a '286. The main difference being the lack of segmentation registers. Why bother? This would do you for a while, but what about next year? What about the fact that the AT doesn't *really* support anything over 1Mb? You'd have to hack and hack, and end up with a system that you'd want to replace in a year, anyway. >I am going to buy a 386 box with S5.3.2. Remeber, the only real superiority of >BSD over 5.3.2 is the Fast file system, and Interactive's version of 386 5.3.2 >has it... in BSD you do not get mapped files, streams etc... I am going to I beg to differ. Granted, I've never used Release 3, but I have definitely used S5.2.1, and found it nasty (I'm fighting to avoid using stronger words). BSD also has sockets. Not to mention 'csh', and many more features. However, you're certainly welcome to your opinion. But, don't expect AT&T to continue to support release 3, when 4.0 is in the wings, and don't expect 4.0 to run on your AT box. >use it for developing th next great capability based distributed operating >system. I would for sure use a 32k based cpu instead of the 386 if I could run >UNIX on it. But (BIG NUMBERS ONLY) NatSemi will never come forward with such a >thing, will it ? My only hope is all of YOU. > Piercarlo "Peter" Grandi INET: pcg@cs.aber.ac.uk *WE* are trying to, but we need your support. As for your first point about the price of the '532 chip (or chipset?), call me an optimist, but I believe NSC won't leave us "high and dry". - Der -- 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? ---
ken@gatech.edu (Ken Seefried iii) (11/22/88)
Yet another in my series of 'For The Record' articles...;') In article <2667@sultra.UUCP> dtynan@sultra.UUCP (Der Tynan) writes: >but slower to the PC-Bus. I don't agree that a '532 is not needed. The >current migration of software (both applications, and OS), is toward Big Time. >Sure, the '332 may be just great today, but what about next year? Will we >have to start over again? I belive you have missed the whole boat here. People are talking about building an *inexpensive* yet reasonably powered machine. Next year, my friend, the 32532 will be passe also, thechnology moves pretty damn fast these days. The 532 cost big $$$, the memory to feed the 532 cost big $$$, and the IO capacity to feed the 532 costs, you guessed it, big $$$. Like the National Semi guys have been saying, if you want a 532, talk to Heurikon, they did it right. > Want to guess how much space >X-Windows eats? >rates, at least In the neighborhood of 45MB+ with source...though I haven't 'du'ed release 3 yet. >>In other words, the aim should be to develop a 386SX not a 386 type of system. > >Huh? I think I missed something. What is a 386SX? > The 386SX is a 16-bit data bus version of the 386 (akin to the 8088/8086 pair). The idea is to reduce system cost by reducing the nessesary memory and IO bandwidth. > > Granted, I've never used Release 3, but I have definitely >used S5.2.1, and found it nasty (I'm fighting to avoid using stronger words). Yes...100% >BSD also has sockets. Yes you've obviously never used Release 3. The STREAMS facility, which replaces sockets, is much nicer... > Not to mention 'csh', I think the 3b2 is the only System V.3 system ive seen that didn't come with csh. > and many more features. That certainly goes both ways. As I said, i like BSD, but that is religion, and religion only. Both systems are getting VERY close feature wise. > However, >you're certainly welcome to your opinion. But, don't expect AT&T to continue >to support release 3, when 4.0 is in the wings, and don't expect 4.0 to run >on your AT box. I beg to differ. There is still an awful lot of active support for System V.2, perhaps not from AT&T, but certainly from vendors (when was the last time AT&T gave support on a System V based, say, Silicon Graphics...?) Vendors support Unix. AT&T barely supports the 3b's, much less anyone else. System V.3.x support is NOT going to vapourise for a few years (and then we'll be running Mach or Unix System V.5 or OSF/OS or Clouds (nawwww) or whatever). And as far as Release 4, think yet again...i have been quote preliminary pricing on System V Release 4.0 for the 386. And for the record, Release 4 looks pretty damn nice.... Another point of interest: the 80386, along with the 3b2 and the SPARC are targets for the official AT&T reference ports of future AT&T Unixes, which means that other than device drivers, Unix will compile off the tape for these processors. This is NOT the case for the NS chips, which (at least) will need the memory management code rewritten, along with tons of other kernal stuff. Not slamming the NS chips, just stating the facts.... > >*WE* are trying to, but we need your support. As for your first point about >the price of the '532 chip (or chipset?), call me an optimist, but I believe >NSC won't leave us "high and dry". Perhaps...perhaps a year from now...perhaps not...I'm pushing the '332... > 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? --- ken seefried iii ...!{akgua, allegra, amd, harpo, hplabs, ken@gatech.edu masscomp, rlgvax, sb1, uf-cgrl, unmvax, ccastks@gitvm1.bitnet ut-ngp, ut-sally}!gatech!ken "Children is more trouble than human beings!" -- Popeye the Sailor
