4 ms·

In 1964 importance

dfox 5y ago
Well, it is an computer that: 1) uses architecture that has at least user-space binary compatible descendands today. 2) introduced 32b wordsize with 8b byte addressable memory. 3) was designed to use ASCII. (which is somewhat ironic as all the IBM big iron things are mostly synonymous with EBCDIC, which was only meant as temporary solution because ASCII was not finalized in 1964)
- iamhamm 5y ago
  Totally agree. The comment I was replying to asked "most important hardware announcement ever made by a tech firm?" and I was noting that in 1964 it was. That said, it's still a great read and I can imagine how cool it was :)
Throwawayaerlei 5y ago
dfox lists some important things, the two most important I'd note are how through extensive use of microcode there was a huge range of systems that could run the same software, and for the lower end ones could also emulate older systems. Going from memory and not counting the very limited 360/20, from the 8 bit internally 360/30 all the way to the 360/91 supercomputer you could run basically the same software. Which eventually solved a huge problem for IBM and the industry in general because not so much effort had been put into comparability in the early days, and IBM had two major lines of big computers, decimal ones for business and binary ones officially for science and engineering but by then often used by businesses. Although stuff at the low end which somewhat like the 360/20 remained incomparable, the business 1401 series remained, also see the scientific 1130 which a lot of people including myself got their first start at programming on. It was a very clever design that used 360 technology to make an affordable system, for quite some time they'd put effort into that niche with the 650 and the CADET, Can't Add, Doesn't Even Try, a backronym from how they based it on a big block of core memory which they were very good at manufacturing by then vs. the drum as used in the 650, and and used that with tables etc. to avoid including an ALU. And leaders of the effort going all the way up to Tom Watson Jr. made a catastrophic mistake in deciding virtual memory, which they called dynamic address translation (DAT) was a bad thing, which hardened so much the first models of the System/370 which used "monolithic" circuits, ICs, didn't have it (and they didn't do right by their customers who bought those first models). The affordability of the 360 series and their ability to make so many but much less than initial demand rested on their avoiding too new at the time ICs and using automation to manufacture small boards logic boards on a ceramic substrate: https://en.wikipedia.org/wiki/Solid_Logic_Technology https://en.wikipedia.org/wiki/Solid_Logic_Technology This was an era in which IBM still massively benefited from its electromechanical wizardry gained in the punched card data processing days. This lost them the high end computer science community which I believe seriously harmed them in the long term, ultimately made what they now call IBM Z a niche product.
- GeorgeTirebiter 5y ago
  The CADET was in fact the IBM 1620, contemporaneous with the IBM 1401, but more for 'scientific' market. The '1620 uses these clever in-core tables and iteration to achieve arbitrary-precision addition/subtraction and different tables for multiplication/division; I implemented the algo in C and it worked great, certainly fast enough for the occasional arb-precision computation. The IBM 1401 could operate on arbitrary strings of digits in hardware ; since that was my first machine, I was disappointed when I used these silly 'binary' machines. However, a typical instruction + operand fetch/store took about 100 microseconds on the '1401; yes, that is 10 kilo-instructions per second! There were many programming tricks to minimize instruction fetch e.g.'chaining' instructions, using left-over contents of the regs in the next instr, and using parts of instr as data, the most common being: "A *-6,FOO" because "*-6" is relative address, which points to the letter 'A', which was BCD +1, etc. Equivalent to ++FOO; The pdp-11/45 I used next took about 2.5 usec per instruction with 2 memory operands -- 400 kilo instructions per second! This "40x" speedup of the 11/45 over the IBM 1401 more than compensated for the hardware arbitrary-precision addition/subtraction! (oh, and on the '1401, the HW 'Multiply/Divide Feature' was another big group of electronic cards, if you bought that) http://ed-thelen.org/comp-hist/ibm-1401.html http://ed-thelen.org/comp-hist/ibm-1401.html (edit: learned that \* needed to not have * gobbled by the markdown italics sentinel https://news.ycombinator.com/formatdoc https://news.ycombinator.com/formatdoc)

