2) The only SPARC implementation available today is the Fujitsu
gate array (details have been posted). I have the data sheet,
the bus interface is *horrible* unless you want to use the
chip *exactly* as Sun does. Future versions by Fujitsu may
be different (rumor).
3) The performance of SPARC implementations does *NOT* necessarily
scale with clock rate! The Fujitsu gate array does LOADs in
2 cycles (I-fetch, data-read) but STOREs in 3 cycles (I-fetch,
i-don't-know-what, write-data). The data sheet doesn't say anything
about that, but the Fujitsu rep. explained it. It has to do with
the funny bus interface. I forgot the exact reason.
All this to say, that another *implementation* of the SPARC
*architecture* may use a different number of cycles, more
busses or god knows what.
4) There is no way the Fujitsu SPARC chip can emulate 68020 code
at 7MIPS (whatever that means). I think the confusion arises
from the "source code compatibility" between Sun-3 (68020) and
Sun-4 (SPARC) claimed by Sun. All that means, is that source code
used on Sun-3 can be compiled for Sun-4 "unaltered".
5) The Fujitsu SPARC chip number is MB86900. There is a floating
point controller to interface the SPARC cpu with Weitek 1164 and 1165
chips. It's number: MB86910.
6) Flame ...
Question: how many pins do you think the SPARC cpu has?
Answer: 256!
Question: how big do you think the package is?
Answer: 2 inches on a side!
Question: is this a joke?
Answer: No!
Thorsten von Eicken
...!research!tve
t...@research.uucp
PS: Please correct if I mistated anything. I'm not trying to make the SPARC
look bad ... nor good ... nor ...
Of course Sun's claim about running Sun-3 code unaltered on a Sun 4
is utter fantasy. Many programs that run fine on a Sun 3 will
not run on a Sun-4 without major modifications. (And of course
many programs that run a Sun-3 will indeed run fine on a Sun -4. The
point is that they all won't.)
The exact claim is "Even though the Sun-4 family is based on a powerful
new architecture, Sun has maintained 100% source-code compatibility
with Sun-2 and Sun-3 product families"
Sun's response to that is that the programs are not portable if they
won't run on a Sun 4. However, thats not the issue. Marketing claims
that any program that runs on a sun3 need only be recompiled to
run on a Sun-4 it doesn't say anything about passing lint with
NO ERRORS (I have never seen a real program that didn't get some
sort of lint error).
---rick