Which Microprocessor
by Hal Chamberlin, Box 295, Cary, NC 27511

At this time there are three microprocessor chips or chip sets readily available
to the hobbyist: the 8008, the 8080, and the IMP-16. The first two were
pioneered by Intel and the last is a National Semicon ductor invention. Chips
and/or kits utilizing each of the three microprocessors are available from at
least two sources catering to hobbyists as of this writing. This level of
availability and popularity is not even approached by other microprocessors,
therefore this discussion is being confined to these three.

Comparing computers is like comparing people: the conclusions depend on the
application, the circumstances, and personal preference. The comparisons will

The comparisons made will be based on use of the microprocessor as a general
purpose computer.  For our purposes a general purpose computer is one which has
read-write memory for the bulk of its storage,  which is expected to run a
variety of programs, and for which the end use is the development and execution
of programs written by the user. General purpose computers are also expected to
be able to control a variety of input-output equipment. Instruction sets will be
compared on the basis of assembly language programming. Speed will be compared
on the basis of the time necessary for the machine to complete a non-trivial
task. Complexity will be compared on the basis of ease of understanding
microprocessor operation as  well as the sheer number of parts required to
implement a system. Finally, cost will be compared on the basis of minimum
systems capable of assembling programs for themselves given the existence of
suitable I/O devices.

Before getting into comparisons, we will take a brief look at the leading
features of each microprocessor. Then the comparisons will be made in each
performance area elaborating on individual features as necessary.

The Intel 8008 Processor
The 8008 was the first microprocessor to be introduced and the first to be
available to the hobbyist. It has an 8 bit instruction and accumulator length.
There are essentially only two memory addressing modes: immediate, and zero
displacement indexed. Subroutine and branch addresses are full length absolute,
allowing branching anywhere with one instruction. Subroutine return addresses
are saved on an internal 8 level stack which puts a 7 deep restriction on
subroutine nesting. Much of the instruction set power is derived from the six
additional 8 bit index registers which may count, save, or address memory. The
maximum directly addressable memory is 16k bytes; in addition, 8 input and
24 output devices may be directly addressed with a one byte instruction. CPU
speed is modest ranging from 20  microseconds for a register operation to 32
microseconds for a memory operation to 44 microseconds for a jump or call. A
selected chip, the 8008-1, reduces these times to 12.5, 20 and 27.5 microseconds
respectively. A single level of interrupt is provided but external hardware
is necessary for complete status saving during interrupts. Interfacing the chip
to the rest of the system is fairly involved and requires from 20 to 70 TTL
packages depending on the system performance desired. The lower figure will
barely function while the higher one includes a console, complete interrupt
system, and dynamic memory interface with direct memory access capability. Most
of the interfacing complexity can be blamed on overzealous designers trying to
make-do with an 18 lead package. Present cost to the experimenter ranges from
$40 to $80 with the "dash one" version bringing roughly 50% more.

The Intel 8080 Processor
The 8080 is Intel's sequel to the 8008. Basically it has more of everything. The
instruction set contains all of the 8008 instructions making it upward
compatible at the  assembly language level. Major additions to the instruction
set include direct load and store of the accumulator, double precision (16 bits)
add and increment for address calculation, and a pushdown stack of indefinite
length in memory thus allowing unrestricted subroutine nesting. Addressable
memory has been increased to 64k bytes and addressable I/O devices have been
increased to 256 inputs and 256 outputs at the expense of 2 byte I/O
instructions. Execution speed has been considerably improved also. Register
operations take 2 microseconds, memory operations require about 3.5
microseconds, and subroutine calls consume 8.5 microseconds. Interrupts work the
same way as on the 8008 but everything required for complete status saving is
provided as well as an interrupt enable/disable flag. Interfacing an 8080 is
generally regarded as being simpler than interfacing an 8008. There us only a
slight improvement in the minimum system, about 15 chips, but a full-bore system
may be cut in half to 35 chips. The 40 lead package allows a separate 16 bit
address bus and 8 bit data bus, as well as simplified timing and control.
Present cost to the hobbyist is about $160.

The National IMP-16
The IMP-16 is one of the older microprocessors and for a long time the only one
with a 16 bit wordlength. The programmer is supplied with four 16 bit
accumulators and a 16 word stack. The instruction set is typical of many 16 bit
minicomputers, and in many ways resembles that of a NOVA. Four general address
modes are provided, base page direct, program counter relative, and indexed
using either accumulator 2 or accumulator 3. In addition, LOAD, STORE, JUMP, and
CALL can be indirect addressed using any of the addressing modes to get to the
address pointer. Two memory modification instructions are provided, ISZ
(Increment memory, Skip if Zero), and DSZ which allows much counting and
indexing to be done in memory freeing the registers for arithmetic. The stack is
used for subroutine return addresses but can also be used for saving registers
and status. A unique feature is the availability of an extended instruction set
chip which provides automatic multiply, divide, double word add and subtract,
and byte manipulation. The CPU can address 64k words but this should be held to
32k if the byte instructions are used. The I/O instructions can also address 64k
devices. Another unique feature is that several bits of input and output are
provided by the microprocessor itself making communication with a teletype
possible without any interface at all. Speed is good ranging from 4.2
microseconds for a register operation to 7 microseconds for a memory operation.
A multiply takes about 160 microseconds which is still considerably faster than
a software routine would be. The IMP-16 provides two priority levels of
interrupt and all of the hardware necessary for complete status save/restore.
Interfacing is conceptually simple and requires 25 to 50 packages depending on
system sophistication. Part of this number is due simply to the fact that 16
bits are to be handled rather than 8. The microprocessor is in the form of five
24 lead packages which for the most part are simply wired in parallel. The
extended instruction set resides in a sixth package. Present cost of the
standard chip set is about $160. The extended instruction set chip is available
only from National at this time for $80.

