@Downix
Well I certainly did find some interesting info on this
subject. Check out this link:
http://www-it.desy.de/systems/linux/meetings/22ndLUM_Benchmarks.pdfOf notable interest to us is page 4, under the heading "Dirac
Operator Benchmark", which rates a P4 @1.4ghz as capable
of achieving 1503 MFLOPS while performing 32-bit maths;
814 computing 64-bit maths (under this particular test).
Now this is entirely consistant with my earlier benchmark,
rating the P3 @750mhz as capable of processing 998
MFLOPS - considering that clock - for - clock, there is
very little difference between the P3 and P4 (some tests
even rate the P3 as being "faster" processing certain tasks
at the same clockspeed).
We also have to realize that MFLOPS is an entirely subjective
rating to begin with - are we measuring float multiplication,
division, linear (static) float or perhaps interleaved (piped)
float? And consider that the increased superscalar nature of
modern chips means that they can process much more
information per clock cycle than those old chips ever could;
note the radical difference between the 68040 and 68060
when performing certain operations due to the 68060's (quite
basic) super-scalar abilities. This site seems an accurate
fundamental description of MIPS and MFLOP ratings:
http://futuretech.mirror.vuurwerk.net/perf.htmlAnother interesting factor from the above (first) link is the
difference between the P4 MFLOPS rating when comparing
the 32 and 64-bit maths tests...the difference is almost linear!
Applying this same logic to our 750mhz P3, we can see that
even if the MFLOPS rating of 998 was achieved while
calculating 8 or 16-bit functions (unlikely), the P3 would still
be considerably faster...and in some respects the jury is
still out on this one, as the particular design of the "32-bit"
CPU will largely influence how well it may calculate
8 or 16-bit maths.
And now we haven't even addressed the issue of memory
bandwidth or "graphics" performance, which surely
would have put the (already) challenged 3000+ at an even
larger disadvantage paired against our P3. Also, in the
end, we have to think about 'real world' factors: How would
the 3000+ compare to our PIII in the theoretical game of
Q2?
-edit-
Check out this cool site for some more accurate benchmarks
on the PII @450mhz:
http://n0cgi.distributed.net/speed/