stevel@eleazar.dartmouth.edu (Steve Ligett) (11/23/88)
First, some facts. These are prices from NSC distributors. They are "book prices"; there's no use trying to twist someone's arm this early in the game to get special prices when I don't have a PO# to give them. The 532. The designer's kit, at $532, is a deal. Um, I don't have all the facts on this in front of me (left 'em at home, I guess), but you get a chip set w mmu, and fpu, 25MHz parts, I believe. Compare that to single quantity prices for the cpu alone: 20MHz is $750, 30MHz is $975. The 332. Sigh - there doesn't seem to be a designer's kit for this one. The cpu costs $177.45 (15 MHz). The databook mentions a 10MHz part, but the distributors don't have it. The 032. The designer's kit is $75. It's 6 MHz. (Please correct me if I'm wrong on that!) The 10 MHz cpu alone costs $88.70. Pc board costs. I called my PC board maker for better costs than I gave Der. A 4-layer setup costs $400 or more. It doesn't go up linearly with board size. So, a 8.5" by 12" board setup might only be $500 or $600, depending on number of holes, assuming that traces are 12 mil or larger, and pads aren't too small in relation to holes, and that inside layers are just power and ground. Then, boards cost $1.50 per square inch. That could go down some in large quantity, but we're talking small quantity here, and this is a prototype house (i.e. small quantity) I deal with. So, maybe $180 for a board, in quantities of 20 (including paying off the setup). You line up 100 people who'll buy a board and it'll be cheaper. Memory costs. If Hitachi delivers, I could sell memory to this group for less than $300 per megabyte (100 ns, 1 meg by 9 simms). I don't mean that as a promise, or as an advertisement. Just to give you an idea of memory costs. Another possibility is to collect 256k simm "pulls" - simms taken from Macs (256k by 8, 150 ns), when they are upgraded to 1 meg simms. Those should be pretty cheap, but I've never tried to buy such an inventory, so my guess of $40 per megabyte (4 simms) is just that. I'll go reread the articles I've saved before going into my opinions. Steve Ligett steve.ligett@dartmouth.edu or (decvax harvard linus true)!dartvax!steve.ligett
pcg@aber-cs.UUCP (Piercarlo Grandi) (11/23/88)
In article <924@bacchus.dec.com> vixie@decwrl.dec.com (Paul Vixie) writes:
[ this piece actually from somebody else, cited by vixie: ]
# [...] For the price of a MINIMUM, which would be FAR from what an OS
# developer would need, '386 based system (16MHz, 2MB, 40-60MB disk, EGA,
# ISC 386ix v.2.0 == $5500+) you should be able to dig up a usable 32k
# based Unix machine.
[ vixie's comments: ]
Heck, Symmetric 375's cost about that much. Not that I particularly
recommend the '375, but I agree that anyone who wants to program an OS
for the 32k should probably be programming WITH a 32k if it's at all
possible.
Ahah. Were it so easy! Actually, the 386 system I am buying is somewhat
larger than the MINIMUM described above, and yes, with all discounts
factored in will be just a tad over $5500 at the end (my own money...
will have to dig DEEP into my relatively shallow pockets, so price, and
price/performance are dominating issues).
There are several considerations here:
I am going to build a portable OS; so initial platform is not so
important, it will not be a 32k specific system.
The 386 is not that terrible for os developers. It is far less
complicated than it seems, and the os related functions are quite
straightforward and well designed, even advanced. This is true,
conceded, only if you do not care at all about backwards
compatibility, and the compiler is already done :-), but I don't
think that the main attraction of an object oriented, distributed
capability nucleus, would be in MSDOS/8088 compatibility :-).
However I would really rather use the 32k because it is obviously
the easiest architecture for which do the machine dependent parts
of an os, and you can easily imagine that I could dispense with any
tougher challenge.
