From: minforth@gmx.net   
      
   Am 11.07.2025 um 03:40 schrieb Paul Rubin:   
   > minforth  writes:   
   >> Kahan was also overly critical of dynamic Unum/Posit formats.   
   >> Time has shown that he was partially wrong:   
   >> https://spectrum.ieee.org/floating-point-numbers-posits-processor   
   >   
   > I don't feel qualified to draw a conclusion from this.  I wonder what   
   > the numerics community thinks, if there is any consensus.  I remember   
   > being dubious of posits when I first heard of them, though Kahan   
   > probably influenced that.  I do know that IEEE 754 took a lot of trouble   
   > to avoid undesirable behaviours that never would have occurred to most   
   > of us.  No idea how well posits do at that.  I guess though, given the   
   > continued attention they get, they must be more interesting than I had   
   > thought.   
   >   
   > I saw one of the posit articles criticizing IEEE 754 because IEEE 754   
   > addition is not always associative.  But that is inherent in how   
   > floating point arithmetic works, and I don't see how posit addition can   
   > avoid it.  Let a = 1e100, b = -1e100, and c=1.  So mathematically,   
   > a+b+c=1.  You should get that from (a+b)+c in your favorite floating   
   > point format.  But a+(b+c) will almost certainly be 0, without very high   
   > precision (300+ bits).   
      
   AFAIK Cuda does not support posits (yet). BFLOAT16 etc. still win the   
   game, until the AI industry pours big money into the chip foundries   
   for posit math GPUs.   
      
   Even then, it is questionable, whether or when it would seep into the   
   general-purpose CPU market.   
      
   For Forthers to play with, of course. ;o)   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)