Vincent BARAT wrote:
> 
> OK, and what about a 386 33 MHz from 1991 (the reference machine of byte
> benchmark, because the fastest PC they found when they issued the test).
> I think it has an even slower I/O bus and no L2 cache (nor L1 by the way !).
> So I can't understand the difference...

Well, sometimes it doesn't mean a thing to do a direct comparison
between two systems running a benchmark program, since any benchmark
program can emphasize few aspects of the system out of propotion.

For example, it has been known that bytebench-3.1 result is greatly
affected by the speed of unlink(2) system call which is noticeably
faster on Linux than on *BSD (by default) due to so-called async
metadata update (I could be wrong on this one...  this is from top of
my head).

So, I suggest that you stop taking the benchmark result too seriously.
Of course you *could* run Linux-m68k on your Centris just to prove L
is faster than N, but I doubt it has any productive value whatsoever
(not to mention L is not fully functional on the Centris/Quadra class
machines just yet).

Ken