In 1964 importance

dfox 5y ago

Well, it is an computer that: 1) uses architecture that has at least user-space binary compatible descendands today. 2) introduced 32b wordsize with 8b byte addressable memory. 3) was designed to use ASCII. (which is somewhat ironic as all the IBM big iron things are mostly synonymous with EBCDIC, which was only meant as temporary solution because ASCII was not finalized in 1964)

iamhamm 5y ago
Totally agree. The comment I was replying to asked "most important hardware announcement ever made by a tech firm?" and I was noting that in 1964 it was. That said, it's still a great read and I can imagine how cool it was :)

Throwawayaerlei 5y ago

dfox lists some important things, the two most important I'd note are how through extensive use of microcode there was a huge range of systems that could run the same software, and for the lower end ones could also emulate older systems. Going from memory and not counting the very limited 360/20, from the 8 bit internally 360/30 all the way to the 360/91 supercomputer you could run basically the same software. Which eventually solved a huge problem for IBM and the industry in general because not so much effort had been put into comparability in the early days, and IBM had two major lines of big computers, decimal ones for business and binary ones officially for science and engineering but by then often used by businesses. Although stuff at the low end which somewhat like the 360/20 remained incomparable, the business 1401 series remained, also see the scientific 1130 which a lot of people including myself got their first start at programming on. It was a very clever design that used 360 technology to make an affordable system, for quite some time they'd put effort into that niche with the 650 and the CADET, Can't Add, Doesn't Even Try, a backronym from how they based it on a big block of core memory which they were very good at manufacturing by then vs. the drum as used in the 650, and and used that with tables etc. to avoid including an ALU. And leaders of the effort going all the way up to Tom Watson Jr. made a catastrophic mistake in deciding virtual memory, which they called dynamic address translation (DAT) was a bad thing, which hardened so much the first models of the System/370 which used "monolithic" circuits, ICs, didn't have it (and they didn't do right by their customers who bought those first models). The affordability of the 360 series and their ability to make so many but much less than initial demand rested on their avoiding too new at the time ICs and using automation to manufacture small boards logic boards on a ceramic substrate: https://en.wikipedia.org/wiki/Solid_Logic_Technology https://en.wikipedia.org/wiki/Solid_Logic_Technology This was an era in which IBM still massively benefited from its electromechanical wizardry gained in the punched card data processing days. This lost them the high end computer science community which I believe seriously harmed them in the long term, ultimately made what they now call IBM Z a niche product.

GeorgeTirebiter 5y ago
The CADET was in fact the IBM 1620, contemporaneous with the IBM 1401, but more for 'scientific' market. The '1620 uses these clever in-core tables and iteration to achieve arbitrary-precision addition/subtraction and different tables for multiplication/division; I implemented the algo in C and it worked great, certainly fast enough for the occasional arb-precision computation. The IBM 1401 could operate on arbitrary strings of digits in hardware ; since that was my first machine, I was disappointed when I used these silly 'binary' machines. However, a typical instruction + operand fetch/store took about 100 microseconds on the '1401; yes, that is 10 kilo-instructions per second! There were many programming tricks to minimize instruction fetch e.g.'chaining' instructions, using left-over contents of the regs in the next instr, and using parts of instr as data, the most common being: "A *-6,FOO" because "*-6" is relative address, which points to the letter 'A', which was BCD +1, etc. Equivalent to ++FOO; The pdp-11/45 I used next took about 2.5 usec per instruction with 2 memory operands -- 400 kilo instructions per second! This "40x" speedup of the 11/45 over the IBM 1401 more than compensated for the hardware arbitrary-precision addition/subtraction! (oh, and on the '1401, the HW 'Multiply/Divide Feature' was another big group of electronic cards, if you bought that) http://ed-thelen.org/comp-hist/ibm-1401.html http://ed-thelen.org/comp-hist/ibm-1401.html (edit: learned that \* needed to not have * gobbled by the markdown italics sentinel https://news.ycombinator.com/formatdoc https://news.ycombinator.com/formatdoc)