Actually the machine is for developing a nucleus, not an entire os;
the rest will be the extensive toolset developed by the FSF. In
other words, wait a couple of years and you will have the luxury of
a choice between *two* GNU kernels. This also means that I have to
choose a system on which g++ is already available. Well, the 386 is
the last to be considered by the FSF people, but something is
better than nothing.
To me the MINIMUM machine listed above looks enourmous; I used to
run 5 users on a 256kb 11/34. Also, I am not a member of the SV.4
team! :-). To me os design is about elegance, frugality, etc...
See SP&E, August 1979, for something I have as a model.
The MINIMUM 386 system configured above is vastly superior in
performance etc... to a symmetric (by several times overall, and by
an order of magnitude in CPU speed) or to anything I could find, on
this side of the Pond, for the same range of price, in the 32k (or
68k for that matter) worlds, not to mention the difficulties of
getting a decent UNIX, virtual memory, peripherals, etc... Hint:
VME bus systems are EXPENSIVE.
I would be prepared to expend some more money, for some less
performance, for the priviledge of working with a 32k system.
My site lost a few articles (partition full...) yesterday and the day
before, so I do not know if anybody took up the other point I made:
Would it be possible (I mean convenient, reasonable, good)
to build either:
a single board 32k CPU like the Ciarcia's one, for use on one
of the now numerous AT clones that do not have a motherboard,
but only an AT bus backplane ?
a daughterboard by way of which one could remove a 286 or a 386
from an AT clone and run in its place a 32K ?
If either of these were possible, and there were a suitable UNIX
implementation, oen could well reap the advantages of being able to buy
*cheap* in the AT clone market and the AT components market (a disk
controller board for an AT clone, for not much worse performance, costs,
retail, one tenth of a board for a VME machine, and one fifth that of a
board for a Q-BUS machine, and so on...).
--
Piercarlo "Peter" Grandi INET: pcg@cs.aber.ac.uk
Sw.Eng. Group, Dept. of Computer Science UUCP: ...!mcvax!ukc!aber-cs!pcg
UCW, Penglais, Aberystwyth, WALES SY23 3BZ (UK)
woods@gpu.utcs.toronto.edu (Greg Woods) (11/23/88)
In article <2667@sultra.UUCP> dtynan@sultra.UUCP (Der Tynan) writes: > In article <256@aber-cs.UUCP>, pcg@aber-cs.UUCP (Piercarlo Grandi) writes: > > I am going to buy a 386 box with S5.3.2. Remeber, the only real superiority of > > BSD over 5.3.2 is the Fast file system, and Interactive's version of 386 5.3.2 > > has it... in BSD you do not get mapped files, streams etc... I am going to > > I beg to differ. Granted, I've never used Release 3, but I have definitely > used S5.2.1, and found it nasty (I'm fighting to avoid using stronger words). > BSD also has sockets. Not to mention 'csh', and many more features. However, > you're certainly welcome to your opinion. But, don't expect AT&T to continue > to support release 3, when 4.0 is in the wings, and don't expect 4.0 to run > on your AT box. Ok, here comes another blast furnace.... I'm a SysVr3 fan. Let's get that out in the open... The only significant kernel features that I miss in SysVr3 are job control, and support for long filenames. EVERYTHING else you should ever need in a kernel is in SysVr3, and often implemented in a much more elegant fashion than the BSD equivalent. Look at STREAMS for example.... Sockets can be built on TLI, and PTY's are just device drivers (and can even be implemented using STREAMS). Admittedly, poll() should work on any file (ala select()), but I can live with it as-is. With the recent declaration that BSD only code will be free, any little feature you miss in the utilities can often be added. As for csh, it's too buggy and hard to use in face of ksh, which is easy enough to find. The only BSD feature I can argue either way for is the symbolic link. If it was trimmed to only allow directory links I could live with it. -- Greg Woods. {utgpu,lsuc!gate,ontmoh}!woods, woods@{gpu.utcs.Toronto.EDU,utorgpu.BITNET} 1-416-443-1734 [h], 1-416-595-5425 [w] LOCATION: Toronto, Ontario, Canada
dtynan@sultra.UUCP (Der Tynan) (11/23/88)
