Issue # 10 DTACK GROUNDED Newsletter - June 1982
HERE IS OUR ESTIMATE OF THE RELATIVE USAGE of various high level languages among those owners of Apples and Pets who program:
BASIC: 98.2%
PASCAL: 1.7%
FORTH: 0.08%
LISP: 0.0001%
SMALLTALK: 0.000000002%It therefore follows that a sensible company attempting to sell a high performance attached processor to the largest number of potential customers would naturally select BASIC as the language to be supported. We did this, even BEFORE we decided to go commercial. After an interim of writing 68000 assembly language code to prove that the 68000 is a heckuva graphics processor, we are about to return to the task of providing EVEN MORE support to BASIC.
(Please do not write and ask when WE, ourselves, are going to support one of those other obscure languages. We are NOT, for largely the reason your local schools do not teach Basque as a foreign language. The demand just isn't there! The support for PASCAL is going to come from those people who use PASCAL. Some such support has already surfaced; more is on its' way. Support for FORTH will come - surprise! - from the FORTH folks.)
THOSE REMAINING TRANSCENDENTAL FUNCTIONS: We have studied the computation of those remaining math functions which we have not yet implemented, and we feel that we have a good handle on all of them except the RND function (after reading several articles on this subject we wonder if ANYBODY has a good RND function). We went to the source book; "COMPUTER APPROXIMATIONS". The one by Hart et al, not the earlier one by Hastings. See newsletter #4, page 8. Knowing in advance just what had to be done certainly simplified the understanding of the algorithms used by Microsoft. We had taken the REVERSE approach on analyzing our first transcendental (LOG). We eventually got there, but what a waste of time!
Page 1, Column 2
We plan to use writing these algorithms in Applesoft format as fill work during the next couple of months, which will be devoted primarily to hardware design. But you can't (WE can't!) concentrate exclusively on ONE project. You get stale that way! So, we will before long have all of the difficult math functions except RND working in the 68000. ABS is NOT a difficult function!
We invite correspondence on the RND problem. Any red-hot algorithms out there? Don't bother sending in the one about taking the last few digits of the LOG function; that one isn't very random!
MATH FUNCTION STATUS SUMMARY: We have long since implemented +, -, *, /, LOG. We have recently implemented SIN, COS, SQR. We have not yet implemented TAN, ATN, EXP but we feel we have a good handle on them. RND has us stumped. Other math functions not listed are either not supported by Microsoft BASIC (Arc Sin, Sinh) or are trivial (ABS) and are being left to the 6502.
A COMPILER STATUS REPORT: We are the proud (?) owners of no less than FIVE basic compilers. We have the DTL and Petspeed compilers for our 8032; we have the Hayden, Speedstar and Tasc compilers for the Apple. ALL of them work with our 68000 board 'hooks' since, as predicted back in newsletter #2, they ALL call the ROM routines when it comes time to do the math. Aside from that, there are some differences which we had not thought about. BIG MISTAKE!
What we had NOT thought about was the question of whether we could distribute compiled programs as examples. It turns out that there are restrictions on the distribution of the compiled programs in some cases. These restrictions are enforced either by run-time 'dongles' or by a license agreement. A 'dongle', by the way, is a hardware key, exact configuration unimportant.
Fortunately, there is ONE vendor of a Pet compiler and ONE vendor of an Apple compiler which places no restrictions on the distribution of the compiled programs. That is Petspeed by Oxford Computer Systems Ltd. and TASC by Microsoft, respectively. We have, in effect, NO CHOICE in the selection of compilers to use for demonstration purposes!
WE CAN THEREFORE ABSOLUTELY CONFIRM that our statement made in early issues of this newsletter that our board would work very well indeed with a compiler are ABSOLUTELY ACCURATE. Speed improvements for the Dtack board/compiler combination have been reported all the way from 5 to 12 times! We can also (sigh) report that our prediction that our board would provide a much SMALLER improvement in the absence of a compiler has ALSO proven accurate. Our statement that the 'long program problem' (outlined in newsletter #3) would NOT be helped by our board (alone) has also proven true.
Page 2, Column 1
THE NEW 16 BIT OPERATING SYSTEMS
There has been a great deal of Sturm und Drang in the personal computer press lately on the subject of operating systems for the new, more powerful 16 bit microcomputers. We have ignored them up to now because N different articles proved conclusively that N different operating systems will become standard!
Let us here examine the question: is it POSSIBLE to develop a standard operating system (please note that this is a VERY DIFFERENT question than: is THIS PARTICULAR system going to become the standard).
We must first decide whether the operating system is to be GENERAL PURPOSE or REAL TIME. Most of the applications we are familiar with - payroll and general ledger, text editing, engineering calculations etc. are general purpose. CP/M is a general purpose operating system, as are the Pet/CBM and Apple II operating systems. However, these general purpose operating systems cannot be used in certain process control and laboratory applications where real time response is needed. Contrarywise, real time operating systems carry around a lot of baggage that is totally unnecessary in a general purpose computing environment. It follows that no single operating system will be completely satisfactory in BOTH computing environments,
Imagine that we have a hundred people in a room who are looking for operating systems. The ones with lighter color hair want general purpose systems but the ones with darker colored hair want real time systems. So we direct the persons with light colored hair to the north end of the room and the dark colored types to the south end.
Since 16 bit systems are such more powerful than 8 bit systems, we next must decide whether the operating system is to be MULTI-USER like UNIX or SINGLE-USER like CP/M. In this crowd, it turns out the men want multi-user systems while the women want single-user (Linda, we are giving the women the best of it!). So, all the men move to the east side of the room while the women move to the west side. We now have FOUR groups, whose need is different from each of the other groups.
Now we tell the ones who want to work with the Intel 8086/8088 line of processors to sit down, but Motorola 68000 fans to remain standing (yes, we know that it is theoretically possible to implement an operating system on two different processors SIMULTANEOUSLY but we don't expect to see it happen in the real world). We now have EIGHT groups with different needs.
Page 2, Column 2
There are MANY competing software houses who wish for THEIR unique, proprietary operating systems to become the standard and thus financially successful. However, we will NOT point out that there are LEGITIMATE DIFFERENCES in opinions of what the features of a good operating system should be BEYOND the factors discussed above, because that would be unfairly stacking the deck.
YOU are in charge of a software house which is going to develop a 16 bit operating system. Do you serve the needs of the dark-haired women who are standing up? Or one of the OTHER seven groups? Does ANYBODY out there see the realistic possibility of a SINGLE standard operating system? If so, you must know something that we don't!
ABOUT WONDERFUL UNIX:
UNIX is widely recognized as the best operating system available (for software development) on the famous PDP11/XX series. One of the reasons it is the best is that it is 'FINE TUNED' to the PDP11/XX environment. That environment features an astonishingly (to Pet/Apple users) fast hard disk and an abysmally slow (for $50,000) CPU. UNIX therefore places as much of the operating system burden as possible on the disk and as little as possible on the CPU.
Now let us consider the situation of an Apple II user with two DISK II minifloppy drives and a brand new Dtack board with an 8MHz 68000. Are you REALLY SURE that user wants/needs UNIX?
A second point is that UNIX is highly useful (if you have a PDP11/XX) for software development but is much too complex and unwieldy for a dentist or an attorney.
And since UNIX is really NOT a good match for a system that uses a floppy disk, you are seeing lots of UNIX 'look-alikes'. Here is what you should know about these systems: 1) There are lots of thee around, 2) All of them deviate from UNIX in some respect, 2) None of them are identical to any other of them, 3) Some of them are poor implementations, 4) It is almost certain that NONE of them are implemented as efficiently as the original UNIX, which had the vast resources of the Bell labs behind it!
Look: UNIX is a VERY GOOD operating system. If you want to use UNIX, grasp $50,000 firmly and go call on your local DEC sales office. And be sure the system you buy includes the fast hard disk!
ALL IS NOT LOST:
Now that we have pointed out that certain rational factors affect (or should affect) your choice of operating system, it is only fair that we tell you which operating system is the BEST operating system for YOU! So, on the next page we will do just that!
Page 3, Column 1
"It is THE WORLD'S VERY BEST OPERATING SYSTEM for the Apple II" asserted the salesman, er, technical representative. "Fortunately, it happens to support that new higher performance 16 bit processor board you have just bought. For a monetary consideration less than that of a 128K expansion board, YOU, sir, can enjoy the enormous benefits of this new operating system."
