*** Please note, this page (and web site) are in early development.
Items are certainly not complete, and may be inaccurate.
Your information, comments, corrections, etc. are eagerly requested.
Send e-mail to Ed Thelen. Please include the URL under discussion. Thank you ***

IBM-7094

Manufacturer IBM
Identification,ID 7094
Date of first manufacture1962
Number produced 130 7094 I's, 125 709 II's
Estimated price or cost-
location in museum -
donor -

Contents of this page:

Photo Photo
IBM 7094

Placard
IBM-7094 - by Ron Mak


IBM 7094

The 7094 was IBM's most powerful scientific computer in 1963. It could perform 500,000 logical decisions, 250,000 additions or subtractions, 100,000 multiplications, or 62,500 divisions in one second. It had hardware to do double-precision floating-point arithmetic.

The computer gained considerable I/O bandwidth from its separate data channels with direct memory access, and so it was also used to run business and general-purpose applications. The 7094 had an operating system called IBSYS, and FORTRAN and COBOL compilers.

A typical system cost $3,134,500. IBM stopped selling them in 1969.

Manufacturer: IBM Memory technology: magnetic core
First introduced: 1963 Memory size: 32K 36-bit words
CPU technology: transistor Cycle time: 2 microseconds (0.5 MHz)

Sources:
  • C. Gordon Bell, et al. Computer Structures: Readings and Examples. New York: McGraw-Hill, 1971. pp. 515-523
  • IBM 7094 Principles of Operation. IBM Systems Reference Library, 1962
  • IBM 7090/7094 IBSYS Operating System. IBM Systems Reference Library, 1964
  • IBM 7090/7094 IBSYS Operating System Version 13 FORTRAN IV Language. IBM Systems Reference Library, 1968
  • http://www.multicians.org/thvv/7094.html
  • http://www.multicians.org/mga.html
  • http://www.ibm.com/news/ls/1999/07/articles/sidebar_13.phtml
- Also see 7094 WORD.doc by Ron Mak

Architecture
The IBM 701, 704, 709, 7090 (709T) and 7094 were 36 bit parallel computers.

The IBM 701 stored 2 instructions per 36 bit word,
the rest stored 1 instruction per 36 bit word and were somewhat upward compatable (code written for a lower number machine might execute correctly on a higher number machine).

Special features
from Terry Harris 11/03/2003
Some years ago I had a chance to work on one of those old beasts. The Physics department at my old school got a 7090 donated to them. Some of us got it running as a project. The oil memory unit had a circulation pump that would move the oil through either a heater or a radiator depending on the oil temperature. As far as I remember the set point on the oil system was 104F.

We were lucky. Some folks at another university ( Purdue ) had gone through this before and were able to offer help. Also the machine came with a complete set of manuals. It took a few days just to cable it up. That thing had bundles of 92 ohm coax going all over the place.

One problem you had with the oil memory was metal fatigue in the plumbing. One night a small diameter tube in the circulation system developed a crack. Fortunately it was not a fast leak. Only lost a couple of gallons. Unfortunately the lab where the computer was located was right above the reading room for the branch of the university library in the Science building. They had a couple of books with translucent pages after that.

This was about 1970/71. There were several Universities then that were collecting old computers and nursing a little more life out of them. It seemed like a reasonable way to get some computing power cheap. In the early 1970's a 7090 or 7094 or a Univac 1103 was still a reasonably powerful machine and you could often get them for hauling them away.

Hmmm - I think the Computer History Museum IBM-1620 Restoration Crew said that the 1620 air cooled memory box had a thermostat set of 104 degrees F. - and the unit would think about executing programs until the memory box was up to temperature?

I will try to verify that today.

from Gordon Bell, 10/26/2000, see web site
BTW: the IBM 7090 core was in oil. Not sure about Stretch, but it probably was too as it pioneered technology that the 7090 exploited.
Ah Yes, the early magnetic core material was quite temperature sensitive. The magnetic characteristics changed quite a bit with temperature, and magnetic flipping used energy and warmed them up. I guess the oil helped stabilize their temperature. I remember magazine articles about the major efforts to find less temperature sensitive core material - which would eliminate the maintenance nightmare of having the core stacks in a tub of oil. Later core stacks (after about 1960) had the cores in air. :-)) A machine I worked on in 1960 had core stacks in air, but the core drivers had temperature sensors to adjust their current with the air temperature. Yes, the good old days were "interesting".

from Coslet, Tim July 2004
I've done a bit of research and found that both the 7090 and Stretch did use the same core memory unit:

IBM 7302 - IBM 7030 Core Storage (16384 - 72-bit words: 64 data bits & 8 ECC bits)

IBM 7302 - IBM 7090 Core Storage (32768 - 36-bit words)

It was probably a different model with some different logic (as Stretch used 8 bits of each 72 for ECC and the 7090 used all the bits as data), but it would have been the same core, the same heated oil bath, the same cabinet, etc.

The following is from Peter A. Goodwin - December, 2009

Comment about oil-bathed core memory
I think the fix for the oil-bathed core memory in the 7090 was nothing more than installing some deflectors to alter the path of the oil currents within the tank.