In article <17658@gatech.edu>, ken@gatech.edu (Ken Seefried iii) writes: > > In article <2667@sultra.UUCP> dtynan@sultra.UUCP (Der Tynan) writes: > >Sure, the '332 may be just great today, but what about next year? Will we > >have to start over again? > > I belive you have missed the whole boat here. People are talking about > building an *inexpensive* yet reasonably powered machine. Hmm. Amusing comment at best. Check the original reference. *I* was the one pushing for an *inexpensive* machine. I *still* am. But, one has to look at the price/performance curve as well. On top of that, if there is already an organization in place (the '532 Manifesto) to build a '532 system, why go in competition to that effort. What I would like to see, is an overall vote or consensus of opinion as to a CPU type. I am open to comment. I happen to have a 6MHz 32032 chipset in my desk, which is slowly ageing. > Next year, my > friend, the 32532 will be passe also, thechnology moves pretty damn fast these > days. Maybe the thing to do is just give up then. I mean, if we can't expect to be reasonably ahead of the game, let's just stick to our 4.77MHz Intel stuff :-) I doubt very much if the '532 will be obsolete next year. This is a gross exaggeration. Ask yourself when was the 68K first designed, and furthermore, what was Moto's best year for shipping the things. I would be inclined to agree that the 32016 won't be worth much next year, but come on. The '532? Sequent is only now shipping their '532 CPU cards. > The 532 cost big $$$, the memory to feed the 532 cost big $$$, and the > IO capacity to feed the 532 costs, you guessed it, big $$$. Like the National > Semi guys have been saying, if you want a 532, talk to Heurikon, they did it > right. This is not true. It may be safe to say that the '532 chip is expensive. However, what kind of extortion can we pull on NSC? On the other hand, whether you use a Mips processor or a Z80, the DRAM will be 150ns SIMMS. This is fixed. Nothing to do with the CPU speed. If you want to avoid wait-states, just use the "paging scheme" hinted at by Steve Wilson. This should give you the 50ns eaten by the MMU at least. And this is without using 15ns cache chips. Furthermore, since when did I/O feed the CPU. The '332 and '532 (and maybe even the '032 -- Steve??) allow dynamic resizing of the bus. In this way, the CPU could access a vanilla IBM-PC bus, without major surgery. > The 386SX is a 16-bit data bus version of the 386 (akin to the 8088/8086 > pair). The idea is to reduce system cost by reducing the nessesary memory and > IO bandwidth. See above. The NSC stuff can support smaller bus sizing, without needing to change the chip. At last, a chip designer with an IQ :-) > That certainly goes both ways. As I said, i like BSD, but that is religion, > and religion only. Both systems are getting VERY close feature wise. You're right. Let's leave the SysV/BSD thing to religious fanatics. Personally, over the long haul, I think Mach is the way to go, because the kernel threatens to be public domain (or equiv). It will still need utilities, but that's what GNU is for. > Another point of interest: the 80386, along with the 3b2 and the SPARC are > targets for the official AT&T reference ports of future AT&T Unixes, which > means that other than device drivers, Unix will compile off the tape for these > processors. This is NOT the case for the NS chips, which (at least) will need > the memory management code rewritten, along with tons of other kernal stuff. > Not slamming the NS chips, just stating the facts.... > ken seefried iii ...!{akgua, allegra, amd, harpo, hplabs, Again, this is the beauty of Mach, that the VM stuff has been designed to be easily ported. As for the AT&T support, you have been misled. I recently attended an AT&T lecture on R4.0, and the three processors listed above, are just people who have ABI's. Motorola will soon be added to the list. If National wants to, they can pay the money and join up. Personally, I don't care. I'd rather not use a brain-dead system just because I didn't have to write any code when I first took it out of the box. Most of us are talking about a system for the "long-haul". Furthermore, the notion that we'll upgrade the OS every time AT&T releases a new version is nice, but unrealistic. - Der -- 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? ---
stevew@nsc.nsc.com (Steve Wilson) (11/24/88)
In article <2677@sultra.UUCP> dtynan@sultra.UUCP (Der Tynan) writes: >This is fixed. Nothing to do with the CPU speed. If you want to avoid >wait-states, just use the "paging scheme" hinted at by Steve Wilson. This >should give you the 50ns eaten by the MMU at least. And this is without >using 15ns cache chips. Furthermore, since when did I/O feed the CPU. >The '332 and '532 (and maybe even the '032 -- Steve??) allow dynamic >resizing of the bus. The 32032 doesn't support dynamic bus sizing. As for the page mode scheme see George Scolaro's recent posting for more details. (Where do you think I got the idea from ;-) Though I still think the 50ns number is ok (George, I really did take address drivers and data bus drivers into account when I came up with that number...) Trying to support burst fills on a 532 with 50ns SRAMS even at 20 Mhz would be an expensive proposition though. There are ways to do it, but managing it gets abit tricky, and how you interface it to your DRAM system is also a consideration. Another detail of the cache is that you couldn't possibly use 50ns SRAMS to implement the TAG. You'd probably want to use some of the faster TAG chips. These guys aren't cheap either. Steve Wilson National Semiconductor
