gbs@stratus.UUCP (George B. Smith) (11/11/88)
In article <16761@agate.BERKELEY.EDU> c60a-1cu@web-1e.berkeley.edu (Drew Dean) writes: > /* lots of stuff about Mac application programs deleted */ I agree that the Mac is hard to beat for out-of-the-box, easy-to-use apps like word processing, grphic production, DTP, etc, etc. I think that what this group is all about is do-it-yourself. Building your own hardware, having source to the OS and compiler, learning and stretching, thats what I am hearing from this group. I am not saying that the Mac doesn't provide this, its just that the 32k looks like a fun way to go on this path. > Yes, I would like a 32 bit Unix box,... >but it better be under $1500....This eliminates the '532, as >I just don't think it can be done....Now as to what I'd do: > 1) Use a cheaper processor (ie 32332 I was wondering about this today. We need some help from George Grenley on this one. With the 332 computing cluster, the following parts are pretty much required for a good system: - 332 cpu - 382 mmu - 381 fpu - 201 tcu - 202 icu while with the 532 computing cluster, you need: - 532 cpu/mmu - 381 fpu - 202 icu Now, the question is, for low volume situtations, how do the clusters compare in price, especially considering the 532 at 30 MHz is a good 5 times faster than the 332 at 15 MHz. So if all the other system stuff is the same (memory, block i/o, character i/o, power, size, etc, etc,) how much does the cluster make a difference to the cost? > 2) Put 2Mb RAM (of 256K chips) on the motherboard, and provide for > expansion to 8Mb via 1M chips. There is an implication in this point that I want to bring out. It is assuming the use of a motherboard like that of the IBM PC system board. I like this configuration better than the IBM PC i/o channel card like Opus and Ziaz. Of course the size of the motherboard should be that of an XT or AT system board. It would allow us to take advantage of very inexpensive cases, power supplies, disks, floppys, etc, etc. The lastest issue of Micro Cornucopia has an article about a 68000 board that uses exactly this strategy and is very attractively priced. As far as the memory configuration, I think we should provide SIMM sockets for up to 8 Meg of RAM. The user can then put in as much memory as they can afford. The latest prices here in Silicon Valley for 1 Meg SIMMs for the Mac are around $340 per Meg, not too bad and it should get better. > 3) Put a SCSI port on the thing, as SCSI drives are getting cheap. Another choice I like. Put a SCSI port right on the motherboard. SCSI chips are getting very powerful, easy to interface to, and inexpensive. This makes the configuration of the block i/o device standard so the software/firmware will be more manageable. > 4) Leave the user the option of using an existing computer/terminal > as the system console, or using a PC clone keyboard and a VGA > monochrome or NEC Multisync GS monitor (ie. 640 * 480 or 800 * 600 > monochrome graphics on a 14" screen). I would like a keyboard port right there on the MB just like true blue. Buy any clone keyboard you like, $20 at any computer swap meet for the el cheapo brand. Have a PC compatible i/o slot for the display adapter. The slot can be made to be *hardware* compatible while the drivers at the system software level will have to be unique, unless we can get the 32k to run the 8088 code on ega and vga cards :-). For herc clone cards or the monochrome display adapter it would be trivial. And then the monitor can be whatever is appropriate for the chosen adapter. Again, for the low cost solution, monochome monitors are about $70 here. > 5) Get a solid Unix on this system...If GNU arrives in reasonable > working condition, that would be great...However, GCC still has > some bugs (read gnu.gcc.bugs), and I think it will take another > 6 months to 1 year for GCC to stablize, and about the same time > for the GNU kernel (even if it's Mach) to become really usable. > If GNU isn't there, National should step in by offering cheap > binary licenses (4.3 BSD would be WONDERFUL, but I'll take Sys V). This is where is gets sticky. I would like to see whatever choice that is made include an OS where the source is available for all to hack, study, learn, have fun. That pretty much leaves out anything derived from AT&T UNIX, i.e. any release of BSD or System V. GNU would be a definite possibility someday and we shouldn't make any decision what would preclude it from running on the box. I just don't think the kernel is is that close. We could use the compiler though. Ron Guilmette of National has GCC 1.30 producing 32k code so we may be able to use that as a boot strap. As for the os, file system, networking code, I can't see any low cost solution other than MINIX. I know that there has already been a port done to the 32k by more than one group. Using GCC on an appropriate host, we could boot strap MINIX to the box without too trouble (I think). >National -- tell the MBA's that you'll be getting a whole bunch of EECS >students across the nation familiar with NS products, and look how well >Apple's done in the last couple of years :-) Even in this group, I think >it's safe to say that the NS32xxx series has not been a big commercial >success for National Semi, although it does have technical merit. >Conventional marketing techniques haven't worked very well, so why don't you >try something a little different.... > >Drew Dean Believe me, I have said these *exact* words (well awfully close) to National management and marketing. Know what its like to talk to a wall? Their response is always something mumbled like how many orders *this* quarter, we can't support them, not a big enough market, not enough profit, ad nauseum. A *lot* of people within National felt as you do but not the ones with the power. But that's OK, an enthusiastic user group can do a better job and have more fun anyway. George B. Smith Stratus Computer, Inc gbs@stratus.stratus.com
rfg@nsc.nsc.com (Ron Guilmette) (11/17/88)
In article <1069@stratus.UUCP> gbs@stratus.UUCP (George B. Smith) writes: >We could use the compiler though. Ron Guilmette of National has GCC 1.30 >producing 32k code so we may be able to use that as a boot strap. Dear George, There no way I can seriously take credit for this. The initial port for the 32000 series instruction set was done by Michael Tiemann, formerly of MCC, now of Stanford U. (the same wonderful fellow who brought you G++) and was brought up on the Sequent-Balance and on the Encore-Multimax before I ever even saw the code (let alone tweeked it). Also, the initial port to Genix was done by a fellow named Jay Stein, who is a member of some computer club back east. All I really did was to keep re-porting the various versions to Genix (starting at about version 1.22) and I reported the bugs & fixes I found along the way back to FSF and to Dr. Stallman. I did fix some rather irritating bugs which were, for the most part, only the results of various bugs in the environment and tools (i.e. Genix & GNX) which I used to do the bootstraping. Most notable was the BUG in Genix (yes George, it is a bug!) where the original kernel implementors failed to save/restore *all* of the registers on context switches. Specifically, I had to redo the GCC code generation to avoid the use of the SB register. Also, I had to get GCC to output ".globl" assembler declarations for each external name which actually gets *referenced* in the source program being compiled. This was needed because of an assembler bug (when using -g). These fixes (and the C_EFCN workaround for the loader bug) are neither needed nor useful if you are NOT working on Genix and/or using GNX tools. Therefore, as I have said, the real credit goes mostly to M. Tiemann. -- Ron Guilmette National SemiConductor, 1135 Kern Ave. M/S 7C-266; Sunnyvale, CA 94086 Internet: rfg@nsc.nsc.com or amdahl!nsc!rfg@ames.arc.nasa.gov Uucp: ...{pyramid,sun,amdahl,apple}!nsc!rfg