"I'm sure glad to see the world's very best operating system become available for my Apple" replied Stem McIntosh, proud owner of a Apple II. "I have been getting fed up with the crummy operating system I've been using these last three years. Especially since I read about all those new, superior op systems in BYTE"
"A commendable attitude!" stated the tech rep greedily. "We are certain that you will be most favorably impressed by our system. In time-honored Apple fashion, we have prepared a demonstration program to demonstrate the superior features of our operating system. This demonstration program, which we call 'GAME', involves shooting rocks. If you will just sit down at the keyboard of this Apple computer..."
"Great!" exclaimed Stem. "I love to shoot rocks!" Stem then proceeded to key 'RUNGAME(return)'. Three quick beeps were heard and an error message appeared on the monitor. "Huh?" Stem inquired.
"Sir" enjoined the tech rep, "you are attempting to use the syntax of your bad old system. Perhaps we should take a moment or two to review the really comprehensive manual which documents our operating system." With that he placed an instruction manual on the table beside the Apple computer. The instruction manual was slightly smaller than the last three BYTE magazines combined.
"You mean I'm going to have to study that MONSTROSITY just to shoot rocks? Why should I have to study something like that? I have been using my Apple for three years now," stated Stem, puffing up in indignation, "and I CERTAINLY know how to run a program!"
"But sir! All of that knowledge you have gained in those past three years applies ONLY to that BAD OLD operating system you have been using! "
"Let me see if I have this straight" replied Stem. "The operating system that I know how to use because I HAVE used it, for three years now, is the BAD OLD INFERIOR operating system. And the world's very best operating system is the one I DON'T know how to use? Look, I am NOT a computer professional. I LIKE my Apple but now that I think about it, it took me quite a while to learn to use what I have. You are telling me I will have to start over. And that is certainly a formidable looking instruction manual you have there!"
Page 3, Column 2
"Our operating system is the very best" stated the tech rep. "It has capabilities far beyond those you are accustomed to. Naturally, the (very complete) manual we provide will be somewhat, er, thicker than the manual for your former operating system."
"And it will take me more than, er, three years to learn to use all those capabilities that are far beyond what I am accustomed to, of course? "
"With daily use and a few evenings of study you should quickly adapt to the superior features of our new operating system. It is the world's very best! Surely it is worth a little effort" the tech rep persuaded, "to enjoy its advantages?"
"Look, I have just explained to you: I am NOT repeat NOT a computer professional! I use my Apple maybe an hour a day, maybe four or five days a week. I am not at all sure that I want to expend the obviously large investment of time, never mind the cash cost, to learn your new operating system when I have one that works, and which I already know." and Stem stood up and walked out of the store.
THE WORLD'S VERY BEST OPERATING SYSTEM, dear reader, is the one YOU already have and already know and which does what YOU need done.
If you happen to be a computer professional and you use your computer 8 or more hours a day the above statement is of course inoperative. But remember that you are part of a very small MINORITY!
SOMEBODY IS ALWAYS STEPPING ON OUR LINES! About a week after we finished writing the preceding three-pronged view of operating systems, MICROCOMPUTING magazine (formerly KILOBAUD) arrived with a three page article on UNIX. A very good article, by the way. Here are a few quotes from this article:
"...the average Unix user of the future will be less sophisticated. Many of those users will only know how to log in."
"If it took you a month to learn the ins and outs of CP/M, it will probably take you a year to have a comparable knowledge of Unix."
"About 5 million bytes of disk storage is required just to support the Unix overhead..."
"CP/M ... has become the de facto standard because it was there, not because of its sophistication."
Page 4, Column 1
HOW TO PROGRAM THE 68000:
First, learn how to program the 6502 (we are talking about assembly language, of course). Oh? You already know how to program the 6502? Well, that does help a bit. You see, most of the techniques you have learned on the 6502 will carry over to the 68000, with one HUGE possible glitch, which we will discuss later.
You will be pleased to know that the 68000 has a 'branch always' command, something lacking in the 6502. But the best news is the 'quick' instructions. Motorola evidently did a statistical analysis of the most frequently performed operations and came up with short, quick formats for these instructions. The instructions are ADDQ, SUBQ and MOVEQ.
ADDQ is a one word (two byte) instruction which can add #1 thru #8 to an operand. The addition size can be byte, word or long word. The operand can be a data register, an address register, or one of seven other addressing modes. SUBQ works just like ADDQ, of course.
EFFECTIVE ADDRESSES: In the previous paragraph we wrote "... can add #1 thru #8 to an operand." Motorola prefers to use the expression "...to an effective address." Which really means 'the operand located at the effective address'. Do not let this 'effective address' business throw you! You have been using 'effective addresses' all along with the 6502, but you have been calling them 'addressing modes'. You can do an (6502) LDA from several effective addresses: zero page, absolute, immediate data and various indexing modes. So when you see '{ea}' in a Motorola publication, that means 'effective address' which in turn is the same as 'addressing modes'.
With the 6502 you can 'LDA' using immediate data as an addressing mode. However, you obviously cannot 'STA' into immediate data! Or, to use a 68000 example, we cannot 'ADDQ' with immediate data as the destination operand, for the same reason. Motorola has formalized these (largely common sense) rules by providing names for the usable 'ea's. For ADDQ, to continue our example, only 'alterable' addressing modes are allowed. Again, this is only common sense.
Page B-1 of the User's Manual (UM) lists the various addressing categories: Data, Memory, Control, and Alterable. Combinations of addressing categories are sometimes used, as in Data Alterable. This means that only those modes which are legal Data addressing modes and also Alterable addressing modes are legal. You will refer to page B-1 OFTEN when learning to program the 68000. UNLESS you use our 'hand assembler's helper' program, which is available for both the Pet and Apple and has been placed in the public domain. This program already knows which addressing modes are legal for each instruction and only 'menus' the legal modes.
Page 4, Column 2
Remember the 'otherwise rational person'? We can hear him snickering and chortling as he reads the last paragraph. WHY deserves an explanation. You see, Motorola promotes the 68000 as having a "highly regular structure". They also assert that "all 14 addressing modes operate consistently" (UM page 1-8). Well, that is not in accordance with the facts. The addressing modes are used fairly irregularly, However, it DOES appear that the National 16032 microprocessor will eventually become available with a MUCH more regular instruction set. So, 'otherwise rational' is laughing! In view this, should we all drop the 68000 and start a stampede to the 16032? Not at all. Here's why:
Let us return for a moment to the 6502, a processor with which you may be familiar. A single byte is used to specify a particular instruction AND ITS ADDRESSING MODE. If the addressing made is implicit, such as CLC (clear carry) we have a one byte instruction. For a zero page or immediate data LDA we have a two byte instruction. For in LDA to absolute memory or AB,X or AB,Y we have a three byte instruction.
Because a single byte always fully specifies the instruction AND ITS ADDRESSING MODE, a maximum of 256 such instructions is possible. In fact, the 6502 uses 156 of those 256 possible instructions. Obviously, if we are to have more than 256 possible instructions including their differing addressing modes, we will need to use more than a single byte to specify the instruction. This additional byte will in turn slow the processor since an additional byte fetch will be required for each instruction.
We trust it will therefore come as no surprise to you that the 68000 uses more than one byte to specify the instruction and its addressing modes AND (this is something new) the SIZE of the operand(s). The 68000 crams ALL of this information into a single word (two bytes). And some instructions (notably MOVE) are two-address mode instructions. This means that either Motorola had to be very, very clever at assigning the 65,536 possible one word codes or else had to add another word to specify the instruction. And that would slow the 68000 down by forcing an extra word fetch for each instruction.
(As in the 6502, the initial word is sometimes followed by additional words, such as immediate data. Nevertheless ALL of the necessary information to specify WHETHER additional words are required, HOW MANY additional words are required and WHAT their function is, is contained in that first word.)
Motorola naturally chose to take the route of clever assignment of the available 16 bits. As a result of this intelligent choice on Motorola's part the instruction set is not particularly regular (!).
Page 5, Column 1
If the National 16032 has a FULLY REGULAR instruction set they will either have fewer available instructions than the 68000 or they will have to use more words to specify all the parameters of the instruction. There is no such thing as a free lunch!
On the other hand, that 'there is no free lunch' bit cuts both ways. It is much easier to design (create? program?) a compiler if the instruction set is fully regular. So it will be easier to bring up FORTRAN, COBOL, ALGOL and such with the 16032. The tradeoff is, it will be easier but the result will be slower. And WE are not particularly interested in compilers, anyhow.