bga@raspail.UUCP (Bruce Albrecht) (11/24/88)
In article <2677@sultra.UUCP>, dtynan@sultra.UUCP (Der Tynan) writes: > Hmm. Amusing comment at best. Check the original reference. *I* was the one > pushing for an *inexpensive* machine. I *still* am. But, one has to look at > the price/performance curve as well. On top of that, if there is already an > organization in place (the '532 Manifesto) to build a '532 system, why go > in competition to that effort. What I would like to see, is an overall vote > or consensus of opinion as to a CPU type. I am open to comment. I happen > to have a 6MHz 32032 chipset in my desk, which is slowly ageing. If we purchase 532DKs, we can get the '532 and all the necessary support chips for $532. It's a 25Mhz '532, which would need approximately 80-100 nsec RAM for zero wait states. If the code is tight, we could have a lot run within the internal cache, which lessens the need for zero-wait state memory. > > The 532 cost big $$$, the memory to feed the 532 cost big $$$, and the > > IO capacity to feed the 532 costs, you guessed it, big $$$. Like the National > > Semi guys have been saying, if you want a 532, talk to Heurikon, they did it > > right. > > This is not true. It may be safe to say that the '532 chip is expensive. > However, what kind of extortion can we pull on NSC? On the other hand, > whether you use a Mips processor or a Z80, the DRAM will be 150ns SIMMS. > This is fixed. Nothing to do with the CPU speed. If you want to avoid > wait-states, just use the "paging scheme" hinted at by Steve Wilson. This > should give you the 50ns eaten by the MMU at least. And this is without > using 15ns cache chips. Furthermore, since when did I/O feed the CPU. > The '332 and '532 (and maybe even the '032 -- Steve??) allow dynamic > resizing of the bus. In this way, the CPU could access a vanilla IBM-PC > bus, without major surgery. Do you lose 50 ns with the '532's internal MMU? I won't be getting a copy of the new databook with the '532 specs for a couple weeks, so I can't check it out. My local 32k specialist said that a 25 Mhz 532 should be able to get by with fairly slow speed RAM. If we make a decision to use the 532, we can get the designer's kit for $532 through February. A well designed board should allow either you to start with 256K or 1M SIMMs, so the people who can afford more can buy them. If we were interested in providing ECC, we could use the NS ECC chip, and use 5 (256K/1M)x8 SIMMs, and have a spare bit. I, for one, would not even consider building a 32016/32 system, even though it was cheap. I think we could build a high-performance '532 system for about $1k more than the ultra-low end machine, depending on how much memory and other "frills" like big disks get added. If the box with a '532, 1M of memory, and 40M disk drive costs about $2k, then we've got a system competitive with the '386 machines in performance and '286 machines in price, and for half the cost of a Mac II. Bruce
pcg@aber-cs.UUCP (Piercarlo Grandi) (11/25/88)
I am very pleased to see all this discussion, here are some comments: ken@gatech.UUCP (Ken Seefried iii) writes: Yet another in my series of 'For The Record' articles...;') In article <2667@sultra.UUCP> dtynan@sultra.UUCP (Der Tynan) writes: but slower to the PC-Bus. I don't agree that a '532 is not needed. The current migration of software (both applications, and OS), is toward Big Time. Sure, the '332 may be just great today, but what about next year? Will we have to start over again? I belive you have missed the whole boat here. People are talking about building an *inexpensive* yet reasonably powered machine. Next year, my friend, the 32532 will be passe also, thechnology moves pretty damn fast these days. The 532 cost big $$$, the memory to feed the 532 cost big $$$, and the IO capacity to feed the 532 costs, you guessed it, big $$$. Like the National Semi guys have been saying, if you want a 532, talk to Heurikon, they did it right. Exactly! that was my point. A 10-15 MIPS machine is way beyond what we can currently afford (me, at least). Also, if you design your system carefully, switching over to 532 may just entail swapping a daughterboard... In