Some Old Computers

This page is really a bit of fun: just details of some of the really old (2nd generation) machines I have known. Other historical sites you may want to visit include:

Some old British Machines

What follows is a summary of some British architectures from the 1960's: I give them in (approximate) chronological order. They share a common philosophy (if not a common architecture), apparently deriving from the vacuum-tube designs of the early 1950's. None of these machines used integrated logic in any form: everything was discrete transistors, and in some cases, ferrite-core logic.

Elliott 803

This machine was introduced around 1962. It was also the first machine I myself ever wrote code for, in 1968!
Built using discrete transistor and ferrite-core technology, and using bit-serial logic. Word length was 39 bits. This architecture seems to have originated in the Ferranti Pegasus, a vacuum-tube machine from 1956.

Sizes & weights (inches, pounds)

Central Processor 66x56x16 680
Extra 4096 words store 33x56x16 250
Power Unit 33x56x16 380

19-bit instructions were packed 2 to a word, with the 39th bit acting as a "modifier" on the second instruction. The effect of this was to add the content of the memory location addressed in the first instruction to the whole of the second: this was the only computed-address mechanism available (short of self-modifying code). The modification acted on the opcode bits as well as the address: this was the basis of many truly incomprehensible coding tricks!
Half-page Algol programs used to take half an hour to compile & execute. Amazingly, the Algol compiler and run-time package (developed under the direction of C. A. R. Hoare) would fit in 8192 words of memory. Standard IO was 5-track paper tape, with magnetic tape an option. The mag-tape was termed "film": the tape format actually being identical to 35mm movie film.

Hardware floating-point was available, with a 39-bit floating-point number, and a (binary) exponent range of +/-256.
With a maximum memory capacity of 8192 words, 13 address bits were used. The remaining 6 bits were available for the opcode, and were represented as 2 octal digits.

Instructions

The content of the single 39-bit accumulator is denoted by a, and that of the addressed location by n. a' and n' denote the values after the operation. N denotes the address part of the instruction.

Groups 0 - 3 (Arithmetic) 576uS

Group 0 Group 1 Group 2 Group 3
Code a' n' Code a' n' Code a' n' Code a' n'
00 a n 10 n a 20 a a 30 n n
01 -a n 11 -n a 21 a -a 31 n -n
02 n+1 n 12 n+1 a 22 a n+1 32 n n+1
03 a&n n 13 a&n a 23 a a&n 33 n a&n
04 a+n n 14 a+n a 24 a a+n 34 n a+n
05 a-n n 15 a-n a 25 a a-n 35 n a-n
06 0 n 16 0 a 26 a 0 36 n 0
07 n-a n 17 n-a a 27 a n-a 37 n n-a

Group 4 (Jumps) 288uS

Code Code Condition
40 44 Unconditional
41 45 If a negative
42 46 If a zero
43 47 If oflo. (clear flag)
40..43 Jump to 1st half of word
44..47 Jump to 2nd half of word

Group 5 (Shifts)

Double-length shifts use a and ar (the "auxiliary accumulator")

Code Function
50 Arithmetic right shift N places
51 Logical right shift N places, clear ar
52 Multiply (double-length product)
53 Multiply (single-length product), ar:= 0
54 Shift left, double-length, N places
55 Shift left, single length, ar:= 0
56 Divide a:ar by n, ar:= 0
57 a:= ar

Group 6 (Floating-point & miscellaneous functions)

Code Function
60 a:= a+n
61 a:= a-n
62 a:= n-a
63 a:= a*n
64 a:= a/n
65 Convert integer in A to floating-point
66 Interrupt return (503 only)
67 Modify next instruction by adding LS 19 bits of n (503 only)

Group 7 (Input/Output)

70 Read panel key-switches to a
71 Read character from device N to a
72 Output 39-bit word from a to device N
73 Write the address of this instruction to location N
74 Output character to device N
75 Input 39-bit word from device N to a
76 Setup hardware for block transfer (503 only)
77 Initiate DMA operation (503 only)

Elliott 503

Introduced in the mid-60s, the 503 was a fully parallel replacement for the 803, running some 60 times faster. Group 0 instructions, for example, took 7.2uS. This machine, fully optioned up, would have weighed several tons.

Sizes & weights (inches, pounds)

Logic Cabinet 78x38x24 1000
Main Store cabinet 78x38x24 850
Power Cabinet 78x38x24 1350
Full Controller cabinet 78x38x24 1000
Half Full Controller cabinet 78x38x24 750
Peripheral Services cabinet 78x38x24 1200

A little exercise

Just for fun, here's a stock bit of code (a subroutine entry/exit sequence) for these machines. Do you see what's happening (tip: the "/" operator between the instructions invokes the "modify" function):
. 1st half . 2nd half
Label Code Operand Modify Code Operand
caller 73 SUBR : 40 SUBR+1
.....
SUBR: 00 0 : 00 0
subroutine code.....
return 22 SUBR / 40 0

You can jump here for an explanation.

Elliott 903

This machine was originally designed for a British military project, and spawned a series of military and civilian derivatives. The basic machine fitted in a desk 37 x 26 x 43 inches, weighing 470lb. It could be likened to a 503 cut in half: the word-length is 18 bits, and accommodates a single instruction. Most instructions executed in around 20uS. The bit assignment within the word was:

Address 13 bits
Opcode 4 bits
Modifier 1 bit
   
Like the 503, address space was 8192 words, with extra memory accessible by paging. Being designed for real-time applications, the 903 featured 4 interrupt priority levels. Separate S (program counter) and B (modifier) registers for each level were mapped into the first 8 memory locations. Unlike the 503, the "modifier" did not affect the operand of the previous instruction, rather the modifier (here designated B) was added to the address part of the instruction. The B register was loaded by Instruction 0.

Instruction codes (notation as above, with B & Q added)

Code Function
0 b:= n; q:= n
1 a:= a+n
2 a:= n-a; q:= n
3 n:= q
4 a:= n
5 n:= a
6 a:= a & n
7 Jump to N if a=0
8 Jump to N unconditionally
9 Jump to N if a<0
10 n:= n+1
11 Store program counter in N
12 A:Q:= a*n This multiply used fractional notation
13 a:= A:Q/n This divide used fractional notation
14 Shift A:Q left or right N places (N signed)
15 Miscellaneous: I/O, interrupt return, etc.

Singer System-10

Being designed in the US, this machine wouldn't normally rate a place on this page. However it is so unusual that it deserves mention somewhere on the Web, and no-one else seems to have done so. So here it is, on its own page. This is put together from memory, so there may be errors. Corrections & additions are welcome.
HomeBack to Home Page

What that code was doing

Couldn't figure it out? OK, step by step:
  1. 73 SUBR writes the address of this instruction in SUBR (that's the all-zero location that starts the subroutine).
  2. 40 SUBR+1 now we jump to the (first half of the) instruction following the SUBR location: this will be the first live instruction of the subroutine.
  3. 22 SUBR this is the return instruction. We begin by incrementing the stored value, so it now points after the 73 instruction that stored its address. That is, it now points at the instruction we want to return to. We then modify the next instruction.
  4. / 40 0 this instruction looks like a jump to zero. But it was modified by the previous instruction, which added the (incremented) content of SUBR to the instruction. Since that would be a 13-bit memory address, the effect is to change the jump target to the required return point, as required.
Of course, 00 SUBR / 40 1 would have returned just as well (wouldn't it?)
HomeBack to Home Page
Valid XHTML 1.0!