Back to how to program the 68000: we NOW have to figure out how to sneak up on perhaps the most important new programming aspect of the 68000 as compared to the 6502, called 'sign extend'. Sign extend is mentioned in the UM but it is NOT explained in detail nor is it illustrated. So we will do that for you here. Let us discuss the instruction 'MOVEQ': this instruction was considered so important by Motorola that they assigned 1/32 of the available code space to it! (See why the instruction set isn't regular?)
MOVEQ is a cutie. This is a ONE WORD instruction whose second byte contains the data byte to be moved to the destination. The destination is ALWAYS a data register. ALL 32 BITS OF THAT REGISTER ARE FULLY DETERMINED BY THAT ONE BYTE OF DATA! This occurs through what is known as 'sign extend'. Sign extend is a term related to TWO'S COMPLEMENT notation, in which the most significant bit of a datum determines the sign. To illustrate this, suppose that data register 3 contains $AAAA AAAA. After:
7607 MOVEQ #7, D3D3 will contain $0000 0007. After:
7680 MOVEQ #$80, D3D3 will contain $FFFF FF80. The way this works is that whatever value B7 has is stored not only in B7 of D3 but also in the 24 high order bits of the data register. (That is 'B7' to distinguish bit 7 from D7 which is, of course, data register 7.) This process is illustrated in Figure 1 below.
Page 5, Column 2
The 6502 has only one equivalent to this 'sign extend' function. It can be quite useful at times and a real pain at others. For instance, suppose that we want to set D5 equal to $FFFF FFFF. Here are three methods:
1) 2A3C MOVE .L #$FFFFFFFF, D5 (3 words)
FFFF
FFFF
2) 4285 CLR .L D5 (2 words)
5385 SUBQ .L #1, D5
3) 7AFF MOVEQ #$FF, D5 (1 word)Thereby proving for the umpteenth time that there is more than one way to skin a cat! Method #1 will always be chosen by the experienced 6502 programmer the first time around.
MORE GOOD STUFF: The 68000 has a 'branch to subroutine' instruction. If a particular subroutine is being called several times in a short sequence of code, a savings of both time and memory can be obtained by locating the subroutine close to the code which calls it and using the one word (two byte) form of the 'branch to subroutine' instruction. This has the additional benefit that the resulting code is relocatable since the address of the subroutine is relative, not absolute. The two word (four byte) mode of the branch can be used if the subroutine is located more than 128 bytes away.
In addition to saving memory, the shortest form of the various instructions will invariably execute more quickly. The reason is that things happen VERY QUICKLY for any data that is already INSIDE that big 64 pin chip. But it takes four clocks (half a microsecond at 8MHz) to fetch each word from memory (or longer if Dtack is not grounded). In the code example given above, method #1 requires 12 clocks, #2 14 clocks (two instructions) and #3 exactly FOUR clocks!
THE 68000 IS BUS BANDWIDTH LIMITED: Please note that the execution time of examples #1 and #3 are EXACTLY EQUAL to the time required to fetch the instructions from memory! This is highly relevant to the following discussion:
Page 6, Column 1
IN THE 6502, ZERO PAGE ADDRESSING is the fastest available method to access memory. The 68000 has an equivalent addressing made, but it is NOT the fastest available method of accessing memory! Address register indirect is the fastest, again because less data bus activity is required.
If we are comparing destination address codes, the difference is precisely equal to the time required to fetch the extra word required to specify the zero page address (in case you have been asleep during the last nine issues of this newsletter, the 'zero page' in the 68000 is 65,536 bytes (64K), NOT 256 bytes like in the 6502.). If we are comparing source addressing modes the matter becomes more complex but address register indirect is consistently faster than the zero page mode.
TRUTH IN ADVERTISING:
THE LATEST BUZZWORD in small computers is MIPS (millions of instructions per second). We are building a board level product with an 8MHz (nominal) 68000 which runs just under 1 MIPS depending on the instruction mix. Since the magic number is OVER 1 MIPS, we miss the brass ring. Also, the product we sell uses a 16 bit microprocessor which has 32 bit registers and some 32 bit instructions.
But the Sage Computer Co. sells an 8MHz 68000 based computer which operates at 2 MIPS! If you do not believe us, check the advertisement on page 27 of May BYTE magazine (a McGraw-Hill publication never lies). Also, Charles River Data Systems, the outfit with the big green frog, sells a 32 (that's THIRTY TWO!) bit system which also uses the 68000.
How come the Sage computer runs at 2 MIPS (YEAH!) while ours runs at a slow-poke (appx) .8 MIPS (BOO!) even though we are both using 8MHz 68000s? Why is our 68000 a 16 bit device but the big green frogs' 68000 is a 32 bit CPU? If YOU have any ideas, write Ann Landers and Dear Abby. They need some new letters.
Incidentally, the big frog's computer also has a system clock (?), TWO serial ports, ONE parallel port and a memory-management unit. The memory management unit locks each of the up to 34 users into their VERY OWN memory space. Although the product bulletin we are getting this information from doesn't mention it, we are POSITIVE that there is a DMA port in there someplace.
Those persons who have DEMANDED that we provide these features on our $495 board now have an alternate source, one which is clearly listening to them! Just bring $18,500 for a single-user model.
Did we ever tell you that there is no such thing as a free lunch?
Page 6, Column 2
DTACK GROUNDED AT 12.5 MHZ!
We have been WORRIED SICK over not having a product which runs at 1 MIPS or faster. So we mortgaged the farm and bought a 12.5 MHz 68000. Then we fetched the premium 'dash-one' memory chips we had hidden for a rainy day and put together a 12.5 MHz 92K board. We peeked at the bus signals with our 200MHz Tektronix 475 oscilloscope and discovered we had a 60 nanosecond margin using 'dash one' Toshiba RAMs. Now, we don't expect you to believe that 100 nsec RAMs can leave a 60 nsec margin with a 12.5 MHz 68000, but it happens to be true, at least on the particular board we (personally) measured. Since our 12BK memory expansion board, which will exist in a production prototype form by the time you read this, has two extra 12nsec buffer delays that margin will drop to about 36 nsec, which is plenty.
Naturally, these experimental projects are not particularly useful unless we can tell you the product is REAL and AVAILABLE. So we approached a wealthy customer and persuaded him to purchase the board so that we could honestly report that we have shipped product that runs at over 1 MIPS! The wealthy customer agreed immediately. We then mentioned, in the course of a phone call to a customer in St. Louis, that we had 12.5 MHz 68000s up and running. 15 minutes later, we received a frantic phone call instructing us to ignore the order for ANOTHER 92K board that was already in the mail and that they would IMMEDIATELY post the additional cost for the 12.5 MHz version. Which is $300.
THERE WAS A VERY EXCITING ARTICLE in April MICRO magazine. We are referring to the one on 32 bit addition using the 6502 and the 6809. There was plot, pacing, drama and all the other stuff that makes for good reading. We even 'learned a few of the facts we never knew before', as the song from Kismet has it. For instance, we learned that 6502 assembly code printed in the middle of a page requires, consistently, 3 clock cycles for a zero page instruction but 6502 assembly code printed at the bottom of the same page requires FOUR microseconds for a zero page instruction (again, quite consistently). Amazing!
The article had HIGH DRAMA because there was some question for a while whether EITHER the 6502 or the 6809 were EVER going to complete a full 32 bit addition! We finally realized why Bob Tripp hasn't published anything on the 68000 yet. It is a dull, boring microprocessor! Look, here's how the 68000 performs a 32 bit add:
D081 ADD .L D1, D0 (remember newsletter #1?)Elapsed time: UNDER half a microsecond with a 12.5MHz Dtack board (OK, we admit the standard 8MHz unit requires all of 3/4 microsecond. You could maybe get a haircut while you wait?).
Page 7, Column 1
LET US TIMIDLY POINT OUT that a 12.5 MHz 68000 runs 56.25% FASTER than an 8MHz 68000. And, since Dtack remains firmly grounded, there are NO WAIT STATES WHATEVER. We ask you, dear reader: just exactly what OTHER companies do you see shipping real 12.5 MHz product? We would ask what company is shipping real 68000 product at ANY clock speed except that we have seen a real, live Wicat. Since the Wicat has TWO 16 bit processors, the second a Z8002 dedicated to the graphics display, one would expect a really SUPER 3-D demo program would come with it! We are told by the proud owners of that Wicat that there is ONE graphics demo: a two dimensional maze program. Whoopie!
We are beginning to see SEVERAL new announcements of 68000 based small computing systems in each week's issue of Electronic News. That previously mythical '100 different announced 68000 systems' say soon become a reality. But who is shipping product, we ask again? Anybody seen a Tandy Model 16 yet?
