In the early '60 IBM Germany started a project with the codename "TINY". The idea was to develop a
computer for the small customers. This system had a small punchcard and was electro-mechanical.
"TINY" was later announced as the IBM System/3000. IBM withdraw this system quickly from
marketing when serious problems surfaced with the cardreader. For more history details of this
system please read this interview with Ralph Mork.

The early end of this system was not the end IBM's idea to develop a small system for the lower end
of the market.
Mid '60 Larry Wilson developed a prototype of the 3.7 system with Roy Harper (electronic) and
Greg Tobin (mechanical) of the San Jose Lab.
In early 1966, it was decided that the experimental program be developed in San Jose should be moved
from that location. After some study the program was moved to Rochester. Larry was not convinced that
Rochester had the technical capability to to undertake a total system project. Dick Trachy invited Larry to
move temporarily to Rochester at Rochester's expense. He did and was given an office in the lab area near
Harry Tashjian whom was the appointed project manager. In only a few weeks, Larry notified Dick that
he was satisfied that Rochestor could handle the project and he moved back to San Jose.

Development work continued using the code name project 3.7 which had been coined by Larry.
The design goals were quite simply in concept but not easily to achieved. Rochester had to develop a stored
program data processing system that satisfy Unit Records customers at a price that would be competitive
with the existing equipment and offer more capability. Most of these customers are small business companies
with limited resources.

The introduction of the small card to replace the 80 column card which was a standard unit record media was
key to the program primarily because of the huge maintenance costs of the 80 column card machines and
their large size.
Work continued on the project through 1966. Early in 1967 a task force was formed consisting of key people
from all IBM divisions, including World Trade, that would be impacted by the introduction of the system in
the market place. The task force spent about 3 weeks in Rochester examining all aspects of the program.
They recommended continuation of the program.
In October 1967, the development program responsibilities were consolidated under a Systems manager
Dick Trachy, who was responsible for all aspects of the development and announcement of the system.
Reporting to him was Harry Tashjian in charge of Engineering, Bob Webster, in charge of  Programming and
Carl Gebhardt in charge of Product Planning, which included forecasting, pricing etc.  

Here is a very nice picture of the IBM System/3 development team.  
Many other key contributors are not in the picture due to reassignments and were not present in Rochester at
the time the photograph was taken.
Some sources state that IBM HQ was not fully aware what was developed at Rochester.
Here is Dick Trachy's comments on this rumor:

""IBM HQ was well aware of our program. Frank Cary was Group executive at the time and all systems development
in the company reported to him.  I met with he personally, and with his staff, weekly reviewing all aspects of the
and resolving problems between divisions as they arose. The question of compatibility came up often and
was always
resolved.  There were dissenters on this question as well as others. In particular, the issue of the use of a
new media,
the small card, came up several times. It was considered to be a major problem by the Sales organization.
More than one study was made during the course of development on both subjects.""

The commercial name of the 3.7 system became "IBM System/3". The old project name can still be seen 
left under on the punch card: "IBM 3700".  

The name “System/3” was chosen after considerable work with the domestic and world trade representatives.
IBM considered numbering the system with a four letter number like the larger systems. This was rejected in favor of
the format that was finally chosen. “System/7” was popular because we were entering the 70s decade but
because a 7 is written differently in world trade countries it was rejected. "3" was selected because the system
had 3 main components: CPU, Printer and Card Reader.

Two S/3 models (a card and a disk system) where built by  IBM, one was used for product testing to
support announcement.  The system was designed with the compact MST-1  logic circuits.

More history details can be found in Emerson W. Pugh, Lyle R. Johnson and  John H. Palmer book
IBM's 360 and Early 370 Systems (MIT Press, Cambridge, 1991) pages 443 - 451.

The system was officially announced by IBM on 30 July 1969 in United States. Announcement in IBM World Trade
came later in the year.
The announcement included:
  • IBM 5410 Model 10 Central Processing Unit 
  • IBM 5424 Multi Functional Card Unit (aka MFCU)
  • IBM 5203 Line Printer 
  • IBM 5444 Disk Storage
  • IBM 5440 Disk Cartridges
  • IBM 5471 Printer Keyboard 
  • IBM 5475 Data Entry Keyboard
  • IBM 5486 Card Sorter (offline)
  • IBM 5496 Data Recorder 
Click here for the technical information released by IBM on the announcement date.
An 12 minute System/3 promotion movie, shot late 1968, can be seen here: Part 1 and Part 2.
Here are some nice model pictures of a pre-production S/3:  Pic 1 , Pic 2 & Pic 3

This rare picture was taken during the press announcement when all the focus of the photographers was on the new hardware.
It was the card deck for the RPG-II compiler. IBM's design point for the card system was 4K storage size.  They almost made
that but found it impossible. So IBM opted for 8K. This System/3 Card System was managed by Donald J. Kastella.

Here are two press reaction, one from Datamation after the S/3 announcement (published on the 15th september 1969)
and one from Design (published in 1970).

On 28 October 1969 IBM announced the IBM 5406 System/3 model 6 developed at Boca Raton.
The model 6 was the only model that had BASIC and was developed under supervision of Glenn Henry. 
This effort was originally intended for new small scientific computer, a follow on to the IBM 1130, but was switched
to use the S/3 model 6.  

In 1969 a scientific oriented machine was developed in San Jose and one day the developemnt team was told that
hardware had been canceled and that they had to use a new machine being developed in Rochester. 

The move of the BASIC project to the mod 6 was a real shock.

Don Kastella visited the team and give a presentation about the S/3.  The group couldn't believe what they were hearing,
a slow memory-to-memory architecture with no registers, a byte-wide datapath, etc.
That is, the direct opposite type of architecture they were working on.  To put scientific-oriented software on it
sounded impossible at first. But, then the challenge became a motivator for the group, make a good BASIC system in
spite of the feeble hardware. The whole group was transferred to Boca Raton to do the project.
Of course, it turned out that the S/3 architecture was fine and the BASIC software ran well on the Model 6.

On January 23 1970 the first IBM System/3 was delivered to Lasko Metal Products, Inc., in West Chester, Pa.
The IBM System/3 was produced at Rochester (USA), Vimercate (Italy), Fujisawa (Japan) and Boca Raton (USA).

Here are some good articles of the IBM Rochester lab and plant history and at the IBM Archives.
This article gives a very good timeline overview of IBM Rochester at the Rochester QCC website.