From Newsgroup: comp.lang.fortran

On 2/6/2026 2:48 PM, David Jones wrote:

Thomas Koenig wrote:

Radey Shouman <shouman@comcast.net> schrieb:

The size of the integer type was specified to be the same as the
size of the real type. The real type need not be 32-bit floating
point, but that is certainly the way to bet.

A famous counter-example: Cray's 64-bit REALs.

And the Univac's 36 bit reals.

I miss the Univac 1108 36 bit reals and integers. They were perfect for
1 ppm (1e-6) calculations, good for most of our calculations.
Unfortunately, we had to walk away from the Univac 1108 in 1981 ? due to
its 1 megaword addressing limit. We had already moved to a Prime 450 in
1977 and a Prime 750 in 1980 ? for our software development.

Double precision in 64 bit reals takes you to 1e-15 calculations
generally. That is actually a problem for some iterative calculations
as the delta variable goes too small and must be tested for routinely.

Lynn

--- Synchronet 3.21b-Linux NewsLink 1.2