But back to our 12.5 MHz board (a product which we are completely unreluctant to talk about!): the run time of program D3.FASTER is now 10.3 seconds, hand-timed. That's down from 15.8 sec as reported in our newsletter #7, page 11. And the Sine calculation in our graphics floating point package now executes in 237 microseconds, which is well over four times faster than a 4MHz 9511 and about NINE times faster than a 2MHz 9511. Bye bye 9511!
True, the 12.5 MHz version of the 68000 is a bit expensive right now. But the EIGHT megahertz version of the 68000 was expensive when we started our little endeavor, no? Both Hitachi and Motorola are racing to see which can produce the 'shrunk' 68000 in working form first.
Although the authorized second sources are supposed to wait for Motorola to produce working mask sets, Hitachi thinks they can get there first (and that Motorola might stall a little after getting a working mask). When the shrunk mask version is available the 68000 will suddenly get a lot faster on average and also, once BOTH H & M are producing, a whole lot cheaper. The reason is, a smaller chip size will provide a higher yield due to a lower probability of a surface defect within the chip area. It is the random distribution of imperfections in the silicon wafers which limits the yield of large chips such as the 68000 (and the 8087, which we have NOT forgotten).
Translation: on Jan. 1, 1983 12.5 MHz 68000s will be routinely affordable. Yawn!
Page 7, Column 2
To answer the obvious question, here is how to convert an 8MHz Dtack board to the 12.5 MHz version: the 74ALS04 now used as the crystal oscillator will have to be replaced with a 74F04, a direct plug-in replacement. The crystal must be changed to 12.5 MHz. A resistor and capacitor must be removed from the crystal oscillator circuit and a resistor and capacitor must be added. The new resistor involves cutting a trace on top of the board. You will need -1 Toshiba RAMs and, we almost forgot, a 68000L12.
LIES, ALL LIES: it may have come to your attention that we occasionally print something in this rag which is not in accordance with the facts. Invariably accidentally, we defensively add. Back in newsletter 12, page 8, we stated clearly that we would be shipping 200 nsec Toshiba RAMs, which are the -2 parts. We lied! The fact is, we have not to date even SEEN a -2 part! Every board we have shipped has been populated with either 'no dash number' parts (150 nsec) or -1 parts (100 nsec). Since this fact is certainly not disadvantageous to a Dtack board buyer, we have not lost any sleep over it!
You see, Toshiba 'shrunk' the chip size 3 or 4 months ago and VERY few chips are (now) remotely close to being as slow as 150 nsec. Once 74ALS245s and 74ALS138s are generally available, even the 'no dash' parts will be fast enough for a 12.5 MHz 68000! (The two ALS part numbers just mentioned are a data buffer and an address decode chip. They don't yet exist - we think - but they will!)
Looks like we are going to have to print that timing analysis one of these days.
LISA WATCH:
In the latest Infoworld (as we write this) an Apple spokesman is quoted as stating that LISA probably will not be introduced for another year. Translation: either within four months or longer than 20 months!
TRUE STORY DEPT:
Once upon a time there was a company called RCA, which is short for Radio Corporation of America. In the middle to late 1960s, they were looking for fertile ground in which to plant a new enterprise so that the wealth of their stockholders would grow. It came to their attention that there was a computer industry which was exhibiting rapid growth and high profitability, at least for a company called IBM. When they discovered that computers were made of transistors the RCA board of directors rejoiced mightily, for RCA was ALREADY in the transistor business.
Noting that IBM was highly profitable, they introduced a line of computers which were compatible with the IBM 360 series. You see, the supreme court of the land where RCA resides had just ruled that it was illegal for IBM to tie their software to their hardware. What transpired after that ruling was called 'unbundling'. It meant that it was equally costly to run IBM software on computers made by IBM or by some other company, such as RCA. The name which RCA placed on its line of IBM 360 compatible computers was 'SPECTRA 70'.
Page 8, Column 1
RCA poured hundreds of millions of dollars into creating the subsidiary to manufacture and sell these SPECTRA 70s. Finally the design was completed, the manufacturing plant in place and the production force hired and trained. Impressive numbers of mainframe computers began to roll off the production line towards the shipping dock. The RCA board of directors rejoiced mightily once again, and awaited the large profits. Instead of lots of cash, what they got was a phone call from the factory manager asking for more money. Lots and lots more money.
So the RCA board of directors poured - and we mean this literally, folks - hundreds of millions of dollars into the SPECTRA 70 operation. Ever larger numbers of SPECTRA computers vanished from the shipping dock, but no cash ever passed from the factory to RCA. Instead, there was another phone call from the factory manager asking for more money. And even MORE hundreds of millions of dollars were poured into the factory from RCA. The line of computers leaving the shipping block became like unto a blur! And behold, there was ANOTHER phone call.
Finally the RCA board of directors said, "Wait just one cotton-picking minute!" and went into a huddle. Emerging from the huddle, they decided to close down the factory and write off the operation lest they lose even MORE hundreds of millions. Did they decide that they had made a mistake? No, in the true manner of the country in which they are located, the RCA directors decided that someone ELSE was at fault and, (you maybe guessed this part?) that they would SUE that someone. Since IBM had lots of money and was in the same industry, they decided to sue IBM.
(Before continuing further, we again remind the reader that this is a TRUE story!)
Having decided to sue IBM, they naturally had to decide HOW MUCH to sue for. On advice of their own council (translation: the shysters in the legal dept.), they hired independent auditors to determine EXACTLY how much money had been lost. In due time, young men wearing serious faces, three piece suits and short haircuts arrived and began pouring over the financial records of the defunct operation.
Page 8, Column 2
The more they examined the records, the more serious the expression on their faces became. After a few weeks, they ceased examining the records, gathered together and whispered among themselves. And then they departed, quietly. The next day they were back on the job accompanied by stern middle aged men with fierce visages. Papers were examined at an ever increasing rate. After nearly two months, the middle aged men with fierce visages gathered together and whispered among themselves. They then called the serious young men to them and all left together, quietly.
At the start of the very next week an enormous number of vehicles converged on the warehouse containing the records of the defunct computer operation. Into the warehouse marched dozens of serious faced young men. Behind them marched many stern middle aged men with fierce visages. Finally, a few elderly men entered, grizzled veterans all, partners in the accounting firm. The elderly men quietly gave orders to the stern middle aged men. The stern middle aged men barked orders to the serious faced young men. A storm of invoices, waybills, credit memos and ledger sheets flew into the air! It was no longer possible to note the short haircuts on the young men since all that could be glimpsed of them was elbows and, um, a certain nethermost portion of their anatomy.
After several weeks the partners gathered together, whispered among themselves, and everyone left, quietly. It was two days before the last memorandum settled out of the air and lightly fell to the floor.
Finally came the day when a chauffeur-driven limousine drew up in front of the office of the Chief Executive Officer (CEO) for RCA. There emerged from the limousine one serious-faced young man with short haircut, carrying a briefcase. Then emerged a stern middle aged man of fierce visage, who held the door open for the most senior partner of the accounting firm. These three marched into the building and were, of course, immediately shown into the office of the CEO.
The CEO, rubbing his hands together greedily, asked "Well? How such do we sue them for, er, how much did we lose?"
Upon being cued by a small gesture, the serious faced young man with short haircut opened the briefcase he was carrying and took out a large envelope. He passed the large envelope to the stern middle aged man. The stern middle aged man opened the envelope and handed the few sheets of paper to the most senior partner of the accounting firm.
The most senior partner of the accounting firm passed the few sheets of paper over to RCA's CEO and stated, "David, the computer operation you shut down was actually showing a tidy profit." And then all of the accountants left.
Page 9, Column 1
FOR THE THIRD TIME we assure you, dear reader, that the above story is true. How, you may ask, can an operation which eats large sums of money be PROFITABLE? Simple. The computer factory did not SELL SPECTRA 70s, it LEASED them. And RCA management was not familiar (evidently!) with leasing operations. It's like this:
Say we have this computer with a retail (!) price tag of one million dollars. Since people who sell things priced at a million dollars generally like to make a profit, the COST OF MANUFACTURING is less. On a less exalted scale, WE are happy to sell a $1143 fully loaded Dtack board because our cost of manufacturing is less. The difference is what is called, like we already said, profit. If YOU don't like the concept of profit, emigrate to Russia. They have this certain economic system which also doesn't like profits.
