pi and LIA

[Gideon Yuval]

Roger's 75-million figure seems high

(I quote: "There are probably 75M PCs in use with FP hardware.  Among 
people who
do FP work and are concerned about that last bit, I doubt very many of
them buy PCs with no FP hardware.")

On the 88, 86, 286 and 386, the X87 was an expensive option. And the 
Intel 486SX (including SX2!), Cyrix 486SLC, IBM486SLC, AMD486SX (&SX2), 
NexGen586, all sell at an interesting discount to the DX version (i.e. 
the one with an FPU).

I have no idea how the market splits on the issue; but most software 
vendors believe that crashing (or running dog-slow) on an SX is a great 
way to lose the marketing war before it has started. The only big 
exception I know is Autocad -- which sells for more than the price on a 
cheap 486 box.



On another issue: precision control can make the Intel FPU round to 53- 
or 24-bit mantissa precision; but the exponent range stays as large as 
woth full 80-bit math. The only way to make that FPU do SP/DP math on 
every intermediate result is to store and reload from memory.
----------
| From: Roger Schlafly  
<netmail!validgh!uunet!attmail.com!rschlaflyauunet.uu.net>
| To: Mitch Alsup  <netmail!uunet!uunet!ross.com!mitchauunet.uu.net>;
| <netmail!uunet!uunet!validgh.com!numeric-interestauunet.uu.net>