Comparisons - Instruction Sets
One of the most important performance areas of a microprocessor is the
instruction set. A good instruction set should be well organized so that it is
easy to learn, powerful so that complex routines can be coded with a small
number of instructions, memory efficient so that complex routines require only
small amounts of memory, and time efficient so that only a small number of
memory cycles is necessary to complete a task. In addition, performance should
be equally high on both character oriented tasks and numerically oriented tasks.

Instruction set organization is best on the 8080 closely followed by the 8008
with the IMP-16 being somewhat disorganized. Consequently, the beginner will
find the 8008/8080 the easiest to learn. Experience has shown that beginners
prefer simple instruction sets and that they retain a certain "fondness" for
their first machine long after they have graduated into much more sophisticated
endeavors. The experienced programmer however should experience little
difficulty keeping the little 'quirks, distinctions, and special cases straight
when working with the IMP-16.

Instruction set power is best on the IMP-16, followed by the 8080 with the 8008
a distant third. Based on actual experience, it may require as few as one half
as many IMP-16 instructions to program a task as 8008 instructions. The 8080
falls about midway between the extremes. There are many reasons why the 1MP-16
is superior. Memory addressing is much more flexiable due to the four addressing
modes and indirect addressing capability. An addtional advantage is that the
arithmetic word length is the same as the address length. Since the return
addresses are put on a stack in all three machines, multiple entrypoint
subroutines are easy but the IMP-16 also allows multiple return points (return
to CALL+1 on error, CALL+2 otherwise, etc.) with no additional instructions. The
8080 is a big improvement over the 8008 because the registers may be saved on
the stack when they are used by a subroutine and then restored unaltered upon
return. This allows subroutines to be called as needed without regard to which
registers they may destroy. Note, however, that this capability may be added to
the 8008 quite simply The direct load and store instructions of the 8080 reduce
the number of lines of code in a program.

Memory efficiency of the instruction set is best on the IMP-16 but is closely
followed by the 8080. The 8008 is not as bad as might be presumed but is
definitely inferior. In terms of numbers, the 8080 may require 10 to 15 percent
more memory bits and the 8008 20 to 40 percent more. Note that these figures are
based on optimized programs written by experienced programmers. The spread can
be much greater with inexperienced programmers or hastily written programs. It
is also interesting to note that instruction set organization and memory
efficiency are usually conflicting requirements. This is because many of the
lesser used possible operation combinations have been culled from a memory
efficient set in order to reduce the number of bits required to encode the
instruction. Implied operands are also utilized in order to free up bits for
other uses. Experienced programmers are able to plan ahead and avoid having
these restrictions become restrictive. The 8008 and 8080 are as good as they
are because many of the instructions are a single word (8 bits) long whereas the
minimum instruction length in the IMP-16 is 16 bits. This is somewhat offset by
the three word (24 bit) instructions of the 8008 and 8080 which in most cases
would only require 16 bits in the IMP-16. A fringe benefit of high memory
efficiency is that the shorter programs will load faster regardless of the
loading method.

Time efficiency is by far the best on the IMP-16 with the 8008 a distant second
and the 8080 a slightly poorer third. On a classic minicomputer, a machine cycle
was the same as a memory cycle in most cases. As a result, a time efficient
instruction set meant a faster machine without faster hardware. Microcomputers
on the other hand may have very few of their machine cycles being memory cycles.
As a result, time efficiency may have little relation to actual machine speed
but does represent the potential speed with an optimized CPU. Time efficiency
can be important in multiprocessor systems with a shared memory where more
memory cycles increase the probability that a CPU will have to await its turn.
The IMP-16 has a high time. efficiency mainly because twice as much data is
fetched in each memory cycle. Further improvement is due to the instruction set
power, requiring fewer instructions to be fetched. The 8080 has poorer time
efficiency than the 8008 mainly because the stack is in memory. A subroutine
call, for example, requires 5 memory cycles, 3 to fetch the instructions and two
to stack the return address.

Historically some minicomputers were better at handling character oriented tasks
and others were well adapted to number crunching tasks. Microcomputers are no
exception. Most micros have been optimized for character handling because of
expected high usage in terminals and the 8008 and the 8080 belong to this class.
The IMP-16 on the other hand is much better at numerically oriented tasks and
was aimed more toward machine tool control and industrial monitoring.
Interestingly, use of the extended instruction set on the IMP-16 greatly
improves both character handling and arithmetic capability.

