One reason that 48 bit words were so popular in this class of machine is that there are some significant differential equation solutions which converge at 48 bit but diverge at 32 bit floating point. So on the S/360 you would have to use 64 bit floating point to solve them. When memory cost USD 1 per bit that difference was a big impact.
On Tue, Nov 17, 2020 at 11:58 AM Wayne Bickerdike <[email protected]> wrote: > Atlas rang a bell and then I remembered it was used at the Rutherford labs > in Chilton. > > I worked at AERE (Atomic Energy Research Establishment) which was next > door. They were basically the same campus but AERE was secure and > Rutherford wasn't. > > Ferranti sold two other Atlas installations, one to a joint consortium > of London > University <https://en.wikipedia.org/wiki/London_University> and British > Petroleum <https://en.wikipedia.org/wiki/British_Petroleum> in 1963, and > another to the Atomic Energy Research Establishment > <https://en.wikipedia.org/wiki/Atomic_Energy_Research_Establishment> > (Harwell) > in December 1964. The AEA machine was later moved to the Atlas Computer > Laboratory at Chilton, a few yards outside the boundary fence of Harwell, > which placed it on civilian lands and thus much easier to access. This > installation grew to be the largest Atlas, containing 48 kWords of 48-bit > core > memory <https://en.wikipedia.org/wiki/Core_memory> and 32 tape drives. > > Funny story: > > Somebody stole a lathe from the Rutherford labs (c. 1968). After that > happened they installed a gate and security guards at the lab. > > At Harwell, we were checked for ID on entry, at Rutherford you were checked > for lathes on exit! > > > > > On Tue, Nov 17, 2020 at 8:42 AM Bernd Oppolzer <[email protected] > > > wrote: > > > Well, I didn't know much about Atlas until now, so I had to do some > > research; > > very impressive, indeed. Atlas is about 3 to 5 times faster than the TR > 4. > > But in contrast to Atlas, the TR 4 was produced in a sort of small > > industrial series (35 machines); > > Atlas seems to be a super computer with very few installations. > > > > What strikes me most are the similarities: > > > > - 48 bit words > > - two 24 bit signed integers or six 8 bit-bytes or one 48 bit integer or > > floating point > > - or one instruction (two on TR 4 / TR 440) > > - 24 bit address space (only 16 on the TR 4, later 22 on the TR 440) > > - 16 k words of core store (32 k max on TR 4, 192 k words on the TR 440, > > maybe more) > > - 8 k words of read only memory for supervisor etc. (4 k on TR 4, > > nothing on TR 440, > > was loaded at startup time) > > > > Atlas was the first implementation of virtual memory, > > but the first paper on virtual memory is from Fritz-Rudolf Güntsch from > > 1957 > > https://de.wikipedia.org/wiki/Fritz-Rudolf_G%C3%BCntsch - > > German wikipedia says: Güntsch is the inventor of virtual memory > > (German title: F.-R. Güntsch: /Logischer Entwurf eines digitalen > > Rechengeräts > > mit mehreren asynchron laufenden Trommeln und automatischem > > Schnellspeicherbetrieb./ > > TU Berlin, 1957 (Dissertation)) - and: Fritz-Rudolf Güntsch later (in > > the 1960s) > > worked for Telefunken and was the designer of the TR 4 and the TR 440 :-) > > > > Kind regards > > > > Bernd > > > > > > > > Am 16.11.2020 um 20:46 schrieb Seymour J Metz: > > > Faster than Atlas? > > > > > > > > > -- > > > Shmuel (Seymour J.) Metz > > > http://mason.gmu.edu/~smetz3 > > > > > > > > > ________________________________________ > > > From: IBM Mainframe Discussion List <[email protected]> on > > behalf of Bernd Oppolzer <[email protected]> > > > Sent: Sunday, November 15, 2020 5:43 PM > > > To: [email protected] > > > Subject: Re: Improve OMVS cp performance? > > > > > > Telefunken TR 4, designed 1958, first delivered in 1962. This predates > > > IBM/360 > > > by at least 4 years. The fastest mainframe built in Europe at that > time. > > > The internal code ("Zentralcode") was an 8-bit code using 256 > characters. > > > Word structure, a word had 48 bits plus 2 tag bits (tagged > architecture) > > > plus some parity bits, not seen by the programmer. > > > Mostly used with ALGOL; the TR 4 was "a Hardware implementation of > ALGOL" > > > (quote from E.J. Dijkstra). > > > A word could hold up to six characters, but some later languages > > > (like Fortran) decided to store only 4 characters in one word, > > > to be more compatible with IBM Fortran. > > > > > > Kind regards > > > > > > Bernd > > > > > > ---------------------------------------------------------------------- > > For IBM-MAIN subscribe / signoff / archive access instructions, > > send email to [email protected] with the message: INFO IBM-MAIN > > > > > -- > Wayne V. Bickerdike > > ---------------------------------------------------------------------- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to [email protected] with the message: INFO IBM-MAIN > ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