If RCA had SOLD those SPECTRA 70s, all would have been happiness and we wouldn't have this story to bore you with. But they LEASED them. This means that it cost, say, $720,000 to build that computer. When it left the shipping dock, did they receive $1,000,000? No, they received the first and last month's lease price - an amount which is VASTLY less than the manufacturing cost! And so the operation ATE MONEY!
Why did RCA operate in that fashion? They were trying to compete with IBM, which also did, and does, lease its products. Why could IBM operate in a way that RCA could not? IBM has been around for (what seems to be) centuries, and has built up an enormous sum of retained profits. They INVEST these retained profits in their leasing operations, profitably. And as an ongoing operation which has been leasing for decades they have, each month, a VERY nice income from those leases. Since RCA was just starting, the monthly income from the leases was NOT enough, not nearly, to carry the operation.
And so RCA management REALLY DID shut down a profitable operation, thinking that it was losing money. The amount they wrote off was in the ballpark of $300,000,000! And it was not until nearly two years later that their independent auditors informed them that they had in fact shut down a PROFITABLE operation!
WHAT DOES THIS HAVE TO DO WITH THE PRICE OF TEA IN CHINA? Simple. A company can get KILLED by an adverse cash flow. We recently read about a number of personal computer retail stores which have been badly hurt by slow payments from some of their large corporate customers (check newsletter #2, p 9, par 7!). We are NOT going to permit Dtack Grounded to suffer over poor cash flow. Those of you who are buying boards will continue to have to send 10% down, balance C.O.D. plus shipping (U.S. customers only) or send full payment in advance, in which case WE pay the UPS charges (U.S. only).
Page 9, Column 2
IN CASE YOU HAD NOT NOTICED, the competitive positions of the various contenders in the microprocessor 'game' change from time to time. Here is our 'short form' assessment of the current status of the various entrants, followed by a prediction of the changes to be expected in the near future. THESE COMPARISONS ARE BASED EXCLUSIVELY ON PERFORMANCE.
The clear winner in the GENERAL PURPOSE microprocessor performance race at the moment is the 12.5 MHz 68000. Even the 8MHz part is faster than the second place entrant.
Clearly in SECOND place is the Z8000 series. In fact, for the specific purpose of string handling, the Z8000 is BETTER than the 68000 at this time! (The Z8000 already has microcoded string handling instructions; the 68000 doesn't YET but instruction set space HAS been reserved for that purpose.) AFTER we started Dtack Grounded, but BEFORE Zilog lost its domestic second source, we told someone in casual conversation that if we were to build a word processing terminal, we would unhesitatingly select the Z8000 as the CPU. If price/performance were the SOLE criteria, that recommendation would retain unchanged today.
In third and fourth place we have a couple of oddballs: the T.I. 99000 'microcomputer' (really a GOOD programmable controller) and the Fairchild 9445, which is now in limited production. The local sales specialist for Fairchild very apologetically explained that the yield wasn't that great yet, but perfectly good 16 MHz parts were available for $150 each. We think we bent his nose when we told his that was nothing; we had paid $249 each for our first two 68000s and at the time we were glad for the opportunity to buy them!
However, as we said earlier, the 9445 is dedicated to the bad old NOVA instruction set. If you need to run NOVA software on a more modern CPU, the 9445 is just the thing for you. Otherwise, forget it. Also forget the 99000 if you have not done so already.
TAIL END CHARLIE is the good old 8086. True, the 8088 is even slower than the 8086 but we have implicitly limited this discussion to the 16 bit (or more!) microprocessors. In fairness to Intel, the 8086 was not INTENDED to be the best. It was intended to be FIRST, with downward software compatibility. Anytime you have DOWNWARD compatibility, you are NOT trying to build the best! (More on this subject later.)
THAT COVERS THE GENERAL PURPOSE MICROPROCESSORS. What about specialized devices such as floating point math processors? (Is THIS one easy!)
Simple; the Intel 8087 is in first place and there IS no second place! Yet.
Page 10, Column 1
HOW ABOUT THE NEXT 6 MONTHS?
Motorola has already seen silicon for 20% and 30% shrunk 68000s. Not working silicon we understand, but they are clearly working on the problem. Six months from now the world's best general purpose microprocessor will be a 15 or 16 MHz 68000. At that time, the NOMINAL speed will be 10 MHz. Since our board (static RAM version) works nicely at 12.5 MHz, we are looking forward to this development.
National Semi is making threatening noises with their 16032; the part actually exists in a 6MHz engineering sample version right now. And they have just CLAIMED that the real part exists (but not 10MHz). Since Motorola's 68000 was in this position as of 1980, National is about 24 months BEHIND Motorola.
The National folks think they are shortly going to be in a struggle with Motorola for first place. WRONG! They are going to be in a struggle for THIRD place with the Z8000! (Or second place if Intel does not do its thing on time.)
PERHAPS (we say it again: PERHAPS) within our six month's time frame, Intel will be sampling their 80286. We understand that this will be produced in a 68 pin package. It will NOT, therefore, be directly substitutable for the 8086. This fact has, er, ANNOYED some of Intel's 8086 customers! Here are the most important things you need to know about the 80286:
A) The part will potentially have VERY high performance. It will (potentially) have performance in the range of a 10 MHz 68000. Since it is very difficult to pin down the EXACT performance of a GENERAL PURPOSE microprocessor, that leaves room for Intel to claim top rank if Motorola does not get its 'shrink' working by the time the Intel part comes along.
8) We emphasized the word 'potentially' above because the 80286 will continue to maintain downward compatibility with the 8080. Which means the device will spend, in practice, much of the time running 8080 code at slightly greater than 8080 speeds. We predict that USERS of Intel micros will clearly identify the fallacy inherent in this a long time before Intel does. The more sophisticated users ALREADY HAVE!
C) It is traditional for microprocessor manufacturers to lie about the performance of their devices (unless, of course, they really DO have the best product). Intel has already started this with their propaganda on the 80286. It's like this: the 8086 is currently available in a 10 MHz version. Intel has announced that they fondly HOPE that the 80286 will come in at 8MHz initially. They have ALSO announced that the 80286 will be about 6 times faster than the 8086!
Page 10, Column 2
APPLAUSE! An 8MHz 80286 6 times faster than a TEN MHz 8086? Um, ah, will you hold that down, fellow? The 'six times faster' was determined by, er, cheating. First, as you might expect, by selecting the most favorable possible algorithms for that comparison. Now, by the rules of the game, THAT is not cheating.
What IS cheating is, the comparison is between a 10 MHz 80286 (a part which even Intel admits will not initially exist) and a FIVE (repeat, FIVE) MHz 8086. Summary: Intel is comparing a part which doesn't exist, at a clock rate which will NOT exist by their own admission for some time, with the SLOWEST available 8086!
Let us emphasize this point further: on the rare occasions when Intel chooses to embarrass itself by comparing its 8086 with Motorola's NOMINAL clock rate (8MHz) 66000, they always choose the 10MHz 8086. (Intel NEVER compares performance with the available 12.5 MHz 68000.) But when they want the 80286 to look good, they compare it to the SLOWEST available 8086! Their hope is that the unwary potential customer will put the two numbers together and conclude that the 80286 is FAR better than the 68000. That is not in accordance with the facts.
LET US MAKE THIS PERFECTLY CLEAR: The Intel 80286 is going to be a VERY high performance microprocessor. Anything that can run with a 10MHz 68000 is not to be sneezed at. The ONLY question, since the part does not yet exist, is WHEN Intel will be able to ship working parts in reasonable quantity.
Unless you have not been following the microprocessor industry very closely, you have probably not heard of the 80286 until now. Since major new parts are (in recent tradition) announced LONG IN ADVANCE, wha hoppen?
Wha hoppen is that Intel's game plan just got drastically changed. The planned progression had been to move the LOW PERFORMANCE end of Intel's 16 bit line from the 8086 to the 80186 eventually, and then to the 80286 a long time after that. In the meantime, Intel saw the iAPX 432 as the answer to the Motorola 68000 at the high end. Well, the 432 is flying from Intel's flagpole, but nobody is saluting. As we reported last issue, Intel has actually announced an intention to develop a 32 bit version of the 8086. Since this part will COMPETE FOR THE SAME SOCKETS with the 432, Intel has tacitly admitted the failure of the iAPX 432.
Some of the readers of this newsletter have been thoughtful enough to send us additional information about the 432 that has not been generally published. Like a beautifully penned 6 page handwritten analysis written by a mathematician from Eindhoven (for the benefit of you locals, Eindhoven is on the other side of the big creek). When this new information is combined with what we had already figured out by ourselves, it's sayonara, 432!
