IAS computer

From Computer History Wiki
Jump to: navigation, search

The IAS computer (sometimes called the IAS machine; it seems not to have had a formal name, although some later contemporary documents call it 'MANIAC', a name used by the Los Alamos copy) was a very early electronic programmable computer. The group that designed and built it began the task in June, 1946, initial operation began in the fall of 1950, and it was formally dedicated on 10 June, 1952. Its most significant contribution was the vast list of first-generation computers, both in the US, and around the world, which were copies of it (below).

The group, the Electronic Computer Project (ECP), was assembled by John von Neumann at the Institute for Advanced Study (IAS) at Princeton (von Neumann's base institution at the time). This machine became his focus after his involvement in the planning for, and design of, the EDVAC had enlightened him to the promise of the field. (He broke off his connection to John Mauchly, and especially J. Presper Eckert, because they were focused on starting a business, the eventual Eckert–Mauchly Computer Corporation, and he wished to make all his work public - as the many copies of the IAS computer testified.)

The engineers on the ECP team initially included Julian Bigelow (Chief Engineer; replaced mid-way with initial team-member James Pomerene), John Davis, Robert Shaw, Ralph Slutz, and Willis Ware; Morris Rubinoff and Richard Snyder joined later. Other members of the team included Arthur Burks, Jule Charney, Hewitt Crane, N. Emslie, Gerald Estrin, E. Frei, Herman Goldstine, T. Hildebrandt, G. Kent, W. Melville, J. Rosenberg, Morris Rubinoff, Richard L. Snyder, and others. Their work was widely distributed via progress reports, which were circulated extensively. The IAS closed the ECP in 1957-58 (the exact date is uncertain); apparently because as an organization the IAS was uncomfortable with such an essentially practical effort. The personnel then scattered, taking their knowledge with them; many became leading lights on other early computers.

Technical details

Not one of the many, many later books which describe the IAS machine (below) give complete coverage of its architecture. This description is mostly gleaned from the Final Progress Report (below); the instruction set is enumerated at the end of that, and is given below.

It eventually used 40 Williams tubes for its main memory (after the Selectron project failed to produce usable memory, although a later copy used them). They contained 1K words in total, each tube using a 32x32 array. (Although this memory had to be refreshed, refresh cycles could be combined with read/write operations when possible.) The logic was constructed using vacuum tubes (although only about 3,000; many fewer than the ENIAC).

It was an asynchronous parallel computer internally, using 40-bit words. It used binary, with the sign bit at the left (and two's complement), and the binary point between it and the next bit; i.e. words can represent values between 1 and -1.

The machine included 3 registers, each 40 bits long:

  • RI (also given as R1 in documentation) - the Accumulator; one input to the adder, receives the output from the adder
  • RII (also R2) - the Arithmetic Register; used in multiplication and division
  • RIII (also R3) - the Memory Register; the other input (possibly complemented, or zeroed) to the adder

Each register is composed internally of a pair of 'sub-registers' (termed "ranks"), denoted as Rn (the "permanent rank") and Rn (the "temporary rank") in contemporary documentation. RI and RII can perform single-bit left and right shifts when copying data from the Rn rank to Rn; an un-shifted copy can take place from Rn to Rn. RIII can also be copied to RII, and RII to RI.

Instructions were 20 bits long, and contained a 10-bit opcode and a 10-bit address. There were about 30 in total; the number changed over time (e.g. when the I/O system was first redone, in 1952).

It had instructions to perform multiplication and division, but no special hardware; both were performed a bit at a time, with shifting (performed with the sub-registers described above) and addition/subtraction. (Throughout the contemporary documentation, a great amount of attention is paid to the details of the arithmetic - as one might expect of a design in which John von Neumann was involved!)

I/O was on paper tape initially, later (1952) switched to punched cards, and supplemented with a 7" CRT for graphics output. It was eventually (June, 1953) given an IAS-built drum of 2K words; a 12K one from ERA was later (1955) installed.

Instruction set

Two instructions were stored in each word, referred to as the 'first' (left) and 'second' (right) 'phases'; executed in that order. A few instructions could be executed in the left phase only, and one in the right only; such limitations are given in the table below (a blank entry means 'either').

The 'step digit', bit 11, may be 0 (halt after executing the instruction) or 1 in many instructions. If so, it is indicated by an 'S' in 'bits' column (and also in the numeric opcode). "S=1" means it must be set in that instruction. The 'clear digit', bit 18, is available in some instructions; if so, it is indicated by a 'C' in the 'bits' column (and will set the appropriate bit in the numeric opcode). It may be 0 (see 'Comment' for action) or 1 (clear R1 before commencing).

Potential operands are generally R1, R2, and R3 (referred to as "R3" in the table); and the contents ('b') of the memory location (address 'x'). Most instruction (unless otherwise noted) leave b in R3. A detailed description of each instruction is provided on the given page of the Final Progress Report.