other words, the aim should be to develop a 386SX not a 386 type of system. Huh? I think I missed something. What is a 386SX? The 386SX is a 16-bit data bus version of the 386 (akin to the 8088/8086 pair). The idea is to reduce system cost by reducing the nessesary memory and IO bandwidth. You get a lot of the benefit of a 386, and the total system cost is much lower. The CPU internally is 32 bits, but externally 24/16 bits. You can run a 332 like that, if you want. However, you're certainly welcome to your opinion. But, don't expect AT&T to continue to support release 3, when 4.0 is in the wings, and don't expect 4.0 to run on your AT box. I beg to differ. There is still an awful lot of active support for System V.2, perhaps not from AT&T, but certainly from vendors (when was the last time AT&T gave support on a System V based, say, Silicon Graphics...?) Vendors support Unix. AT&T barely supports the 3b's, much less anyone else. Actually, you probably will not be able to afford 5.4 very easily... There are ominous rumours. System V.3.x support is NOT going to vapourise for a few years (and then we'll be running Mach or Unix System V.5 or OSF/OS or Clouds (nawwww) or whatever). Well, as I said, just hold your breath until I send to the FSF the full NONE kernel (so far mostly a paper design, unfortunately :-). And as far as Release 4, think yet again...i have been quote preliminary pricing on System V Release 4.0 for the 386. And for the record, Release 4 looks pretty damn nice.... Please send me mail about it. There was another message that said that it was going to cost A LOT. Another point of interest: the 80386, along with the 3b2 and the SPARC are targets for the official AT&T reference ports of future AT&T Unixes, which means that other than device drivers, Unix will compile off the tape for these processors. This is NOT the case for the NS chips, which (at least) will need the memory management code rewritten, along with tons of other kernal stuff. Not slamming the NS chips, just stating the facts.... Sad truth, but truth. Moreover the 386 will be the second machine in order of importance, and it is for AT class machines, not multibus based ones, starting with 3.2. *WE* are trying to, but we need your support. As for your first point about the price of the '532 chip (or chipset?), call me an optimist, but I believe NSC won't leave us "high and dry". Perhaps...perhaps a year from now...perhaps not...I'm pushing the '332... OK, OK! dlr@daver.UUCP (Dave Rand) writes: In article <17659@gatech.edu> ken@gatech.UUCP (Ken Seefried iii) writes: In article <8084@daver.UUCP> dlr@daver.UUCP (Dave Rand) writes: Ok. Everyone WANTS BSD. I'm game to do the port. How do I get it? I know how to get a Binary license for Sys V.3 (and the binary is cheap). How can ^^^^^^ I get one for BSD? Source is nice, but usually too expensive. So your going to do a BSD port to the 32k? And from the binary license, the source being to expensive... [stuff deleted] Yes, I did say BINARY. Not source. There are several companies offering NS32K System V Binaries (ZAIAZ being one of them). There are at least two ports of BSD to the NS32K (one by the U of Toronto, I think, and the other by Symmetric (sp?)). HCR and NatSemi. So - back to the question. I know how to get a NS32K Binary for System V. How do I get a NS32K Binary for BSD? Phone to NatSemi. Well, not really. They have ported BOTH BSD and Sys5 to their development machines, and are prepared to supply you (for a price...) the sources, if you are already a UNIX source licensee. By the way, the paper they had at some old USENIX on their BSD port was really interesting. Great guys. Neat job. meo@stiatl.UUCP (Miles O'Neal) writes: In article <7648@pasteur.Berkeley.EDU> heppell@sim.berkeley.edu (Kevin G. Heppell) writes: (some VERY sensible suggestions) Obviously, I concur with Kevin. I don't care whether it's Nu-Bus or not, but it seems as good a suggestion as any, and a LOT better than the PC buses, altho the new standard from Compaq et al may be worthwhile. But since the ONLY reasons I will be on a PC-derivative are for developing PC software (and I VERY seldom do this!), or for using existing business products while I wait for someone to do them in UNIX (or I do them), I really have no use for anything else to be PC-compatible. That's what RS-232, Ethernet, and the like are for! and bga@raspail.UUCP (Bruce Albrecht) writes: In article <7648@pasteur.Berkeley.EDU>, heppell@sim.berkeley.edu (Kevin G. Heppell) writes: 3) The PC bus architecture is just as slow and awkward as the rest of the system. NuBus, at least, is an IEEE standard, supports 32 bits, and has a connector which can perform at reasonable