Page 11, Column 1
Therefore, since there will be about a three year lead time on that Intel 32 bit 8086, Intel has placed TOP PRIORITY on the next best thing they have in the wings, which is the 80286. Since this part has been (quietly) in development for some time, they may well be able to bring the part in relatively quickly, like in the six months we guess. However, Intel has some problems.
Intel is accustomed to TELLING others what kind of processor to use. They 'did good' with that approach for nearly ten years! The reason, very simply, was that Intel LED the technology race until recently. At this point we hasten to interject that the 6502 clearly outperformed the 8080, but came along after most of the design decisions had been made. The Z80, superior to the 8080 and in its Z80A version technologically on a par with the 6502 and software compatible with the 8080, actually SUPPORTED the Intel image of performance leadership. By coming out with the first viable 16 bit processor (the 8086) three years ago, Intel temporarily reaffirmed its performance leadership.
Intel kept a weather eye out for what the peasants were up to, like Motorola with their 68000. No sweat, they thought. We got that 60 foot tall gorilla called the 432 to step on the 68000 with! Intel had a game plan to stop the Motorola off tackle plays and has been hit by three quick touchdowns scored by long passes over the safety. At this point one changes one's game plan or one quits! Intel has changed its game plan.
Since the newly announced Intel 32 bit machine is at least three years away, we need not discuss the technical aspects now. Especially since even Intel doesn't know anything about it yet! The LAST 32 bit processor Intel announced had a 16 bit bus. Will the new machine have a 21 bit bus? A 27 bit bus? Will Intel cast aside all corporate precedent and put a 32 bit data bus on their NEXT 32 bit processor?
A summary of our projected performance ratings, six months from now:
1st place: 15 or 16 MHz 68000 (doggone expensive)
2nd place: 12.5 Mhz 68000 (affordable then)
3rd place: tie: 10 MHz 68000
8 MHz 80286
5th place: tie: 10 Mhz Z8002
8 MHz 16032
7th place: tie: 24 MHz 99000
16 MHz 9445Page 11, Column 2
WHAT ABOUT MATH CHIPS?
Boy, do we ever have a surprise for you here! You see, National Semi has already seen partially working silicon on a math chip that is - get this - even faster than the 8087! Before you get too excited, the fully functional production part is slated for production late this year, and consider this carefully: National Semi has never, repeat not ever, been PESSIMISTIC about delivery dates in their microprocessor department! We are trying to tell you, don't start holding your breath yet.
So why mention the National part at all since the 8087 is available and it also works? Four reasons: A) as we have already stated, the part is potentially even faster than the 8087, B) the only thing that is going to cause Intel to drop the very high current price of the 8087 is competition, C) since the National part has a MUCH smaller chip area than the 8087, it can be profitably produced at a much lower price, D) the National part is configured as a peripheral and can therefore be hooked more or less directly to the 68000 without the need for a 40 pin clock generator.
If you want more details, look up Apr. 14 EDN magazine, p. 61. Bob Cushman, the special features editor of EDN, did a very complete survey article on math processors, and much of the information printed here comes from that article.
The National part number is 16081, and the 64 bit F.P. multiply time is given as 6.2 microseconds (!) at 10MHz. Since we think the clock speed will be 5 or 6MHz when the part initially becomes real, that translates very roughly as a 12 microsecond 64 bit F.P. multiply! And that is MUCH faster than the 8087.
HOW CAN IT BE FASTER? That takes some explaining. You see, the Intel 8087 developed (evidently) out of the iAPX 432 system. Since a high speed math chip was planned early on as a part of the 432, Intel very sensibly decided to also make the part available for use with the 8086 as a coprocessor. Otherwise, the 8087 really doesn't make sense for use with the 8086; it is like bolting a Rolls Royce Conway jet engine onto a Piper Cherokee airframe. As we told you plainly back in newsletter #2.
Now, the systems boys were COMPLETELY in charge during the development of the 432. So they sat down and developed an INCREDIBLY COMPLEX specification as to how floating point arithmetic should work. This specification was then spread among the academic community and the academic community loved it! So the specification became the basis for the IEEE F.P. draft standard which was published last spring in COMPUTER magazine (an IEEE publication). We will now quote from newsletter #1, page 5:
Page 12, Column 1
"(our) 61 bit floating point package... will NOT be IEEE compatible for reasons to be given in a future issue of this newsletter."
Well, THIS is the future issue! You see, the IEEE draft standard at that time reflected the inputs and desires of the systems boys and the academics ONLY! Nobody asked us working engineers, who want something which runs accurately and VERY, VERY swiftly! So the (then) draft standard was VASTLY too complex. Let us note as an aside that this is the exact same thing that is wrong with the 432. You should NEVER, EVER let a systems group proceed without a few experienced working engineers looking over their shoulders making rude comments about things such as practicality.
The plain fact is, we perceived the draft as being on a railroad track to approval last summer. After all, COMPUTER magazine essentially dedicated a full issue to the draft, including a FULL ARTICLE defending ONE of the MOST IMPRACTICAL features of that draft. So, as quoted above, we essentially turned our (Dtack Grounded) back on the standard as far as a software implementation using 68000 machine code. And, we might add, nothing has changed our mind since.
But the 8087 is that early draft standard, set in concrete (oops! make that SILICON!). And since a lot of those crazy features are carefully implemented in hardware, not software, the 8087 STILL represents a TREMENDOUS technological advance, as we have reported here in the past, and as we CONTINUE to report now.
It now appears that a few OTHER working engineers spotted the obvious (WE think they are obvious) problems with that standard. And some of them must work for companies large enough to have some clout on the committee in charge of the draft. Because Bob Cushman reports, as an aside to the article we just mentioned, that the draft has, glory be!, been revised and much of that excessively complex nonsense has been DROPPED from the draft! HOORAY FOR US ENGINEERS!
So the National 16081 is designed to conform to the new draft? No, National simply adopted the DATA FORMAT of the IEEE (draft) standard. They (sensibly, in our opinion) simply ignored the systems/academic fooraw. This decision on Nationals' part is so eminently sensible that we think the decision MUST have been made by others than the persons who have been running the regular microprocessor section!
And because of that EMINENTLY SENSIBLE decision, the 16081 (WHEN available) will be MUCH faster than the 8087 (which is NOW available). The committee did throw Intel this one bone: the VERY COMPLEX nonsense was ALLOWED (but not required) under the new drift. If the current draft is voted in (which seems likely) the 8087 WILL meet the standard, the 16081 will NOT meet the standard and WE will switch (?) to the 16081 when it is real.
Page 12, Column 2
A NUMBER OF READERS HAVE DETECTED A DISCREPANCY between an earlier mention of the 'data path analysis' problem and our teaser on page 14 of the last issue. As it turns out, the discrepancy is apparent ONLY, not real! Also, many readers skipped page 14 entirely. If YOU did, stop NOW and come up to speed with the rest of us.
OK, all together now. We stated on page 14 that we would build an accessary board with up to 8 68000 processors. We did NOT state, you will notice, that such a board would be GENERALLY useful! However, there are certain specific applications where such a board would be useful indeed. ANY application in which the computational tasks can be readily subdivided among multiple processors and where the extent of the computations slows the application to in undesirable degree when performed by a single CPU is a logical candidate where budgets are not a constraint.
The original writeup on 'data path' cited the computation of a single transcendental function (such as ARCTAN) as an example of a procedure which would not particularly benefit from multiple CPUs. That remains true. Since the multiple CPU board will not be GENERALLY useful, how come, you may ask, are we threatening to build such a board?
For THREE reasons, we answer, only TWO of which we are willing to share with you this issue (us girls need to keep some secrets, after all!). First, since the board in question is an add-on to the basic Dtack board in the same sense as the memory expansion board, there is absolutely NO requirement that it be GENERALLY useful! We can see the university profs salivating over the thought of 9 to 13 68000 CPUs at a price equal to a third that of ONE Charles River Data system!
The second is for promotional purposes. Is there an editor of any personal computer rag, including the hallowed BYTE, who could resist publishing a product release featuring a photo of a 9 to 13 CPU 68000 system? The publicity alone would MORE than pay for the (minimal) cost of developing the system!
HERE IS WHAT YOU MUST DO:
We would like to thank the appx. 387 people who have taken the time to patiently explain to us EXACTLY what we have been doing wrong. They explain that as soon as we correct the problem, we will naturally immediately take over first place in the marketplace from Tandy, never mind the puny upstarts like the fruity company and Admiral (Captain? Boatswain?).
