The IBM 360 single precision floating point has a range of 10**-79 to 10**75; double precision and extended precision has the same number of bits for the exponent.
From: "Paul Koning via cctalk" <cctalk@classiccmp.org> To: "cctalk" <cctalk@classiccmp.org> Cc: "David Wise" <d44617...@hotmail.com>, "CAREY SCHUG" <sqrfolk...@comcast.net>, "Paul Koning" <paulkon...@comcast.net> Sent: Thursday, October 3, 2024 10:39:47 AM Subject: [cctalk] Re: Might be antique computer parts --1620 > On Oct 3, 2024, at 12:01 PM, CAREY SCHUG via cctalk <cctalk@classiccmp.org> > wrote: > > I worked on a model 1 with 40k memory (my very first computer experience) and > floating point, and later a model 2 stripped. I believe the model 2 still > used table lookup for multiply. Sure enough, that's what the Model 2 manual I just found on Bitsavers says. I always though that both add and multiply were in hardware. Interesting. Another major enhancement in model 2 is index registers. > floating point in model 1 (and I think model 2) was limited to a 98 digit > mantissa, still more precision than the hardware in any subsequent computer > AFAIK. since the exponent was 10**-99 to 10**99 a broader range than any > computer till many years later I think. 100 digits says the model 2 manual. Exponents larger than that do appear in machines of that era. The CDC 6000 series goes up to 10**322 (11 bit signed binary exponent) and the Electrologica X8 to 10**644 (12 bit signed binary exponent). Both are early 1960s. I don't remember what the IBM 360's range is; the exponent field is fairly small but it's a power of 16 rather than the usual power of 2 (or power of 10 as in the 1620). > ... > don't know if the disk i/o RPQ overlapped, I saw one through a glass window > once. No, just like all the other I/O instructions they block the CPU. paul
