The Bendix G-15 computer was introduced in 1956 by the Bendix Corporation, Computer Division, Los Angeles, California. It was about 5 by 3 by 3 ft (1.5m by 1m by 1m) and weighed about 950 lb (450 kg). The base system, without peripherals, cost $49,500. A working model cost around $60,000. It could also be rented for $1,485 per month. It was meant for scientific and industrial markets. The series was gradually discontinued when Control Data Corporation took over the Bendix computer division in 1963.
The chief designer of the G-15 was Harry Huskey, who had worked with Alan Turing on the ACE in the United Kingdom and on the SWAC in the 1950s. He made most of the design while working as a professor at Berkeley, and other universities. David C. Evans was one of the Bendix engineers on the G-15 project. He would later become famous for his work in computer graphics and for starting up Evans & Sutherland with Ivan Sutherland.
The G-15 was a serial-architecture machine, one of several inspired by the ACE. It used a magnetic drum to simulate the recirculating delay line memory of other serial designs. Each track had a set of read and write heads; as soon as a bit was read off a track, it was re-written on the same track a certain distance away. The length of delay, and thus the number of words on a track, was determined by the spacing of the read and write heads, the delay corresponding to the time required for a section of the drum to travel from the write head to the corresponding read head. Under normal operation, data were written back without change, but this data flow could be intercepted at any time, allowing the machine to update sections of a track as needed.
This arrangement allowed the designers to create "delay lines" of any desired length. In addition to the twenty "long lines" of 108 words each, there were four more short lines of four words each. These short lines recycled at 27 times the rate of the long lines, allowing fast access to frequently needed data. Even the machine's accumulators were implemented as drum lines: three double-word lines used for intermediate storage and double-precision addition, multiplication, and division in addition to a one single-word accumulator. This use of the drum rather than flip-flops for the registers helped to reduce tube count.
A consequence of this design was that, unlike other computers with magnetic drums, the G-15 did not retain its memory when it was shut off. The only permanent tracks were two timing tracks recorded on the drum at the factory. The second track was a backup, as the tracks were liable to erasure if one of their amplifier tubes shorted.
The serial nature of the G-15's memory was carried over into the design of its arithmetic and control circuits. The adders worked on one binary digit at a time, and even the instruction word was designed to minimize the number of bits in an instruction that needed to be retained in flip-flops (to the extent of leveraging another one-word drum line used exclusively for generating address timing signals).
The G-15 had 180 vacuum tube packs and 300 germanium diodes. It had a total of about 450 tubes (mostly dual triodes).  Its magnetic drum memory held 2,160 words of twenty-nine bits. Average memory access time was 14.5 milliseconds, but its instruction addressing architecture could reduce this dramatically for well-written programs. Its addition time was 270 microseconds (not counting memory access time). Single-precision multiplication took 2,439 microseconds and double-precision multiplication took 16,700 microseconds.
One of the G-15's primary output devices was the typewriter with an output speed of about 10 characters per second for numbers (and lower-case hexadecimal characters u-z) and about three characters per second for alphabetical characters. The machine's limited storage precluded much output of anything but numbers; occasionally, paper forms with pre-printed fields or labels were inserted into the typewriter. A faster typewriter unit was also available.
The high-speed photoelectric paper tape reader (250 hexadecimal digits per second on five-channel paper tape for the PR-1; 400 characters from 5-8 channel tape for the PR-2) read programs (and occasionally saved data) from tapes that were often mounted in cartridges for easy loading and unloading. Not unlike magnetic tape, the paper tape data were blocked into runs of 108 words or less since that was the maximum read size. A cartridge could contain many multiple blocks, up to 2500 words (~10 kilobytes).
While the G-15 had an optional high-speed paper tape punch (the PTP-1 at 60 digits per second) for output, the standard punch operated at 17 hex characters per second (510 bytes per minute).
Optionally, the AN-1 "Universal Code Accessory" included the "35-4" Friden Flexowriter and HSR-8 paper tape reader and HSP-8 paper tape punch. The mechanical reader and punch could process paper tapes up to eight channels wide at 110 characters per second.
The CA-1 "Punched Card Coupler" could connect one or two IBM 026 card punches (which were more often used as manual devices) to read cards at 17 columns per second (ca. 12 full cards per minute) or punch cards at 11 columns per second (ca. eight full cards per minute). Partially full cards were processed more quickly with an 80 column per second skip speed). The more expensive CA-2 Punched Card Coupler read and punched cards at a 100 card per minute rate.
The PA-3 pen plotter ran at one inch per second with 200 increments per inch on a paper roll one foot wide by 100 feet long. The optional retractable penholder eliminated "retrace lines".
The MTA-2 could interface up to four drives for half-inch Mylar magnetic tapes, which could store as many as 300,000 words (in blocks no longer than 108 words). The read/write rate was 430 hexadecimal digits per second; the bidirectional search speed was 2500 characters per second.
The DA-1 differential analyzer facilitated solution of differential equations. It contained 108 integrators and 108 constant multipliers, sporting 34 updates per second.
A problem peculiar to machines with serial memory is the latency of the storage medium: Instructions and data are not always immediately available and, in the worst case, one must wait for the complete recirculation of a delay line to obtain data from a given memory address. The problem was addressed in the G-15 by what the Bendix literature called "minimum-access coding." Each instruction carried with it the address of the next instruction to be executed, allowing the programmer to arrange instructions such that when one instruction completed, the next instruction was about to appear under the read head for its line. Data could be staggered in a similar manner. To aid this process, the coding sheets included a table containing numbers of all addresses; the programmer would cross off each address as it was used.
A symbolic assembler, similar to the IBM 650's SOAP (Symbolic Optimal Assembly Program), was introduced in the late 1950s and included routines for minimum-access coding. Other programming aids included a supervisor program, Intercom, a floating-point interpretive system, and ALGO, an algebraic language designed from the 1958 Preliminary Report of the ALGOL committee. Users also developed their own tools, and a variant of Intercom suited to the needs of civil engineers is said to have circulated.
Floating point was implemented in software. The "Intercom" series of languages provided an easier to program virtual machine that operated in floating point. Instructions to Intercom 500, 550, and 1000 were numerical, six or seven digits in length. Instructions were stored sequentially; the beauty was convenience, not speed. Intercom 1000 even had an optional double-precision version.
The G-15 is sometimes described as the first personal computer, because it had the Intercom interpretive system. The title is disputed by other machines, such as the LINC and the PDP-8, and some maintain that only microcomputers, such as those which appeared in the 1970s, can be called personal computers. Nevertheless, the machine's low acquisition and operating costs, and the fact that it did not require a dedicated operator, meant that organizations could allow users complete access to the machine.
Over 400 G-15s were manufactured. About 300 G-15s were installed in the United States and a few were sold in other countries such as Australia and Canada. The machine found a niche in civil engineering, where it was used to solve cut and fill problems. Some have survived and have made their way to computer museums or science and technology museums around the world.
Huskey received one of the last production G15s, fitted with a gold-plated front panel.
See also 
- The Bendix G-15
- Bendix G15 computer
- Another G-15 reference
- Bendix G-15 documentation
- inside photo
- info page with photo
- Describes Harry Huskey's involvement with ACE
- Extensive G15 site list, photos & technical info