Group Opcode Name Page Bits Phase Brief description Comment
Summation S712 Plus clear 22 S   Load b into R1
S710 Plus hold 23 S   Add b to R1
S732 Minus clear 24 S   Load 2-b into R1
S730 Minus hold 24 S   Subtract b from R1
S752 Plus absolute clear 25 S   Load abs(b) into R1
S750 Plus absolute hold 25 S   Add abs(b) to R1
S772 Minus absolute clear 26 S   Load 2-abs(b) into R1
S770 Minus absolute hold 26 S   Subtract b from R1 It is not clear if/how this differs from 'Minus hold'
Other S702/700 Multiply no round off 27 SC   Multiply R2 and b; high order result to R1, low order to R2 If C=0, add the old contents of R1 to the product
S706/704 Multiply round off 28 SC   As previous Ditto
S760 Division 28 S   Divide R1 by b; leaves result in R2, twice the remainder in R1
S714 Load RII 29 SC   Load R2
Trivial 1650 Store 30 S=1   Store R1; clear R3
1652 Store clear 30 S=1   Clear R1, R3, and memory
S640 Un-conditional transfer 31     Next instruction will be from 'b' (opposite phase from this one, if step bit is set); clear R3 b is stored in R3
S660 Conditional transfer 32     If R1 >= 0, then as above; if < 0, a no-op
Special 1711-1773 Quick sum 33 S=1 first See entry in Final Progress Report
Non-memory S500 Right shift, no round off 34 SC   Shift R1/R2 double-width register right N bits, always retaining sign bit in R1; copy old contents of R2 to R3 N is given in bits 4-9, and must be > 0 and < 47; if C=0, no effect, if C=1 clear R1 before shift
S504 Right shift, round off 35 SC   Not implemented
S520 Left shift 35 SC   Shift R1 and R2 independently left N bits, with the sign bit of R1 copied into the low bit of both; copy old contents of R2 to R3 As for 'right shift, no round off'; but if C=1, R1 is cleared after shift, and the sign bit of R1 is cleared before being copied to R2
S510-S572 R2 to R1 36 S   Same as 'Summation' group, but the operand is in R2, not memory
Input-output 1200-1376 IBM and drum priming 37 S second See entry in Final Progress Report S is only recommended
1010 IBM input to memory 39 S=1 first Ditto; the second instruction in the word must be a jump to the next instruction
1116 IBM output to [sic] memory 39 S=1 first Ditto
1210-1230 Drum input to memory 39 S=1 first Ditto
1316-1336 Drum output from memory 40 S=1 first Ditto

NOTE: the document uses "digit" where we would now say 'bit'.

The operation codes are given in a table in the Final Progress Report, but not in numeric form; and the column for each bit in the opcode has an individual name, not a number. It is not known if the columns are in numeric order; if they are, the step bit is in the bit number indicated elsewhere in the document (11), but the clear digit bit is not (it is given elsewhere as bit 18, but is 19 in the table). The opcodes are given above in octal, as if the bit columns in the Final Progress Report are in numeric order. The bit columns (numbered by apparent value, based on position) are labelled:

Bit '0' Value '1' Value
11 No step Step
12 Ext. Int.
13 Arith. Wms.
14 AT NAT
15 # Abs.
16 +R -L
17 x/÷ ∑
18 No RO RO
19 Hold Clear
20 Spare

Copies

As mentioned, many copies of it were built, early on (and a few later), some with the help of people who had worked on the IAS machine. The faithfulness of the copies varied from machine to machine (apparently including a last one, built with transistors); the later ones used core memory. The list of the US ones, with their dates of completion, is:

  • AVIDAC(Argonne) - January, 1953
  • CYCLONE (Iowa State University) - July, 1959
  • ILLIAC (Illinois) - September, 1952
  • JOHNNIAC (Rand) - March, 1954
  • MANIAC (Los Alamos) - March, 1952 (slightly before the original)
  • MISTIC (Michigan State University) - November, 1957 (a copy of the ILLIAC)
  • ORACLE (Oak Ridge) - September, 1953
  • ORDVAC (Aberdeen) - November, 1951

(The MANIAC was said to have been so named in protest against the 'cute' names which had become the fashion.) Elsewhere, they included:

  • BESK (Royal Institute of Technology, Sweden) - November, 1953 (included hardware floating point)
  • DASK (Academy of Technical Sciences, Denmark) - 1957 (a copy of BESK)
  • EDB-1 (Facit, Sweden) - 1957 (another copy of BESK, without floating point)
  • MUSASINO-1 (NTT, Japan) - March, 1957
  • SARA (SAAB, Sweden) - 1957 (another tweaked copy of BESK)
  • SILLIAC (University of Sydney, Australia) - June, 1956
  • SMIL (Lund University, Sweden) - June, 1956
  • TRASK (Datasystem AB, Sweden) - 1965 (another copy of BESK, using transistors and core)
  • WEIZAC (Weizmann Institute, Israel) - October, 1955

The IBM 701 was apparently a 'cleaned up' copy as well (a different word length, with half-word addresses), as were the EDB-2 and EDB-3 families built in Sweden from 1957 on, and the FACOM 201 (a copy of the MUSASINO-1) produced by Fujitsu from 1960.

Several other machines (such as the BESM 1) were built after close study of the IAS computer's documentation (and in the BESM 1's case, that of the BESK as well), but were not copies of it.

Further reading

Books

  • Herman H. Goldstine, The Computer from Pascal to von Neumann, Princeton University, Princeton, 1972 - contains details of the entire effort, in which Goldstine participated
  • William Aspray, John von Neumann and the Origins of Modern Computing, MIT Press, Cambridge, 1990 - not much technical detail, but a good overall history, and a lot about problem areas the machine was used for
  • Julian Bigelow Computer Development at the Institute for Advanced Study; pp. 291-310 in Nicholas Metropolis, Jack Howlett, Gian-Carlo Rota (editors), A History of Computing in the Twentieth Century, Academic Press, New York, 1980 - the IAS computer is covered in the first section of Part IV, but with a focus on the electrical engineering; many descendants are covered in detail in other sections
  • George Dyson, Turing's Cathedral: The Origins of the Digital Universe, Pantheon, New York, 2012 - focused on the creation of this machine at the Institute for Advanced Study
  • Raul Rojas, Ulf Hashagen, The First Computers: History and Architectures, MIT Press, Cambridge, 2002 - the IAS machine is covered in chapter II/4; no detail of the machine, but interesting coverage of the background and influence

Papers

External links

Retrieved from "https://gunkies.org/w/index.php?title=IAS_computer&oldid=35185"