Page 13, Column 1
Aside from the few hardware freaks who want us to mount DMA ports (but without a price increase), most of these helpful suggestions involve software. Now, we would be the LAST to suggest that any of these suggestions are self-serving but it is a maybe coincidental fact that the Pascal programmers tell us that we absolutely MUST support Pascal. The Fortran types CERTAINLY DON'T suggest that we support COBOL (maybe you can guess what language they suggest we support?).
What is even worse, we receive LOTS of suggestions that we provide UNIX, C, MODULA 2 (Wirth's successor to Pascal) and perhaps a selection of OTHER incredibly complex operating systems. Mind, these suggestions come from intelligent persons who are mostly friendly towards what we are doing and who wish for us to succeed (in the way they suggest, of course). We have decided that it is time for the worm to turn!
THE REBELLION OF THE WORM:
WE are going to tell YOU what WE feel is necessary for Dtack Grounded to succeed as a business enterprise. But before we do that we have a complaint to air!
Why is it that SOFTWARE houses are permitted to happily go their way peacefully, with no one EVER complaining that they do not manufacture hardware? But a HARDWARE house like us is CONSTANTLY criticized for not providing (in the judgement of the complainer) enough software? That is a double standard, folks!
FROM THIS INSTANT FORWARD, we will NOT ACCEPT criticism over our lack of software unless the critic provides documentation to prove that he has ALSO CRITICIZED A SOFTWARE HOUSE FOR NOT MANUFACTURING HARDWARE!
End of complaint. Now we are going to tell you our business plan: we are, for the remainder of the calendar year 1982, going to design and manufacture several different 68000 systems. Although we are not going to turn our backs on software, we are definitely going to put the software at a lower priority FOR NOW than the development of hardware. As we write this (key this?) the 128K expansion board layout is nearly completed. There wasn't room for a clock after all, unfortunately.
The 68000 board that we are now making, along with its 128K expansion board, is what we feel is the correct product for serious software developers and for serious business and scientific applications. It is also priced low enough that some of the people who want to get 'into' a new microprocessor early in the game can, and have, done so.
However, no ONE hardware design can be all things to all men. Either we brush off our blue suede shoes and try to flim-flam you into buying the WRONG product or we manufacture the RIGHT product. Since we do not own a single pair of blue suede shoes, we are going to try the LATTER approach. We need a SMALL, CHEAP 68000 board for two groups: those who want to educate themselves about the next generation of microprocessors and those who want to run Microsoft BASIC very, very swiftly.
Page 13, Column 2
WHILE YOU WEREN'T NOTICING, we have hooked our 60000 board into BASIC on both the Pet and the Apple II. By using PETSPEED or TASC, respectively, your BASIC programs are guaranteed to run 5 to 12 times faster! We think THAT FACT ALONE justifies the purchase of a 4K or 12K Dtack board. . But since very little memory is required for this boost in performance, we can build a SMALLER board and leave out the expansion port, the data and address buffers and make a really cheap 68000 board. Certainly less than $400 INCLUDING the 8MHz 68000! And we are going to do just that. And the very SAME board will serve, for now, very nicely as a vehicle for self-education.
As we mentioned in an earlier issue, Motorola has announced the 68008 which is identical to the 68000 except for an 8 bit data bus. Those of you who can see past the nose on your face will recognize that there will be about 37 companies manufacturing Apple II boards using the 68008, boards which will not have any memory on board (pun intended) because they can use the Apple II memory space (DMA AT LAST!) and therefore will be small enough to be contained ENTIRELY within the case of the Apple II. WE are ALSO going to build such a board just as soon as the 68008 becomes available!
The constraints on that board will be the availability of the 68008 itself, which will hopefully be available in limited production quantities around Oct. of this year. If there is a problem with the first mask, that will slip to about Jan. of next year. To give you a reasonably accurate idea of the status of the 68008, we have just asked Motorola for a pinout of the package and have been told that it is not yet available!
But when that part IS available, we will sell the Apple II version of the board for about $200 to $250, depending on the price of the chip itself. Remember, that board will not be available until five to eight months from now!
WHOOPEE! DMA AT LAST!
Naturally, since each word fetch will require two microseconds, the 68008 board will be MUCH MUCH slower than our other boards which have their own on-board memory. But the DMA freaks, who have clearly decided that speed equates to DMA regardless of the facts, will be enormously pleased when this board becomes available.
And WE are ALSO pleased because a 68008 fetching instructions at two microseconds per word will be able to monitor keyboards even more efficiently than the 6502, and will therefore be a better I/O processor for our 12.5 MHz 68000 board with its attached 8087 math processor!
Page 14, Column 1
A SHORT DIGRESSION:
If you want one of these boards NOW, you do not (we are told) have to wait until the end of the year to obtain one. We have heard from several sources that an outfit in England has designed an Apple II board which has a 68000 and some LSTTL logic circuitry to fool the 68000 into working directly with the 8 bit Apple II bus. We have also heard that you will shortly be able to buy this board in this country for a mere $695 retail. If you have the right connections you eight even be able to pick one up at the wholesale price of five hundred and something dollars! Unfortunately, we cannot give you further details. This is NOT because we are unwilling, as competitors, to provide such details. It's just that we do not HAVE more details, honest! End of digression.
So, right around the turn of the year we should have the following HARDWARE products available;
- The 92K board we have now,
- The 128K expansion board we ALMOST have now,
- A small, cheap, fast non-expandable board,
- A dynamic RAM version, 124K or 252K, parity optional, reliability slightly questionable,
- A 'mother board' holding 6 to 12 small, fast, reliable boards,
- A 68008 board to fit inside the Apple II,
- A math processor accessory board, conditions permitting.
WE THINK THAT PRETTY WELL COVERS THE POSSIBILITIES except for the MINICOMPUTER VERSION of the 68000, which is being introduced by at least 100 other companies. Therefore, WE need not bother with such a product! (THIS is the product that all of our 'friends' have been, in effect, suggesting we produce!)
DUMMY! YOU ARE COMPETING WITH YOURSELF!
RIGHT! Now let us tell you how STUPID that is! In the late 1950s and early 1960s there was this company named Heathkit. Heathkit at that time was experiencing rapid growth and high profitability. They sold via direct mail exclusively. Their products arrived in plain brown cardboard boxes. They INSISTED that you PAY for their products, either in advance or via COD. And they COMPETED WITH THEMSELVES!
Page 14, Column 2
Heathkit made 6 watt audio amplifiers, 12 watt amplifiers, 20 and 25 and 40 watt amplifiers. They made amplifiers with built in preamplifiers and amplifiers without preamplifiers and they made the preamplifiers to work with the amplifiers that did not have preamplifiers. They made some remarkably ugly amplifiers and they made others with decorator designed bezels and gold anodized covers. And we tell you again: while Heathkit did this, they were experiencing HIGH PROFITS AND RAPID GROWTH!
Heathkit is still around, of course. But their business plan has changed. It has changed a whole bunch. We are not sure about their profitability (the company has changed ownership a couple of times lately) but they have certainly become 'with it'. They sell in retail stores, the boxes are shiny plastic with pretty colors, and they don't compete with themselves much anymore. While the more recent business practices of Heathkit are clearly in the mainstream of American marketing, WE think there is room for their OLD marketing techniques.
AND WE ARE GOING TO ADOPT those old highly successful marketing techniques regardless of all the well-meant advice we have received to do otherwise. Let us make this perfectly clear: we do not care WHO emerges as the successful vendor(s) of 68000 MINICOMPUTERS because we have NO INTEREST WHATEVER, either personally or professionally, in MINICOMPUTERS.
WHAT WE ARE INTERESTED IN is PERSONAL computers. Personal computers with a modern CPU such as the 68000. Unfortunately no such animal exists nor is there any evidence of such an animal on the horizon. ALL (we repeat: ALL) of the announced 68000 based computers are in fact minicomputers! Each of then features an incredibly complex operating system and each supports many of the most complex high level languages known to mankind (if one believes the advertisements, which WE don't!).
MANY PEOPLE ARE SUPPORTING the idea of using 'user friendly' operating systems such as featured in the XEROX STAR. Surprise! We are IN FAVOR of such a system. It is just that we do not think such a system will be generally affordable until near the end of this decade! That is MUCH too long a time period to worry about for a company which must justify its existence to its employees every payday.
But Apple Computer's LISA will apparently feature some such operating system for those what got the bucks. We predict that LISA will become popular among those wealthy companies which wish to bring the benefits of computerization to those members of its work force who are not your basic computer freaks (to use current computer jargon). Mobil Oil, DuPont, 3M: these are the companies who will be buying LISA.