How About Running System Software?
One performance area of interest to hobbyists is the suitability of a machine
for running a BASIC system. The IMP-16 and the 8080 are about equal in their
ability to compile BASIC quickly but the IMP-16 without the extended instruction
set may execute BASIC twice as fast. This is due mainly to the all floating
point arithmetic that BASIC requires. The extended instruction set may double
the speed again if a lot of multiplies and divides are done. The 8008 can of
course run BASIC also but compile and execution speeds are ikely to be one
tenth of the 8080.

One other property of an instruction set is the ease with which it may be
assembled, either by hand or with an assembler program. In this respect, the
8008 comes out on top with the 8080 next and the IMP-16 last. Use of the
mnemonics and format recommended by the manufacturer is assumed in making this
comparison. 8008 code is easy to hand assemble because the octal notation used
corresponds to the various fields in the instruction word. Assemblers for 8008
code can also be quite simple because instructions require at most one operand
and very few instruction formats exist. Further simplification results from all
addresses being absolute and all mnemonics being three characters. 8008
assemblers run on an 8008 can be as small as 2.5k bytes for a limited
implementation but 4k bytes is more realistic when providing an easy to use
assembler. If the hexadecimal notation recommended by Intel is used with the
8080, hand assembly is definitely more difficult. The assembler also has a
tougher time with the two operand format and other niceties defined for the
8080. The Intel version of the 8080 assem bier requires 8k bytes but it should
be noted that it provides macro capability. Hand coding and assembling for the
IMP-16 is harder yet due mainly to relative addressing considerations and a
wider variety of instruction formats. National's version of the assembler
requires 4k words and can produce relocatable object code and handle external
symbols.

Is Speed Useful?
Speed in a hobby computer system can be a two-edged sword. A high speed
microprocessor requires higher speed in other system to components such as
memory in order to realize its higher speed. An 8008 for example can run at full
speed with memories as slow as 3 microseconds access but the 8080 will have to
wait on memories slower than 520 nanoseconds and the IMP-16 requires 420
nanoseconds. If the ready line is used on the 8008 and 8080 to permit the use of
slower memories, the wait will be in increments of whole machine cycles which is
4 microseconds on the 8008 and 500 nanoseconds on the 8080. Thus if memory is a
tad slow, one cycle will be added to each three cycle memory access sequence
slowing the system down an average of 25 percent to 30 percent. The IMP-16 does
not have a ready line, rather the user stretches one of the clock periods in a
cycle long enough to permit memory access. This scheme has the advantage that
the stretch can set to the exact amount needed. The higher time efficiency of
the IMP-16 reduce the performance impact of a slow memory as compared to the
8080.

How Complex is the Interface?
The ranking on interface complexity is 8080 (least), IMP-16, and 8008. The
comparison is based on sophisticated general purpose implementations having a
complete I/O interrupt facility, software console using ASCII I/O, an a
generalized input/output memory bus allowing simultaneous direct memory access
without affecting the CPU. The ranking is the same whether parts count or
conceptual complexity is being considered. The difference between the 8080 and
the IMP-16 is primarily due to the wider word of the IMP-16 and less confusing
discussion of chip interfacing in the Intel manual. The 8008 is just plain
difficult to understand and interface correctly but once that is done, either by
the user, a magazine, or a manufacturer, the system should operate just as well.

Software?
Software support is often a big issue among industrial users of microprocessors.
Unfortunately, the majority of the software they are fighting over is
unavailable to the hobbyist because of high prices. It is not unusual for a
program such as an assembler to cost as much as a handful of microprocessors.
8008 and 8080 users can look to Scelbi and MITS for some software at reasonable
prices even if they did not purchase their machines from these sources. National
has an excellent body of software for the IMP-16 but the package price is $200
for object tapes and source listings. The hobbyist will have to depend on
himself, kit manufacturers, and publications for most of his software in the
near future. Ultimately, the level of software support will be directly
proportional to the popularity of the microprocessor.

And Finally...
There are a number of other performance areas that are only of minor interest to
hobbyists. Although there is a large spread in the maximum memory size, all
three machines are likely to have ample addressing capability for the hobbyist's
memory budget. The same may be said relative to addressable I/O devices. Power
supply voltages and power consumption are also usually of minor importance. All
of the microprocessors can be successfully operated from standard +15, +5 and
-15 system supply voltages using simple, . fail-safe, zener regulators. Package
size and pinout are unlikely to be factors in hobbyist use. This brings us to a
comparison of overall system cost. First, the spread in chip cost is roughly
from $50 to $150, so the spread in system cost would be $100 at most. An 8008
requires more interfacing circuitry however which reduces the spread somewhat.
After enough memory to do assemblies or run BASIC and a few I/O devices are
added, the $75 difference left may be small compared to the total investment.
Nevertheless, an 8008 system will be the least expensive followed by an 8080
system closely followed thereafter by an IMP-16. Which microprocessor for you?
The answer still depends on the ap plic at ion , circumstances, and personal
preference, but hopefully the decision can be made with more authority after
reading this article.