The following is from Peter A. Goodwin - December, 2009

Comments about circuits, cabling, packaging, ...
Robert is correct regarding the machine packaging. While I never heard the term "Rolygon", the 7090 was packaged in frames about three feet wide and maybe six feet tall. Remember that the 7090 was derived from the STRETCH (7030) machine. STRETCH used a 16K memory of 72-bit words; the '90 cut that down the middle, to get 32K 36-bit words; the memory still delivered 72 bits, but the '90 interface selected which half to use at any given time. In the frame, the contents rolled out from the enclosure, and then separated into two "pages" each of which had backplanes in four quadrants, with wire-wrapped interconnections. I believe that the backplane assemblies were identical to those used in the "Cube" (e.g., 1401 and 1610) systems. The "hinge" area at the back where the two pages were joined contained the power supplies. Of course, this meant that all of the cables had to pass over hinged cable trays in order to manage the flexing. By the way, cabling in the 7090 was a manufacturing nightmare. Every signal had its own coaxial cable; 36 cables per 36-bit bus. The 7090 used germanium transistors because that was what was available when the STRETCH was designed, and current-mode logic (what we later called "ECL") in order to maximize speed; even then it had a 2.18-microsecond cycle time. An ADD instruction took two cycles (instruction fetch, data fetch) to complete.

The following is from Peter A. Goodwin - December, 2009

Time Sharing adaptation to the 7090
The time-sharing machine that I worked on WAS at Mohansic. The guy who got the $7,000 prize was probably my boss, Andy Kinslow. An IBM Outstanding Contribution award. Andy is still alive -- at least he was a couple of years ago or so -- lives in western Connecticut near the New York border. He got a prize, I got a prize (not nearly so much) and two other guys got prizes for this. The system took the CTSS approach as a starting point. CTSS added memory relocation and protection registers to the CPU and trapped the attempted execution of all I/O or supervisory instructions. Our machine went further that this: It used high-speed (relatively!) drum storage for inactive program residence (program swapping) and a 1301 disk for the program library. I added relocation and protection registers to the 1301 data channel and to data channel A (card reader, card punch, printer, and a bank of 729 tape drives), and added other hardware that would prevent a user program from grabbing control of the machine. The result was that we created a set of virtual 7090s using an 8K-word resident supervisory program in a 32K-word system. You could run any program that didn't absolutely depend upon CPI-I/O synchronization (like the CPY instruction on the 704) or invoke compatibility-mode alterations (which would make the 7090 operate like it was a 704). You could even run standard IBM machine diagnostics in the time-sharing environment! (Of course, a memory diagnostic would only find 24K words available.) See H. A. Kinslow, The Time-Sharing Monitor System , AFIPS Spring 1964, http://portal.acm.org/citation.cfm?id=1464052.1464092&coll=ACM&dl=ACM&CFID=64753258&CFTOKEN=53621636 .

Historical Notes
-

This Specimen
-

Interesting Web Sites

Other Information
I was curious why BIG IBM systems had so many tape drives
for instance,
http://www.ed-thelen.org/comp-hist/BRL61-0549.jpg
linked from
http://www.ed-thelen.org/comp-hist/BRL61-ibm7070.html#IBM-7090
with 12. - Effective sorting can be done on six or less
John Van Gardner commented: (July 2008)

The 7090 at Lockheed Georgia had 10 729 tape drives.

Drive 1 contained a tape with the IBSYS Operating System.
Drive 2 was the input tape containing programs and data.
Drive 3 was for the printed output sent to a peripheral print system.
Drive 4 was for punched card output sent to a peripheral punch system.
That left 6 drives for the programs to use as needed for work space and to create tapes to be sent to the tape library.

The following is from Peter A. Goodwin - December, 2009

placed here because of an "altercation" about the usefulness of SAGE vs jamming - this hint might eventually help determine if/how_much SAGE input could handle jamming and prevent/reduce "GIGO" (Garbage In Garbage Out).
BMEWS: There were two 7090s in each installation. Raw radar data were digitized and fed to the computers for threat analysis. The radar data consisted of target position -- elevation, azimuth, and range -- radial distance = radar-echo-return time), and range rate (a Doppler-effect measurement). GE made the radars, Sylvania made the gear that performed the digitizing. IBM's ASDD Mohansic Lab made the 7090 radar-data real-time data channel boxes. Both 7090s operated on all data, but only one reported to NORAD; the idea was that if either machine failed, the other would be ready to perform actively. The 7090 program that performed threat analysis had operating levels to avoid being overwhelmed with incoming data: If the data stream was moderate, it would perform thorough analysis; if the data stream became heavy, it would only perform cursory analysis, the theory being that by that time, the world was going to hell anyway so who cared? The 7090 program was written in assembly language in order to optimize size and speed. The instruction code was written to be non-volatile, and every couple of seconds the system would perform a Hamming checksum to ensure that nothing in the instruction code had been altered.

If you have comments or suggestions, Send e-mail to Ed Thelen

Go to Antique Computer home page
Go to Visual Storage page
Go to top

Updated December, 2009