Page 15, Column 1
Joe Hacker is NOT, repeat NOT, going to carry LISA home. In addition to the fact that LISA will not be personally affordable unless you had the foresight to be born into a wealthy family, Joe Hacker (and yours truly) is going to be PERSONALLY OFFENDED at the idea of placing a picture of a (paper) file on the CRT to indicate a disk file operation!
SO WHAT DO WE REALLY NEED?
We thought you would never ask. Our problem is, this industry is as inbred as the old Hapsburg monarchy. Carl Helmers talks to Jim Butterfield who talks to Bill Sieler who talks to Bill Godbout who talks to Sol Libes who talks to Jim Strassa who, of course, talks to Carl Helmers! That 0.1% (at most) of the people involved in this industry talk among themselves, write the magazines and newsletters, influence what products are to be built and, most importantly, COMPLETELY IGNORE THE NEEDS OF THE OTHER 99.9%! So the answer to the question posed above is, who are 'WE'?
'WE', in this particular case, are those people who are going to be using 68000 based personal computers THREE YEARS from now. Most of 'WE' do not NOW own a computer and do not NOW know how to program a computer! Maybe the anointed 0.1% need an elaborate, complex operating system but 'WE' need the simplest, most readily learnable, affordable and understandable computer language. 'WE' need BASIC! And because 'WE' have MORE MONEY clutched in our grubby little paws than Mobil Oil, DuPont and 3M all rolled together SOMEBODY is going to give 'WE' what 'WE' need!
WE REALLY NEED A 68000 PERSONAL COMPUTER!
SOMEONE is going to introduce a small, simple, crummy personal computer based on the 60000 at a small, crummy price and they are going to prosper mightily as a result. The only reason us folks at Dtack Grounded (or Digital Acoustics, if you prefer) are not doing this is that we do not have the necessary resources. We (not 'WE') are restricting our activities to the arena of the possible.
But what is desperately needed is a 1982 Trash 80, um, Trash 68 with a small, simple, cheap BASIC interpreter which works and a small, simple, cheap operating system which sort of works. If such a product is built around the 68000 it will not even need a $2.50 attached 8 bit processor to be successful!
Page 15, Column 2
BACK TO DTACK GROUNDED:
We wanted to be honest and tell you that what the mass market REALLY NEEDS is not what we are offering. Only the people who ALREADY have a substantial investment in a personal computer system such as an Apple II will need one of our products. Fortunately, that is a sizeable potential marketplace because, no matter what G.L. asserted last summer, Apple owners are NOT going to dump their Apple computers into a hole in the ground!
HERE IS THE REST OF OUR GAME PLAN: As we said, we are going to concentrate primarily on hardware for the rest of 1982. What we are trying to do is to take it VERY DIFFICULT for the software houses to ignore our existence. One VERY well financed firm has already chosen our board for use with their complex Apple II CAD system as an available option, although they are not ready to formally announce for about another month. And these people are finding the 'hooks' to UCSD Pascal! Unfortunately, their software will of course be proprietary.
But what our board really needs to become a viable mass market product is a 68000 BASIC. A usable BASIC combined with an affordable 68000 board would, we believe, be a very attractive product to many, if not most, current Apple II owners. Especially after they find they can address lots of memory without bothering with bank switching! Unfortunately for Dtack Grounded, although a couple of software houses have nibbled at this concept, we have not been able to set the hook!
BUT THE TIMING IS STILL RIGHT! Back in newsletter #1, mailed toward the end of July, we stated that the market window for the 68000 was "just opening". Well, the market window is STILL opening, slowly. It is not yet WIDE open. We think the reason for that is that the many companies who have announced 68000 systems have announced minicomputers with incredibly complex software systems, and incredibly complex software systems take TIME to develop! Anybody see LISA lately?
If anything, Dtack Grounded was TOO EARLY into the marketplace! But we would much rather be early than late. And we are building up some assets. Believe it or not, the last nine issues of this newsletter contain quite a bit of useful information about the 68000, information which is not in print elsewhere. And there is more software available right now than most people give us credit for. Of course, you will have to purchase a BASIC compiler to take advantage of that software, but what the hey!
AFTER THE ORGY OF BOWL GAMES ON JAN 1, 1983 we should be all out of hardware projects. If between now and then some software house writes a 68000 BASIC which works with our board we will have the Devil's own sweet time trying to manufacture that hardware fast enough. If that BASIC has NOT appeared, we will then have time to write it ourselves! And, in our opinion, the 68000 market window is opening slowly enough that we will STILL be on time when the window is wide open!
Page 16, Column 1
AFTER AN INVESTMENT of several years and several million dollars, Modcomp has just deep-sixed the iAPX432 and switched to a DIFFERENT 32 bit processor for their transaction processing system. Modcomps' chairman stated, "we are no longer doing R&D on the 432."
GOOD NEWS DEPT: We have just received the Phase Zero 68000 cross-assembler announced in issue #7. Here is the short-form description, supplied by Phase Zero:
"ASSEM68K is a disk-based MC68000 cross assembler for the Apple II. It runs under DOS 3.3 in an Apple II or II+ with 32K or memory and one Disk II disk drive.
"All Motorola assembler opcode formats are supported, plus a complete selection of pseudo-ops including some which give more addressing-mode flexibility than the Motorola assembler. Alternate mnemonics are fully supported. The SHORT pseudo-up allows Dtack Grounded customers to make full use of the DG 64K zero-page (as opposed to the normal two 32K zero-pages). The price is $95 including postage and handling."
PHASE ZERO LTD.
2509 N. Campbell
Tucson, AZ 85719
ACKNOWLEDGEMENTS: Apple; singular, II and soft are trademarks of the Apple Computer Co. Pet and CBM are trademarks of Commodore business machines. UNIX is a trademark of Ma Bell. DTACK GROUNDED is OUR trademark. CP/M is a trademark of Digital Research of CA. Is Heathkit a trademark or a company name?
SUBSCRIPTIONS: You can subscribe to this rag (lucky you!) by sending $15/6 issues U.S. and CANADA or $25/6 issues elsewhere. Tell us which # issue to start with. Make the check out to DTACK GROUNDED. The address is:
DTACK GROUNDED
1415 E. McFadden, Ste. F
SANTA ANA CA 92705
REDLANDS WILL BE BACK! Next month we hope to print a threaded 'programmable calculator-like' language featuring very little interpretive overhead and with user-extendable macros. Stick around!
WE RECENTLY RECEIVED A LETTER FROM DR. JSK, which stated in part: "You wrote about reading Poe in reference to the inversion of the meaning of 'No More Dr. Jeckyl' used in Datamost's advertisement of Randy Hyde's new book on assembly language for the 6502. Edgar Allan Poe (he was the Poe you meant, no?) had been dead 37 years when Robert Lewis Stevenson had published his story 'The Strange Case of Dr. Jekyll and Mr. Hyde' in 1886."
Well, we have a perfectly logical explanation. You see,
Page 16, Column 2
CLASSIFIED ADS:
WANTED: suitable burial site for 60 foot tall gorilla. Need is urgent; respond soonest.
P.O. Box 432
Aloha, Oregon
WANTED: one professional software writer who has complained about software theft but has never had illegal possession of other authors' software. Send resume to:
Diogenes
the Parthenon, Athens
LOST: corporate integrity. If found, return to IBM. There is no reward; you were not supposed to notice it was missing.
Dept. of Permanent Math Errors
Armonk, NY 00.001
LOST: 24 month lead time in 16 bit microprocessor marketing. If found, return to:
J. Van Poppolen
V.P. Marketing
A National Semiconductor Company
POSITION AVAILABLE: newsletter editor. Ability to connect words and sentences to express ideas in a coherent way essential. Some knowledge of 68000 microprocessor optional. Apply in confidence:
P.O. Box DG
Santa Ana, CA 92705
WANTED: incredibly sophisticated operating system which can be learned by an average hacker in one day and will fit onto one 37 sector 5-1/4 inch floppy. Contact:
Mr. O'Rourke
Fantasy Island
WANTED: swifter heels. Contact:
Hector
c/o Charon
the River Styx
WANTED: extraordinarily shoddy 68000 based personal computer. Punk keyboard. Lousy CRT display with poorer 'graphics'. Incredibly simplistic BASIC language (level 1/3). Audio cassette program storage. 32K initial memory configuration. Must sell for under $800 complete with cassette and monitor. Contact:
10 million potential buyers
Everytown, USA
