Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Path: utzoo!utgpu!water!watmath!clyde!rutgers!sunybcs!bingvaxu!leah!uwmcsd1!uwmacc!hobbes!root From: root@hobbes.UUCP Newsgroups: comp.sys.ibm.pc Subject: Re: Intel Microprocessors (History) Message-ID: <182@hobbes.UUCP> Date: Tue, 18-Aug-87 02:05:29 EDT Article-I.D.: hobbes.182 Posted: Tue Aug 18 02:05:29 1987 Date-Received: Wed, 19-Aug-87 07:11:27 EDT References: <1112@lznv.ATT.COM> <399@aucs.UUCP> <3225@cucca.columbia.edu> <892@looking.UUCP> <79@LBI.UUCP> Reply-To: root@hobbes.UUCP (John Plocher) Followup-To: comp.sys.ibm.pc Organization: U of Wisconsin - Madison Spanish Department Lines: 119 +---- Robert Cain writes the following in article <79@LBI.UUCP> ---- | Brad Templeton writes: | > In the real world, there are lots of other constraints in choosing a | > chip than what the architecture looks like to software. AMEN! | The one headache we could all have lived without, is segment registers, | and nothing Intel can say or do will ever change that fact. Hindsite is always best. Granted, segment registers can be a pain, but AT THE TIME, they were very useful! What were *you* doing with micros in September of 1980 (when the PC was starting design life)? CP/M was THE OS for Micros, 64K was HUGE - I managed several 6 user systems which ran NorthStar BASIC with 2 floppy disk drives, a Z80, and 64K! Actually, it only had 56K, and of that, 8K was unused by the OS for anything! The OS was so configured as to allow each user a 4 or 5K workspace and rarely did the students ever run out of room! (How many of you can write a program which solves 5 linear equations with 5 unknowns while not using more than 5K? Including input and output routines? I just tried it with C on a Gould - the executable was 10K... :-) What has this to do with segments? CP/M 3.0 and MP/M-II both used bank selected memory (a 64K chunk was made up of (say) 56K of "unique" memory and 8K of "shared" memory, sort of like the LIM memory boards in the PC do today). With this very hardware dependent method, the OS was able to use as much memory as was needed, just not very easily. (ie, to load a program into bank 5, one had to switch to bank 0 where the OS was located, read in the desired program, and copy 4K of it to a data buffer in the 8K of "shared" memory. When this was done, you had to switch to bank 5, copy the 4K to the right place in this bank, and go back to bank 0 to repeat this process until everything was copied. Along came the 8086 family. WOW! DIRECT access to 1 Meg of memory! No more bank switching. Compared to the current Minis out there (the pdp-11 family; Vaxen were still very new) this was hot stuff! The pdp family was limited to 256K, 512K, or 1Mb for various machines smaller than the 11/70. Unix kernals were running about 80-100K in size (v6, and v7), so this CPU was right in the middle of things. The development machines of the time were mostly S-100 boxes. The reliable ones used static memory. 64K cost $795! 256Kb cost $1695. The dynamic memory boards had timing problems when used with anything other than 2 to 4Mhz Z80s, and they cost $350 per 64K ($1300/256K). A full blown 1 Meg 8086 system with two 8" floppies and a (gasp) 10MB hard disk would set you back about $12,000. I know, I built a few. The 10Mb Morrow hard disks (19" RETMA Rack mount, no less) were going for the low price of $2,950! (Prices taken from BYTE, Vol7, No 1 January 1982) Where were the others? Perusing the Bible of the time, Adam Osbourn's series "An Introduction to Microcomputers", Volume 2, "Some Real Microprocessors", 1980 update, and culling the recent trafic on comp.sys.{intel,motorola,nsc}, and the Byte's from Oct to Dec 1981 I find that: Fairchild had the F8, but it was geared for the imbedded processor market. National had the SC/MP, but it, too was suited for only simple applications. National's new CPU was nowhere close to ready - the 16032 wasn't readily avaliable till 1982, and then it had so many bugs in it that the only thing offered for it was a Forth OS kernal. Moto's new 68K was looking real good, but the peripheral chips were nonexistant. It wasn't even listed in the Osbourn book! On the other hand, Dual was advertizing a S-100 68K system with 256K and v7 Unix for $8295 (PLUS hard disk, monitors...) Moto was marketing their chip in the "mini" nitch, and didn't want to be bothered by these hobbiests and their micros, after all, wasn't the good old reliable 6800 good enuf for them? (The MC6800 is a contemporary of intel's 8080a. In the Jaemco/ACP/Jade.. ads in Byte you could get 6 MHz 8088s for $40; the ads had prices for 6800s but NO mention of the 68K! This in 1981, after the PC had been released!) Apple had shown that the 6502 from MOS Tech was a good processor; MOS had a range of peripheral chips avaliable as well as TEN different versions of the basic CPU! But if IBM had gone with the MOS 6502 series, they would have been the ones playing catch up and johnny come lately. Not a position IBM ever wants to admit publicly. | In fact IBM chose Intel because they were a good takeover target | at the time. Where are your facts to back this up? Or is this an armchair assumption? Sure, IBM bought intel stock; wouldn't you when a sizable percent of your income depended on their product? | Not because Intel had any handle on the marketplace at large. Intel didn't have a handle...? You're kidding, right? EVERY CP/M system out there (8080, 8085, or Z80) has intel parts in it! Intel CPUs, DMA chips, USARTs, DRAM controllers, Floppy controllers, and other "glue". Many of the new 68K boxes (CompuTex...) were built on Multibus (an intel bus) cards! They had their fingers into almost everyones pie, and they knew it. | In one fell swoop, IBM released an inferior machine that became a | standard Sure, by TODAYS standards it is inferior, but in 1980, it was a step above the other stuff that was AVALIABLE. | Oh BIG BLUE if only you would have known the Frankinstein you created! They have sold much more than 5 MILLION of these beasts - Figure a standard 35% markup from wholesale to retail, and you still get several billion dollars income! Quita a profitable Frankenstein! And that's what drives IBM, profits! Not state of the art, not segments, not CPU speed, but $$$$. Sorry about SHOUTING so much, but let's try to realize that there is a BIG difference between 1987 and 1980 in terms of computer power and knowledge. In it's time, Multics (which was a segmented OS) was in the forefront of the OS research, and segments were A Good Thing. They looked promising and were an easy extention from the 8080 for intel. If we remember that hardware always lags the current state of the art computer programming *theory* then we realize that the current crop of CPUs (68020, 80386, 32332) is mirroring the theory of yesterday also. What's on the horizon? The 68030, 80486, and the 32532. And by the time we get them, they too will be "old theory". But without them, we wouldn't get anywhere! Here's to tomorrow's CPUs; let's keep on improving them! -- John Plocher uwvax!geowhiz!uwspan!plocher plocher%uwspan.UUCP@uwvax.CS.WISC.EDU