speeds. From what I've seen of the inside of a Mac II, the AT-chassis form factor can be met with NuBus cards. I don't have any problems with using NuBus. Whose version would we use, since I believe Apple and NeXT have both modified it. I have a problem. Tipically AT bus components cost a fraction of the same things for other standards. Just look at the price differential with components for the MCA, the MacIntosh, the MacII, not to speak of the VMEBUS. Thanks Taiwan! Of course the AT bus is almost hopeless for memory access, but a private ememory bus is ok; I would have no objection to using the AT bus only for I/O (as 386 machine do, essentially, and just like the QBUS is used today). An alternative would be a machine with ONLY a memory bus, and all peripherals accessed via SCSI. Neat solution, but we are starting to see non disk, non tape SCSI peripherals (serial and ethernet interfaces) only now, in the MacIntosh market, and compare the prices with AT bus components, and width of choice. The MacIntosh Plus after all is just a machine like that, and that is not encouraging. Since the 32k series is supposed to be upwardly compatible, it should not matter whether or not the on-board 032 or an added 332 or 532 is doing the work. An add-in card will keep initial costs down, allow for easy up-grade, and provide for an I/O processor with the high-perf one. For those who _really_ want a 532 based motherboard, consider this: you will get a much larger user base through the method I have suggested, which makes software and hardware support much cheaper. And by designing in the upgrade, the initial $1000 outlay doesn't disappear when you go up a class. See below for software considerations. I don't think we should base it on an 016 or 032. That's like building a 68k system on a 010. Sure it's cheaper, but it's also going to be slower and need an upgrade all that much sooner. Also, if you use the 082 MMU, you'll probably want to get a '382 for an upgrade to the 332/532, but you may not need to get a new 382 for an upgrade of the 332 to 532. Also, the 082 and 382 are programmed differently, so I think it would be better if we only used one. I think there may be a number of people who would not be interested in this design if it uses the 032. If I had a choice between a home-brew 032, and a commercial 68030 (i.e., Atari or Commodore) for $3-4k, I'd probably go for the commercial system, but a $1500-2500 32332 or 32532 system would be my first choice. Agreed. 032 and 016 are backwards. Too slow; on the other hand we can afford a 386 level machine, probably not a R2000 level one. If upgrading were possible, so much the better. I'd pay more than $2000 for a full 32332 system. I would pay about the same that I'd pay for a 386 system. woods@gpu.utcs.Toronto.EDU (Greg Woods) writes: In article <2667@sultra.UUCP> dtynan@sultra.UUCP (Der Tynan) writes: In article <256@aber-cs.UUCP>, pcg@aber-cs.UUCP (Piercarlo Grandi) writes: I am going to buy a 386 box with S5.3.2. Remeber, the only real superiority of BSD over 5.3.2 is the Fast file system, and Interactive's version of 386 5.3.2 has it... in BSD you do not get mapped files, streams etc... I am going to I beg to differ. Granted, I've never used Release 3, but I have definitely used S5.2.1, and found it nasty (I'm fighting to avoid using stronger words). BSD also has sockets. Not to mention 'csh', and many more features. However, you're certainly welcome to your opinion. But, don't expect AT&T to continue to support release 3, when 4.0 is in the wings, and don't expect 4.0 to run on your AT box. Ok, here comes another blast furnace.... I'm a SysVr3 fan. Let's get that out in the open... I am not a Sys5 fan. I am not a BSD fan. I find all Unixes except the PDP ones distasteful. Long life V7! Long life 2BSD! Also I find UNIX a dead horse, born dead. But this is not the point. Please do not comment on the above paragraph; the point I want to make, as I have said, is that BOTH Sys5 and BSd have their pluses, compared to the alternatives (OS/9, Minix ?). Until we get GNU (nee Mach, nee Accent) from FSF, the best we get is UNIX, rare, medium or well done. We are lucky enough that NatSemi can give us either version, even if I'd rather have Sys5, because it is not too bad, and is the same I get on a 386, and NatSemi probably prefers to expend effort on Sys5 than BSD now. The real challenge is to get a machine off the air and make NatSemi or somebody else supply a cheap binary UNIX for it. This will be the trickiest part. -- Piercarlo "Peter" Grandi INET: pcg@cs.aber.ac.uk Sw.Eng. Group, Dept. of Computer Science UUCP: ...!mcvax!ukc!aber-cs!pcg UCW, Penglais, Aberystwyth, WALES SY23 3BZ (UK)